Release: CLI v3.6.1, extension v1.3.3

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.agent/skills/impeccable/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: impeccable
 description: Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, settings, onboarding, and empty states. Handles UX review, visual hierarchy, information architecture, cognitive load, accessibility, performance, responsive behavior, theming, anti-patterns, typography, fonts, spacing, layout, alignment, color, motion, micro-interactions, UX copy, error states, edge cases, i18n, and reusable design systems or tokens. Also use for bland designs that need to become bolder or more delightful, loud designs that should become quieter, live browser iteration on UI elements, or ambitious visual effects that should feel technically extraordinary. Not for backend-only or non-UI tasks.
-version: 4.1.0
+version: 4.1.2
 license: Apache 2.0
 allowed-tools:
   - Bash(npx impeccable *)

.agent/skills/impeccable/reference/craft-floor.md

@@ -38,6 +38,7 @@ Surface habits:
 - Monospace as a costume for "technical" rather than for code, data, or measurement.
 - A system display face (Impact, Arial Black, the platform sans) as the display voice of an own-world page. Source and self-host a face whose character matches the approved lettering; the closest installed font is a failure, not a fallback.
 - Unicode glyphs or emoji standing in for an icon system. Icons are drawn, from a real library or authored SVG, in one consistent stroke and weight.
+- Geometric masks standing in for organic contours. A circle, polygon, or radial-gradient cutout approximating a photographic subject's edge is the cheap version of the effect and reads worse than omitting it. Derive an alpha matte from the actual image, or produce a cut-out asset.
 - Light or dark picked by category. Pick it from the use scene: who, where, under what ambient light.
 
 The floor holds the mechanics; it never picks the direction. With every check green, spend the page on the committed world, and when torn between refined and committed, commit.

.agent/skills/impeccable/reference/degraded/asset-producer.md

@@ -3,17 +3,15 @@ This harness has no subagent capability, so you are running this role inline. St
 
 # Impeccable Asset Producer
 
-You are the asset production agent for Impeccable craft.
-
-Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent agent gives you. The assets you create will be used to build a real site, so treat every raster as a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
+You are the asset production agent for Impeccable craft. Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent gives you. Every raster you create is a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
 
 ## Core Rule
 
-Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; if CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
+Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; when CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
 
 ## Decision Comps
 
-When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so your entire contract is this card; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; a card too thin to brief a comp is reported back, not padded from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (its regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world; a native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment is what keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
+When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so this card is your entire contract; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; report a card too thin to brief a comp, never pad it from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world. A native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
 
 ## Input Contract
 
@@ -25,42 +23,42 @@ Expect:
 - Required dimensions, format, transparency needs, and avoid list.
 - Notes on what should remain semantic HTML/CSS/SVG instead of raster.
 
-If the source mock is attached but has no filesystem path, use it for visual planning. Ask for a path only before cropping or writing assets.
+If the source mock is attached but has no filesystem path, use it for visual planning; ask for a path only before cropping or writing assets.
 
-Use defaults unless contradicted:
+Defaults unless contradicted:
 
 - `.webp` for opaque photos, backgrounds, and textures.
 - `.png` for transparent cutouts, seals, tickets, and illustrations.
-- Target production size or at least 2x display size when dimensions are known. Do not use small full-page mock crop size as the default shipping size.
-- Remove UI text, navigation, buttons, labels, and body copy by default.
+- Target production size, or at least 2x display size when dimensions are known. Never default to the small size of a full-page mock crop.
+- Remove UI text, navigation, buttons, labels, and body copy.
 - Keep physical marks only when the parent says they are part of the asset.
-- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic to the asset.
-- Keep the final assets directory clean: only files the build will consume belong there. Put source crops, reference crops, masks, and contact sheets in a sibling `_sources`, `sources`, or review folder.
+- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic.
+- Keep the final assets directory clean: only files the build will consume. Source crops, reference crops, masks, and contact sheets go in a sibling `_sources`, `sources`, or review folder.
 
-Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not block; choose defaults and report them.
+Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not; choose defaults and report them.
 
 ## Workflow
 
 1. Inventory the full approved mock or every assigned crop.
 2. Put each visual role in exactly one bucket:
    - `produce`: needs generation, image editing, cleanup, cutout work, or a clean plate before it can ship.
-   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source asset, a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
+   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source: a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
    - `semantic`: build in HTML/CSS/SVG/canvas, no raster output.
-3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, not asset grade, and a shipped crop, however close it looks, is how a beautiful comp turns into a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
+3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, and a shipped crop, however close it looks, is how a beautiful comp becomes a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
 4. Give the parent an execution order for the `produce` bucket.
-5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or semantic HTML/CSS/SVG recommendation if raster is wrong.
+5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or a semantic HTML/CSS/SVG recommendation when raster is wrong.
 6. Use the harness's native image tool by default when generation or editing is needed; otherwise use the skill's generate-image.mjs.
 
 7. Remove baked-in UI text, navigation, buttons, body copy, and mock chrome unless the text is part of the asset.
 8. Think through the final DOM/CSS representation before generating. If CSS will own radius, clipping, shadows, borders, perspective, responsive cropping, captions, or card frames, do not bake those into the bitmap.
-9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .agent/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt is embedded in the image itself, because the build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
+9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .agent/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt lives inside the image itself. The build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
 10. Compare each output against its source crop, opening every image by its workspace-relative path; sandboxed viewers reject absolute paths. If a review/QA tool is available, run it before the final manifest, then retry each major/fatal finding once before finalizing.
 
-Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed to make a reusable texture or background, classify it as crop-derived cleanup or clean-plate work.
+Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed first, classify it as crop-derived cleanup or clean-plate work.
 
-Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Only ship a screenshot raster when the parent explicitly says the screenshot itself is the final asset.
+Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Ship a screenshot raster only when the parent explicitly says the screenshot itself is the final asset.
 
-Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: name the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it should compose with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
+Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it composes with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
 
 ## Prompt Pattern
 
@@ -81,10 +79,10 @@ For transparent cutouts: use true alpha when the tool supports it; otherwise gen
 
 Return a complete manifest, grouped by `produce`, `direct`, and `semantic`. For each asset include: `id`, `source_crop`, `output_path` when applicable, `strategy`, `prompt_used` when applicable, `dimensions`, `format`, `transparency`, `deviations`, and `qa_status`.
 
-For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` must be a concrete build handoff, not a short explanation that no asset was produced. It should name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities that code owns.
+For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` is a concrete build handoff, not a note that no asset was produced: name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities code owns.
 
-`qa_status` must be `accepted`, `needs_parent_review`, or `blocked`. Use `accepted` only after visual comparison passes. Use `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. Use `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
+`qa_status` is `accepted`, `needs_parent_review`, or `blocked`. `accepted` only after visual comparison passes. `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
 
-End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal. Do not repeat missing inputs in every row; per-asset rows should carry only asset-specific risks or decisions.
+End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal; per-asset rows carry only asset-specific risks or decisions.
 
 Do not modify implementation code. Do not edit the approved mock. Do not produce final page copy. The parent craft agent owns implementation and final mock fidelity.

.agent/skills/impeccable/reference/degraded/finish-reviewer.md

@@ -3,35 +3,36 @@ This harness has no subagent capability, so you are running this role inline. St
 
 # Impeccable Finish Reviewer
 
-You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You do not edit anything; the parent agent applies your fixes.
+You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You edit nothing; the parent applies your fixes.
 
-You have no browser. Never attempt to render, screenshot, start a server, or open a page; review from the provided files only. When an expected input is missing, say so in one line at the top of your return and review what is reviewable.
+You have no browser. Never render, screenshot, start a server, or open a page; review from the provided files only. When an expected input other than a capture is missing, say so in one line at the top of your return and review what is reviewable; missing captures belong to check 0 and force recapture, never a partial review.
 
-A hard turn ceiling ends the run without warning; a run that ends before the five sections are written returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads into each turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
+A hard turn ceiling ends the run without warning; a run that ends before its contracted sections are written (five, or the single recapture section) returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads per turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
 
 ## Input Contract
 
-Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, which live in `.impeccable/review/` (on the web, `desktop.png` and `mobile.png`; on native, device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive); a screenshot path the calling brief names is authoritative when the file exists, and `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent; the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths and, on a comp-led build, the approved comp path (a code-led build has no approved comp; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing in this file that binds “the approved comp” binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet also carries the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor and judge every check in the platform's own conventions, the screenshots are device captures rather than browser viewports, and your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
+Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, in `.impeccable/review/` (web: `desktop.png` and `mobile.png`; native: device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive). A screenshot path the calling brief names is authoritative when the file exists; `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent. Also expect: the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); the PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths; on a comp-led build the approved comp path (a code-led build has none; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing here that binds "the approved comp" binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet adds the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor, judge every check in the platform's own conventions, treat the screenshots as device captures, and know your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
 
 ## Checks, in order
 
-1. **Persistence.** PRODUCT.md exists. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too, the surface brief naming the approved comp or an `approved` flag in its sidecar; comp-round comps with no recorded pick mean the approval point was skipped, and that is a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and they imply no approval whatever the build path; a code-led build has no comp round at all.
-2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element, and its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Two rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement, because medium is part of the promise. When no approved comp was supplied, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality, CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never actually renders, as contradicted on its face; imitation material is the single most reliable mark of machine-made design. A critique-reference comp, when one arrived on such a build, is provocation rather than spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is the question of what the image dared that the build did not, and the dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. In every material_fixes list, a fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
+0. **Evidence.** Before any other check, verify the required captures exist and every capture is valid. Required: the platform's full viewport set (web: `desktop.png` and `mobile.png`; native: one capture per shipped device class), plus every capture the calling brief names as required, a reported user viewport (`user-<width>.png`) included. Valid: no black or blank regions, content matching what the filename claims (a visit capture showing the About section is invalid), the document top visible where the file claims a full page, dimensions that make sense for the named viewport. A required capture that is absent fails exactly like one that is malformed: a viewport nobody captured is a viewport nobody inspected, and it cannot ship. When any capture fails, the whole review changes shape: return `disposition: recapture` as the first line, then one section, `recapture`, listing each missing or invalid file and what a valid capture of it shows, and stop. Never build a matrix on malformed evidence; a verdict derived from a broken capture launders the breakage into an approval, and the parent owes you a full re-review on valid captures, not a scoring round.
+1. **Persistence.** PRODUCT.md exists. On a comp-led build, `.impeccable/review/hero-repro.png` exists: the hero reproduction checkpoint's capture at the comp's own dimensions; its absence means the reproduction phase ran unproven, a material finding. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too: the surface brief naming the approved comp, or an `approved` flag in its sidecar. Comp-round comps with no recorded pick mean the approval point was skipped, a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and imply no approval whatever the build path; a code-led build has no comp round at all.
+2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element; its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Three rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement; medium is part of the promise. GROUND: the page field's value and temperature against the comp's, sampled from pixels on both sides when tooling allows rather than judged from memory, and read as the net on-screen result where a texture or tile paints over the base color; a ground warmer or cooler than the comp's is contradicted however faithfully the layout matches, and drift toward the rendition prior (warm cream on light grounds, blue-black slate on dark) is the direction to hunt. With no approved comp, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality (CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never renders) as contradicted on its face; imitation material is the single most reliable mark of machine-made design. GROUND narrows rather than lapses: with no comp to sample, a color OWN-WORLD names is the target and the same warmer-or-cooler judgment applies; when OWN-WORLD names none, there is no GROUND authority, and the review says so in place of a verdict, because a target the reviewer invents turns the check into taste. A critique-reference comp on such a build is provocation, not spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is what the image dared that the build did not, and dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. A fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
 3. **Ceiling.** Against the QUALITY BAR card: name the world's native devices the build left unused, frame, depth, lettering treatment, ornament density, motion. The card governs commitment and finish, never composition.
-4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped and that is a material fix ahead of any craft point. Then, for each of the five blocks, does the render keep the promise? Apply the memory test to the first viewport.
+4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped, a material fix ahead of any craft point. Then, for each of the five blocks: does the render keep the promise? Apply the memory test to the first viewport.
 5. **Truth.** Demonstration data authored and labeled synthetic; no invented commercial claims; unanswered claims present as marked placeholders, not omissions. Every image-native region of the approved comp shipped as a real asset, not a gradient standing in for one, and every produced asset visibly present in the screenshots; an asset applied at near-zero opacity or buried behind other paint is a compliance token, not a shipped material.
-6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp, because the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
+6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp: the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
 
 Do not run a second detector pass; mechanical findings belong to the parent's hooks.
 
 ## Disposition
 
-The first line of your return is `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. It is derived, never felt: rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
+The first line of your return is `disposition: recapture`, `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. These four words are the whole vocabulary; never invent another. The word is derived, never felt: recapture when the evidence check failed, rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
 
 ## Output Contract
 
-Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). Missing inputs are named in one line above the sections. No praise, no summary prose.
+Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). A recapture return replaces the five sections with the single `recapture` section from check 0. Missing inputs are named in one line above the sections. No praise, no summary prose.
 
 ## Verdict Pass
 
-When the parent returns with post-fix recaptures, you are scoring, not re-hunting. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths for this round; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open; unresolved or partial material findings can never recompute to ship.
+When the parent returns with post-fix recaptures, you are scoring, not re-hunting. Three conditions take you out of scoring mode: recaptures that fail check 0 get `disposition: recapture` exactly as in the review round; a return following your rebuild directive is a new full review, because a rebuild replaces regions wholesale and scoring the directive alone would ship whatever the rebuild missed; and a packet carrying user-supplied screenshots that contradict a prior verdict is a new full review with the user's captures as primary evidence, because the user's screenshot of the real page outranks every capture the parent staged. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open, in the same four-word vocabulary. Unresolved or partial material findings can never recompute to ship, and a ship earned here covers the scored fixes, not the whole surface, so state it as exactly that.

.agent/skills/impeccable/reference/hooks.md

@@ -2,9 +2,9 @@
 
 Manage the **design detector hook** for the current project.
 
-The hook runs the impeccable design detector on direct file edits to design-relevant files (`.tsx`, `.jsx`, `.html`, `.vue`, `.svelte`, `.astro`, `.css`, `.scss`, `.sass`, `.less`, `.ts`, `.js`). Claude Code, Codex, and GitHub Copilot use a post-tool-use hook and push a short system reminder into the agent's context after the edit; findings get a correction prompt, pending issues get a re-nudge, and clean UI-ish files get a short ack unless quiet mode is on (`hook.quiet` in config). Plain `.ts` and `.js` files are still scanned, but stay quiet unless the detector finds something. Cursor uses `preToolUse` to block bad proposed writes before they land and stays silent when it allows a clean write.
+The hook runs the impeccable design detector on direct file edits to design-relevant files (`.tsx`, `.jsx`, `.html`, `.vue`, `.svelte`, `.astro`, `.css`, `.scss`, `.sass`, `.less`, `.ts`, `.js`). Claude Code, Codex, and GitHub Copilot use a post-tool-use hook and push a short system reminder into the agent's context after the edit; findings get a correction prompt, pending issues get a re-nudge, and clean UI-ish files get a short ack unless quiet mode is on (`hook.quiet` in config). Plain `.ts` and `.js` files are still scanned, but stay quiet unless the detector finds something. Cursor uses `preToolUse` to block bad proposed writes before they land and stays silent when it allows a clean write. Grok Build fires the same PostToolUse scan to mark touched files, then surfaces findings on Stop `additionalContext`. Do not expect a Grok per-edit reminder: Grok discards that stdout.
 
-The detector rules run in two tiers. The per-edit hook surfaces only the immediate tier: mechanical, unambiguous problems worth interrupting an edit for, such as broken images, overflowing or clipped content, contrast and legibility failures, gradient text, glow shadows, and design-system drift. Everything else (copy cadence, palette and typography taste, layout rhythm) is deferred to a deep pass on the `Stop` hook event, which runs the full rule set over every UI file touched in the session and surfaces the remaining findings once, deduplicated against what the per-edit pass already reported. A session with nothing left to report stops silently. Set `hook.perEditRules` to `"all"` in `.impeccable/config.json` to restore the full rule set on every edit. The Stop deep pass is wired for Claude Code and Codex, which both dispatch a native `Stop` hook event. Cursor does not get one (its stop hook is not consistently dispatched; the pre-write gate covers it), and GitHub Copilot's stop-style events do not feed context back to the model, so they keep the full detector per edit.
+The detector rules run in two tiers. The per-edit hook surfaces only the immediate tier: mechanical, unambiguous problems worth interrupting an edit for, such as broken images, overflowing or clipped content, contrast and legibility failures, gradient text, glow shadows, and design-system drift. Everything else (copy cadence, palette and typography taste, layout rhythm) is deferred to a deep pass on the `Stop` hook event, which runs the full rule set over every UI file touched in the session and surfaces the remaining findings once, deduplicated against what the per-edit pass already reported. A session with nothing left to report stops silently. Set `hook.perEditRules` to `"all"` in `.impeccable/config.json` to restore the full rule set on every edit. The Stop deep pass is wired for Claude Code, Codex, and Grok Build, which dispatch a native `Stop` hook event. Cursor does not get one (its stop hook is not consistently dispatched; the pre-write gate covers it), and GitHub Copilot's stop-style events do not feed context back to the model, so they keep the full detector per edit. Grok also fires an observe-only Stop with `reason: "shutdown"` after `end_turn`; skip that one, scan only `end_turn`.
 
 Every hook is a mechanical pass. The reflexes no scanner catches live in [craft-floor.md](craft-floor.md), which the skill loads before it edits UI, so they apply whether or not a hook is wired. A session with no automatic hook gets one `MANUAL_DETECTOR_REQUIRED` directive from `context.mjs` asking for a single detector run at the end.
 
@@ -14,7 +14,7 @@ Declare server-side template extensions under **`detector.extensions`** when the
 
 Manual `npx impeccable detect` scans use the same project filter config by default: `detector.ignoreRules`, `detector.ignoreFiles`, `detector.ignoreValues`, and `detector.designSystem.enabled`. `hook.enabled` only controls automatic hook execution, not manual CLI scans. Use `npx impeccable detect --no-config ...` for a raw detector run that ignores project config/context. Use `npx impeccable ignores ...` for direct CLI CRUD on the same detector ignores.
 
-Supported harnesses: Claude Code (`.claude/settings.local.json` in the project, which is gitignored so the hook stays machine-local; a hook you move into the shared `settings.json` is honored in place too), Codex (`.codex/hooks.json` in the project), Cursor (`.cursor/hooks.json` in the project), and GitHub Copilot (`.github/hooks/impeccable.json` in the project, a team-shared committed file that both the Copilot CLI and the cloud agent read). For the Copilot CLI, repo-level hooks fire once `.github/hooks/impeccable.json` is committed to the repository's default branch.
+Supported harnesses: Claude Code (`.claude/settings.local.json` in the project, which is gitignored so the hook stays machine-local; a hook you move into the shared `settings.json` is honored in place too), Codex (`.codex/hooks.json` in the project), Cursor (`.cursor/hooks.json` in the project), Grok Build (`.grok/hooks/impeccable.json` in the project; requires `/hooks-trust` or `--trust`), and GitHub Copilot (`.github/hooks/impeccable.json` in the project, a team-shared committed file that both the Copilot CLI and the cloud agent read). For the Copilot CLI, repo-level hooks fire once `.github/hooks/impeccable.json` is committed to the repository's default branch.
 
 On **Cursor**, `preToolUse` checks proposed Write/Edit/Shell write content and denies only when the real detector finds an issue. The denial message is visible to the agent as the tool error, so the agent can reconsider before the bad write lands.
 
@@ -44,7 +44,7 @@ The first argument is the action. Defaults to `status`.
    ```
 
 3. If `<action>` is `off`, follow up with a one-line note: "Done. New edits will not trigger the design hook in this project until you run `/impeccable hooks on`."
-4. If `<action>` is `on`, follow up with: "Done. The design hook will fire after the next Edit/Write/MultiEdit on a UI file."
+4. If `<action>` is `on`, follow up with: "Done. The design hook will fire after the next Edit/Write on a UI file."
 5. If `<action>` is `ignore-value`, `ignore-file`, or `ignore-rule`, just print the script output. The default scope is shared `.impeccable/config.json`; add `--local` only when the user explicitly asks for a private exception.
 6. If `<action>` is `status`, just print the script output. Do not add commentary unless the user asked a follow-up question.
 

.agent/skills/impeccable/reference/new-work.md

@@ -1,34 +1,34 @@
 # New visual work
 
-Use this flow when making a new surface or replacing a visual identity. PRODUCT.md owns product truth. DESIGN.md owns durable visual decisions. A surface brief keeps strategy that belongs only to one route or artifact. Complete [init.md](init.md) first when PRODUCT.md is missing; a missing DESIGN.md does not route back to init.
+Use this flow for a new surface or a replacement visual identity. PRODUCT.md owns product truth. DESIGN.md owns durable visual decisions. A surface brief keeps strategy that belongs to one route or artifact. Complete [init.md](init.md) first when PRODUCT.md is missing; a missing DESIGN.md does not route back to init.
 
 ## 1. Decide what is already true
 
 Read DESIGN.md, representative code, tokens, components, and assets.
 
 - **Redesign:** preserve product truth, content, function, constraints, and explicit brand commitments; replace the old visual world rather than polishing it. The old look is evidence of what the subject is, not authority over what it becomes.
-- **Established world:** inherit it. A missing DESIGN.md does not erase a coherent identity already present in code; document that identity instead of inventing a replacement.
-- **Incomplete brand:** preserve confirmed assets and recognizable traits, then help the user expand the system for this new surface.
+- **Established world:** inherit it. A missing DESIGN.md does not erase a coherent identity already in code; document that identity instead of inventing a replacement.
+- **Incomplete brand:** preserve confirmed assets and recognizable traits, then expand the system with the user for this surface.
 - **No visual authority:** create a new world with the user.
 
-A section, component, feature, or state inside an established surface inherits that surface. Do not turn a local addition into a new identity exercise.
+A section, component, feature, or state inside an established surface inherits that surface. Never turn a local addition into a new identity exercise.
 
 ## 2. Ask what will change the work
 
 Ask one round of two or three related questions through the structured question tool when available. Skip settled facts; a precise request may need only a compact confirmation.
 
-- **Persuade:** clarify who must act, what they should believe, and which real proof, content, or assets can earn that belief.
-- **Operate:** clarify the task, information, important states, frequency, and constraints.
-- **Read:** clarify the reader's question, source material, structure, and wayfinding.
-- **Experience:** clarify what leads, how exploration unfolds, and which interaction or transition matters.
+- **Persuade:** who must act, what they should believe, which real proof, content, or assets earn that belief.
+- **Operate:** the task, information, important states, frequency, constraints.
+- **Read:** the reader's question, source material, structure, wayfinding.
+- **Experience:** what leads, how exploration unfolds, which interaction or transition matters.
 
-Across modes, ask what success looks like, what must remain untouched, and what would make a polished result feel wrong. Do not ask for CSS values or canned aesthetic lanes.
+Across modes, ask what success looks like, what must remain untouched, and what would make a polished result feel wrong. Never ask for CSS values or canned aesthetic lanes.
 
 ## 3. Choose the right amount of invention
 
 ### Extend an existing surface
 
-Inherit its world and composition. Resolve only the new purpose, content, hierarchy, states, interaction, and how the addition joins the surrounding experience. Do not run a concept tournament or change DESIGN.md unless the user approves a durable system change.
+Inherit its world and composition. Resolve only the new purpose, content, hierarchy, states, interaction, and how the addition joins the surrounding experience. No concept tournament, and no DESIGN.md change unless the user approves a durable system change.
 
 ### Create a whole surface inside an established world
 
@@ -36,21 +36,21 @@ Keep the visual system fixed. Derive five to seven materially different structur
 
 `node .agent/skills/impeccable/scripts/concept-seed.mjs --scope surface --mode <mode>`
 
-The script deals three of your structures to the table; the dice decide which three reach the user, so the ranking rut stays broken while the user still holds a real choice. Present the three dealt structures on the decision page as full cards of equal salience, the dealt lead carrying kicker THE ROLL, with steer and re-roll; the user locks one in. No canon card and no pick card at surface scope: the world is settled, so every card visualizes composition, not identity. With image generation available and a comp-led default (the build-path paragraph below: `.impeccable/config.json`, the toggle handles the exception), each card declares a `comp` under `.impeccable/mocks/decision/`, generated after serving in reading order under the comp discipline in [visualize.md](visualize.md); anchor each of these comps on the established identity by passing a captured screenshot of a representative existing page as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`) beside a prompt that leads with the new surface's structure and names DESIGN.md's palette, type, and component character, because a prose paraphrase of a design system drifts where a pixel reference does not. Without image generation, or under a code-led default, each card instead carries a `wireframe` layout schematic (see `serve-question.mjs --schema`) that the page draws itself. Locking a card is the approval and sets the build path: a locked comp builds comp-led with that comp as the approved comp, discharging [visualize.md](visualize.md)'s three-option round with no second approval point; a locked wireframe builds code-led, its ambition carried by the direction contract. Never run the script for a local extension or a precisely specified narrow request; shape those directly.
+The script deals three of your structures; the dice pick which three reach the user, breaking the ranking rut while the user keeps a real choice. Present them on the decision page as full cards of equal salience, the dealt lead under kicker THE ROLL, with steer and re-roll; the user locks one. No canon card and no pick card at surface scope: the world is settled, so every card visualizes composition, not identity. With image generation and a comp-led default (`.impeccable/config.json`; the build-path paragraph below), each card declares a `comp` under `.impeccable/mocks/decision/`, generated after serving, in reading order, under [visualize.md](visualize.md)'s comp discipline. Anchor each comp on the established identity: pass a screenshot of a representative existing page as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`) with a prompt that leads with the new surface's structure and names DESIGN.md's palette, type, and component character; prose paraphrases of a design system drift, pixel references do not. Without image generation, or under a code-led default, each card carries a `wireframe` schematic (`serve-question.mjs --schema`) the page draws itself. Locking a card is the approval and sets the build path: a locked comp builds comp-led with that comp as the approved comp, discharging [visualize.md](visualize.md)'s three-option round with no second approval point; a locked wireframe builds code-led, its ambition carried by the direction contract. Never run the script for a local extension or a precisely specified narrow request; shape those directly.
 
 ### Create or replace the visual world
 
-1. Name the product's unique mechanism in one sentence, the audience's real scene, its cultural home, and what this first surface must prove. Note the page this category always ships and its predictable opposite; name both as the rut and keep them out of the seven-candidate list. A brief that paints its own picture, a product name, a titled artifact, a governing metaphor, adds its literal reading to the rut: spend at most one candidate on it and derive the rest from elsewhere in the audience's world.
-2. From that cultural world, list seven concrete visual systems, artifacts, places, or rituals the audience knows by heart, each with one line on why it resonates and can carry the mechanism, ordered by resonance. The audience's world includes its graphic and screen traditions, not only its physical objects: the notation, publications, identity programs, data graphics, and interfaces it reads daily; a nameable abstract system (a school of poster, a documentation standard) is as concrete a candidate as any artifact. What would this thing look like as a physical object; what did its world look like before the web? Near-duplicates count once. When more than three of the seven share one material family, the derivation stopped at the subject's most obvious artifact; dig until the list spans at least three families.
+1. Name the product's unique mechanism in one sentence, the audience's real scene, its cultural home, and what this first surface must prove. Note the page this category always ships and its predictable opposite; both are the rut, kept out of the seven-candidate list. A brief that paints its own picture, a product name, a titled artifact, a governing metaphor, adds its literal reading to the rut: spend at most one candidate on it and derive the rest from elsewhere in the audience's world.
+2. From that cultural world, list seven concrete visual systems, artifacts, places, or rituals the audience knows by heart, each with one line on why it resonates and can carry the mechanism, ordered by resonance. The audience's world includes its graphic and screen traditions, not only its physical objects: the notation, publications, identity programs, data graphics, and interfaces it reads daily. A nameable abstract system (a school of poster, a documentation standard) is as concrete a candidate as any artifact. What would this thing look like as a physical object; what did its world look like before the web? Near-duplicates count once. When more than three of the seven share one material family, the derivation stopped at the subject's most obvious artifact; dig until the list spans at least three families.
 3. Turn that material into complete directions: each joins a reusable visual world to a concrete first-surface experience.
-4. Run `node .agent/skills/impeccable/scripts/concept-seed.mjs --scope direction --mode <mode>` and follow what it prints. This step has no substitute and no skip condition: on a new or replacement world, writing artifact code before this script has run and its assignment is acknowledged is a contract violation, whatever the harness, the model, or the time pressure, because the roll is the mechanism that keeps every run from converging on the category default. The script assigns which direction gets built and deals catalog challengers. Fuse each challenger before judging it: the challenger supplies the form and its system grammar, the product supplies every fact, and clarity wins conflicts. Weigh fused challengers against the assigned direction on exactly two axes, audience identification and product clarity; losing to strong grounded material is a valid outcome, and beating a thin or tool-monoculture list is the point. The weighing closes with a verdict per challenger, decided before any borrowing is considered: wins (beats the assigned direction on both axes; it becomes the build candidate), competitive (holds one axis; it stays a full alternate), or declined (loses both). A declined challenger is not spent: name the one discipline of its system the assigned direction lacks, and raise the assigned direction to match before presenting it. A donation transfers ambition and system discipline (a palette's total commitment, a grid's density courage, a form's structural honesty), never the challenger's clothes; a motif lifted from a declined world is a costume note, not a raise, and one world owns the page. Write each raise into the presented direction as its own line, named for its donor; a raise nobody can read did not happen.
-5. Present one direction, fully committed and already raised by the hand it beat, its raises visible as named lines: its world, first viewport, visitor path, signature interaction, cross-surface reach, and honest risk. Alongside it, route each dealt challenger by its verdict: winning and competitive challengers are full alternates carrying their QUALITY BAR cards and one-line case, while declined challengers render demoted, compact and quiet, each carrying its verdict plus what the direction kept from it, never full-size and never silently dropped, each still adoptable on request. The verdict informs the user's choice, it never pre-empts it; the demoted row is the hand's proof of judgment, showing why the dealt worlds made the presented direction better. A hand holds at most three full-card challengers: when the roll deals more, the three strongest join the hand and the rest wait in the re-roll pool, noted in one line; dropping a challenger from the hand itself takes a named product-truth failure, disclosed. Add one card for your own top-ranked grounded candidate when it is not the assigned direction, kicker IMPECCABLE’S PICK, same anatomy as every card, with an honest risk line naming its familiarity when true: the strongest grounded direction is often the one most runs in this category land on, and the user deciding that trade is the point of showing it. Familiar and effective is a legitimate destination, not a failure of nerve; the pick card and the standing exit serve it at two depths. One pick card, never two, never a ranked list: the rest of your grounded candidates stay yours, because a lineup of them hands selection back to a taste function and invites the safest card. The pick never takes the lead position, and when the dice assign your top candidate there is no pick card; the assigned card notes it also topped your list. Add re-roll with an optional one-line steer, offered in three registers: plain (a fresh hand, same spread), safer (the familiar register: your remaining conventional grounded candidates plus the canon against named competitors), and bolder (foreign forms only, at full commitment). A register is the user's steering on the familiar-to-bold axis, never yours to pre-select; when the answer carries one, re-run the seed with `--register <value>` and the next `--reroll` round, and follow what it prints. A user saying "bolder" or "safer" while a direction round is open means these registers, never the bolder or harden commands. The two channels share this structure and differ only in richness: cards and boards on the decision page, names and one-liners through the structured tool; the structured tool's option list carries the assigned direction, the pick, the winning and competitive challengers, and the standing exit as its last option, while declined challengers fold into the assigned option's description as their kept lines, so the raise survives the text channel too.
+4. Run `node .agent/skills/impeccable/scripts/concept-seed.mjs --scope direction --mode <mode>` and follow what it prints. No substitute, no skip: on a new or replacement world, writing artifact code before this script has run and its assignment is acknowledged is a contract violation, whatever the harness, the model, or the time pressure; the roll is what keeps every run from converging on the category default. The script assigns the direction to build and deals catalog challengers. Fuse each challenger before judging it: the challenger supplies the form and its system grammar, the product supplies every fact, clarity wins conflicts. Weigh fused challengers against the assigned direction on exactly two axes, audience identification and product clarity. Losing to strong grounded material is a valid outcome; beating a thin or tool-monoculture list is the point. Close with a verdict per challenger, decided before any borrowing: wins (beats the assigned direction on both axes; becomes the build candidate), competitive (holds one axis; stays a full alternate), or declined (loses both). A declined challenger is not spent: name the one discipline of its system the assigned direction lacks, and raise the assigned direction to match before presenting it. A donation transfers ambition and system discipline (a palette's total commitment, a grid's density courage, a form's structural honesty), never the challenger's clothes; a lifted motif is a costume note, not a raise, and one world owns the page. Write each raise into the presented direction as its own line, named for its donor; a raise nobody can read did not happen.
+5. Present one direction, fully committed and already raised by the hand it beat, raises visible as named lines: world, first viewport, visitor path, signature interaction, cross-surface reach, honest risk. Route each challenger by verdict: winning and competitive challengers are full alternates with their QUALITY BAR cards and one-line case; declined challengers render demoted, compact and quiet, each carrying its verdict and what the direction kept from it, never full-size, never silently dropped, still adoptable on request. The verdict informs the user's choice, never pre-empts it; the demoted row is the hand's proof of judgment. A hand holds at most three full-card challengers: when the roll deals more, the three strongest join and the rest wait in the re-roll pool, noted in one line; dropping a challenger from the hand itself takes a named product-truth failure, disclosed. Add one card for your own top-ranked grounded candidate when it is not the assigned direction, kicker IMPECCABLE’S PICK, same anatomy as every card, with an honest risk line naming its familiarity when true: the strongest grounded direction is often where most runs in this category land, and the user deciding that trade is the point of showing it. Familiar and effective is a legitimate destination, not a failure of nerve; the pick card and the standing exit serve it at two depths. One pick card, never two, never a ranked list: a lineup of your candidates hands selection back to a taste function and invites the safest card. The pick never takes the lead position; when the dice assign your top candidate there is no pick card, and the assigned card notes it topped your list. Add re-roll with an optional one-line steer, in three registers: plain (a fresh hand, same spread), safer (your remaining conventional grounded candidates plus the canon against named competitors), bolder (foreign forms only, at full commitment). The register is the user's steering on the familiar-to-bold axis, never yours to pre-select; when the answer carries one, re-run the seed with `--register <value>` and the next `--reroll` round, and follow what it prints. A user saying "bolder" or "safer" while a direction round is open means these registers, never the bolder or harden commands. The two channels share this structure and differ only in richness: cards and boards on the decision page, names and one-liners through the structured tool, whose option list carries the assigned direction, the pick, the winning and competitive challengers, and the standing exit last; declined challengers fold into the assigned option's description as their kept lines, so the raise survives the text channel.
 
-The standing exit: every direction round offers one quiet, permanent alternative, the category standard, played straight. It is the user's door, never yours: never recommend it, never weigh it against the roll, never let it soften the dealt directions; the counterweights bind the unchosen default, not the chosen one. When the user takes it, in the canon action, a safer-steer, or plain words asking for the familiar or competitor-like path, convention becomes the commitment: ask once for two or three products this should sit alongside, make their craft level the bar, and execute the canon at full fidelity, without irony or smuggled quirk. A standing preference gets recorded as a brand commitment in PRODUCT.md. Re-roll eliminates every direction already shown, grounded and challenger alike; after two consecutive re-rolls, ask what quality is missing. You may re-roll on your own only on named factual grounds, when the assigned direction cannot carry the product's truth or task; taste is never grounds. The user may re-roll freely, and a user- or brief-pinned direction beats the roll, always. Present the decision visually: write an options payload with the assigned direction leading and its raised lines included, the pick card when one exists, the dealt challengers as alternates carrying their QUALITY BAR cards plus each challenger's verdict and kept line, re-roll with its safer and bolder registers, steer, plus canon enabled, and `buildPath` carrying the recorded default with `toggle: true` whenever image generation exists (the build-path paragraph below owns the details); a degraded roll with no challengers still uses the page, as a single text-only card with re-roll. Give every card the same anatomy, thesis, palette, materials, first viewport, honest risk, and the challengers' case lines (run the script with `--schema` for the exact shape); the page renders identity from these fields, routes declined challengers to a demoted row on its own, and a challenger's catalog image rides as labeled inspiration, never as the promise of the build. Author `canonCard` too: the category standard as one honest card with the same anatomy; the page keeps it subordinate, and the counterweights still bind you. Run `node .agent/skills/impeccable/scripts/serve-question.mjs --start --payload <file>` (run it with `--schema` first for the exact payload shape). It daemonizes, prints the page URL and a key, and exits immediately; now open that URL for the user, in-app browser first, then the system opener, then showing the URL. Collect the choice with `--wait --key <key>`, repeating while it exits 3; the ANSWER prints as JSON. Exit 4 means the page was closed without an answer: re-present once through the structured question tool, and with no answer there either, proceed unattended with the assigned direction and state the assumptions. A harness that can leave a shell blocked in the background may instead run the script without `--start` and let it auto-open and block. The fallback to the structured question tool is never yours to predict: run the script, and only exit code 2 from starting it routes the decision there; treat that exit as the fallback, never as an error to retry.
+The standing exit: every direction round offers one quiet, permanent alternative, the category standard, played straight. It is the user's door, never yours: never recommend it, never weigh it against the roll, never let it soften the dealt directions; the counterweights bind the unchosen default, not the chosen one. When the user takes it (the canon action, a safer-steer, or plain words asking for the familiar or competitor-like path), convention becomes the commitment: ask once for two or three products this should sit alongside, make their craft level the bar, and execute the canon at full fidelity, without irony or smuggled quirk. Record a standing preference as a brand commitment in PRODUCT.md. Re-roll eliminates every direction already shown, grounded and challenger alike; after two consecutive re-rolls, ask what quality is missing. Re-roll on your own only on named factual grounds, when the assigned direction cannot carry the product's truth or task; taste is never grounds. The user may re-roll freely, and a user- or brief-pinned direction beats the roll, always. Present the decision visually: write an options payload with the assigned direction leading, its raised lines included; the pick card when one exists; the dealt challengers as alternates with their QUALITY BAR cards, verdicts, and kept lines; re-roll with its safer and bolder registers; steer; canon enabled; and `buildPath` carrying the recorded default with `toggle: true` whenever image generation exists (details in the build-path paragraph below). A degraded roll with no challengers still uses the page, as a single text-only card with re-roll. Give every card the same anatomy: thesis, palette, materials, first viewport, honest risk, and the challengers' case lines (`--schema` prints the exact shape); the page renders identity from these fields, demotes declined challengers to their row on its own, and a challenger's catalog image rides as labeled inspiration, never the promise of the build. Author `canonCard` too: the category standard as one honest card, same anatomy; the page keeps it subordinate, and the counterweights still bind you. Run `node .agent/skills/impeccable/scripts/serve-question.mjs --start --payload <file>` (`--schema` first for the payload shape). It daemonizes, prints the page URL and a key, and exits; open that URL for the user, in-app browser first, then the system opener, then showing the URL. Collect the choice with `--wait --key <key>`, repeating while it exits 3; the ANSWER prints as JSON. An ANSWER of `{"optionId":"reroll"}` keeps the server alive and the page open on a loading hand: rerun concept-seed with the same `--scope` and `--mode` plus `--from <seed-key> --reroll <n>` (1 on the first re-roll, counting up), build the next payload, deliver it with `--update --key <same key> --payload <file>`, then return to `--wait` on that key. Never `--start` a second server or fall back to chat here: either strands the open tab on a hand that never arrives. Exit 4 means the page closed unanswered: re-present once through the structured question tool, and with no answer there either, proceed unattended with the assigned direction and state the assumptions. A harness that can leave a shell blocked in the background may run the script without `--start` and let it auto-open and block. Never predict the fallback: run the script, and only exit code 2 from starting it routes the decision to the structured tool; that exit is the fallback, never an error to retry.
 
-When image generation exists, every card also declares a `comp` path under `.impeccable/mocks/decision/`, the canon card included. Where the harness sandboxes its shell, start the page through the least-sandboxed command path it offers: a sandboxed shell cannot bind the board's port, and the first-attempt failure costs a retry every session. Serve the page first, then produce the comps; the page shimmer-waits per slot and the user may answer before they land. Each card's image is that direction's north-star comp at full fidelity, produced under the comp discipline in [visualize.md](visualize.md): the requested surface's first viewport, structure-led prompt, real product name and real content, no invented commercial claims, in that card's own palette, type character, and material world, committed all the way; visualize.md's self-checks bind decision comps identically. Generation takes the same time at any fidelity, so an unfinished draft pays draft quality for comp cost; fairness between cards comes from equal fidelity in each card's own grammar, one surface, one aspect, never from shared unfinishedness. The frame's aspect is the surface's own: a native app or mobile-first surface comps portrait at its device viewport, a desktop web surface landscape, and the decision page adapts to either, so a phone screen comped landscape is a broken frame, not a neutral default. Produce in the order the user reads, the assigned card, then the pick, then the full-card hand, then canon, each file written with its prompt sidecar the moment it is done, so a re-roll's spend front-loads onto the cards read first; declined challengers get no comp, their catalog thumb is their face. When the harness runs subagents in parallel, fan the set out as one agent per card: each spawn is the shipped asset producer with a single-comp packet, that card's fields, PRODUCT.md, the shared frame, and the card's declared path, up to four in flight at once. A slot still empty when its agent returns is regenerated inline, and a slot still empty when the user answers is dropped without ceremony; no other supervision is owed. Without parallel subagents, generate in the main thread after serving, in the same reading order, and let the harness's own generation display carry the progress; the wait for the answer follows the last file. The chosen card's comp is not spent by the choice: on a comp-led build it enters the comp round as compositional option one, and on a code-led build it returns at the finish review as the critique reference, what the image dared that the build did not. The unchosen comps stay in `.impeccable/mocks/decision/` as the round's spent hand; they carry no approval and imply none. With no image generation, the cards carry their identity in palette chips and facts, and that page is complete, not a lesser version; the page then also demotes every challenger's catalog art to a labeled thumbnail on its own, because salience must encode the verdict, never the accident of which cards have images.
+When image generation exists, every card also declares a `comp` path under `.impeccable/mocks/decision/`, the canon card included. Where the harness sandboxes its shell, start the page through the least-sandboxed command path it offers: a sandboxed shell cannot bind the board's port, and the first-attempt failure costs a retry every session. Serve the page first, then produce the comps; the page shimmer-waits per slot and the user may answer before they land. Each card's image is that direction's north-star comp at full fidelity under [visualize.md](visualize.md)'s comp discipline: the requested surface's first viewport, structure-led prompt, real product name and real content, no invented commercial claims, in that card's own palette, type character, and material world, committed all the way. Generation takes the same time at any fidelity, so an unfinished draft pays comp cost for draft quality; fairness between cards is equal fidelity in each card's own grammar, one surface, one aspect, never shared unfinishedness. The frame's aspect is the surface's own: portrait at device viewport for a native app or mobile-first surface, landscape for desktop web; the decision page adapts to either, and a phone screen comped landscape is a broken frame, not a neutral default. Produce in reading order, the assigned card, then the pick, then the full-card hand, then canon, each file written with its prompt sidecar the moment it is done, so a re-roll's spend front-loads onto the cards read first; declined challengers get no comp, their catalog thumb is their face. With parallel subagents, fan out one agent per card: each spawn is the shipped asset producer with a single-comp packet, that card's fields, PRODUCT.md, the shared frame, and the card's declared path, up to four in flight. Regenerate inline any slot still empty when its agent returns; drop without ceremony any slot still empty when the user answers. No other supervision is owed. Without parallel subagents, generate in the main thread after serving, same order, and let the harness's own generation display carry the progress; the wait for the answer follows the last file. The chosen card's comp is not spent by the choice: comp-led, it enters the comp round as compositional option one; code-led, it returns at the finish review as the critique reference, what the image dared that the build did not. Unchosen comps stay in `.impeccable/mocks/decision/` as the round's spent hand; they carry no approval and imply none. With no image generation, cards carry their identity in palette chips and facts, and that page is complete, not a lesser version; the page then also demotes every challenger's catalog art to a labeled thumbnail on its own, because salience must encode the verdict, never the accident of which cards have images.
 
-The execution contract, comp-led or code-led, is a workflow preference, not a per-surface decision, so no round asks it: the recorded default rides every round and the page's toggle handles the exception. Read the default from `.impeccable/config.json` (`buildPath`), with the gitignored `.impeccable/config.local.json` winning where one machine differs from the team's committed value; with neither, comp-led is the default whenever image generation exists. Author every direction and surface payload with `buildPath: { "value": <default>, "toggle": true }`; the page renders a footer toggle with the trade stated beside it, and the ANSWER returns `buildPath` plus `buildPathFlipped`. A flipped value binds that session only and is never written back, with one exception, and it is the only thing inside a round that earns a question about this preference (init records it up front on projects that get the chance): when `buildPathFlipped` comes back true on a project that records no `buildPath` at all, ask once after the round closes whether to keep it as the standing default. Either answer ends in a write to `.impeccable/config.json`; the answer picks the value, never whether to record one. Yes writes the flipped value, and "no, just this once" writes the value they flipped away from, which is the standing default they just confirmed by declining. Ask on the flip and never on the untouched default, because a user who left the toggle alone has told you nothing. A declined offer nothing writes down is an offer the next session makes again. When the user asks in words to change the standing default, update the file without asking. **Comp-led**: the chosen card's comp is law, generated before building when it does not exist yet, and the finish review audits the build against it; boldest composition on the table, fix rounds expected; comp-led makes the comp non-optional, no silent skipping. **Code-led**: no comp of this page and no apology for it; the QUALITY BAR boards still calibrate finish, and the ambition moves into the written contract, the FIRST VIEWPORT block plus a named signature interaction and motion grammar, which the finish reviewer audits in behavior; code-led is not a discount on commitment, the direction still lands fully committed in code. A code-led round still declares each card's comp path as a flip reserve: when the user flips the toggle to comp mid-round, `--wait` returns once with BUILD PATH FLIPPED while the page shimmers the slots; generate each open card's comp into its declared path then, lead first, and wait again. The flip back is free, and a comp that already rendered rides at the finish review as the critique reference. Without image generation there is no toggle and no choice: code-led is the only path, stated in one line rather than asked. The old two-card execution-contract round is retired; `followup: true` remains the general mechanism for delivering any later round over the same table via `--update`.
+The execution contract, comp-led or code-led, is a workflow preference, not a per-surface decision; no round asks it. The recorded default rides every round and the page's toggle handles the exception. Read the default from `.impeccable/config.json` (`buildPath`), the gitignored `.impeccable/config.local.json` winning where one machine differs from the team's committed value; with neither, comp-led is the default whenever image generation exists. Author every direction and surface payload with `buildPath: { "value": <default>, "toggle": true }`; the page renders a footer toggle with the trade stated beside it, and the ANSWER returns `buildPath` plus `buildPathFlipped`. A flipped value binds that session only and is never written back, with one exception, the only question this preference ever earns inside a round (init records it up front on projects that get the chance): when `buildPathFlipped` comes back true on a project that records no `buildPath` at all, ask once after the round closes whether to keep it as the standing default. Either answer writes `.impeccable/config.json`; the answer picks the value, never whether to record one. Yes writes the flipped value; "no, just this once" writes the value they flipped away from, the standing default they just confirmed by declining. Ask on the flip, never on the untouched default: a user who left the toggle alone told you nothing. A declined offer nothing writes down is an offer the next session makes again. When the user asks in words to change the standing default, update the file without asking. **Comp-led**: the chosen card's comp is law, generated before building when it does not yet exist, and the finish review audits the build against it; boldest composition on the table, fix rounds expected, and the comp is non-optional, no silent skipping. **Code-led**: no comp of this page and no apology for it; the QUALITY BAR boards still calibrate finish, and the ambition moves into the written contract, the FIRST VIEWPORT block plus a named signature interaction and motion grammar, which the finish reviewer audits in behavior; code-led is not a discount on commitment. A code-led round still declares each card's comp path as a flip reserve: when the user flips the toggle to comp mid-round, `--wait` returns once with BUILD PATH FLIPPED while the page shimmers the slots; generate each open card's comp into its declared path then, lead first, and wait again. The flip back is free, and a comp that already rendered rides at the finish review as the critique reference. Without image generation there is no toggle and no choice: code-led is the only path, stated in one line rather than asked. The old two-card execution-contract round is retired; `followup: true` remains the general mechanism for delivering any later round over the same table via `--update`.
 
 Catalog worlds are working systems, not mood references. When one survives, carry its palette and material, type and composition, topology, controls and state, and responsive rules into the product. When the source is itself an interface language, commit to its native grammar across navigation, content, controls, and states. Open the QUALITY BAR board and hero for the world you build the moment the choice lands, even if you viewed another card earlier; the ANSWER line names the chosen card's images (when the harness only reads files or runs sandboxed, download them into the workspace and open the relative path; sandboxed viewers reject absolute paths outside it). They set the craft level the build must reach, a rendered reference's finish, commitment, and art direction, never the composition; your surface serves this product.
 
@@ -64,11 +64,11 @@ Pick a color strategy before picking colors: Restrained (neutrals plus one accen
 
 Choose faces like objects from the subject's world, in the mode's register. Operate and Read surfaces are well served by system stacks and workhorse UI faces; Persuade and Experience surfaces want faces with a point of view, and these training-data defaults mean you stopped looking: Fraunces, Playfair Display, Cormorant, Lora, Crimson, Newsreader, Syne, Space Grotesk, Space Mono, IBM Plex, Inter-as-display, DM Sans, DM Serif, Outfit, Plus Jakarta Sans, Instrument Sans. Naming one of these faces anyway requires a reason no other face could satisfy, and a subject association is never that reason: books wanting a serif, bookshops wanting hand-lettering, and tech wanting a mono are the associations the list exists to break.
 
-Calibration: AI-generated interfaces cluster around a few looks regardless of subject: warm cream ground, high-contrast serif display, and a terracotta or signal-red accent; near-black with one neon accent and glowing edges; broadsheet-editorial hairlines, italic display serif, and small tracked mono labels. All are legitimate when the brief calls for them. Where the brief leaves the aesthetic free, landing in one means the self-check failed: if someone could guess your aesthetic from the category alone, or from category-plus-avoidance, rework until neither answer is obvious. Energy is not the enemy of trust: a brief's negative constraints (no gamification, no hype) rule out those devices, not exuberance, and adjectives describing the product's behavior (quiet support, calm coaching) do not dictate the surface's energy. A bookish, warm, or child-facing subject does not soften the calibration: book cloth, thread, jackets, endpapers, and shelf ephemera span the whole saturated spectrum, and cream paper is the smallest corner of that world; landing on cream plus serif for a book subject is the default wearing the subject's clothes. A brief-pinned world pins the world, not its softest rendition: the pinned world's full material range stays in play, and a rendition that matches what any model ships for that world failed the self-check at execution rather than selection.
+Calibration: AI-generated interfaces cluster around a few looks regardless of subject: warm cream ground, high-contrast serif display, and a terracotta or signal-red accent; near-black with one neon accent and glowing edges; broadsheet-editorial hairlines, italic display serif, and small tracked mono labels. All are legitimate when the brief calls for them. Where the brief leaves the aesthetic free, landing in one means the self-check failed: if someone could guess your aesthetic from the category alone, or from category-plus-avoidance, rework until neither answer is obvious. Energy is not the enemy of trust: a brief's negative constraints (no gamification, no hype) rule out those devices, not exuberance, and adjectives describing the product's behavior (quiet support, calm coaching) do not dictate the surface's energy. A bookish, warm, or child-facing subject does not soften the calibration: book cloth, thread, jackets, endpapers, and shelf ephemera span the whole saturated spectrum, and cream paper is the smallest corner of that world; landing on cream plus serif for a book subject is the default wearing the subject's clothes. A brief-pinned world pins the world, not its softest rendition: the pinned world's full material range stays in play, and a rendition matching what any model ships for that world failed the self-check at execution rather than selection.
 
 ## 5. Record the decision
 
-Before code, state the chosen direction as a contract in the artifact's opening comment, five short blocks, 150 words at most, in a form that survives the production build: an HTML comment in the emitted markup, never only a templating-frontmatter comment, placed as the first child of the document's body in the root layout, never inside a slotted or child component (some compilers, Astro among them, strip a slot's leading comment while keeping deeper ones). After the first production build, grep the built output for the seed key; a contract the build erased is a contract nobody can audit. THESIS: the one idea this surface owns and the category-default arrangement it refuses. OWN-WORLD: the palette and component language, specific enough to be recognizable with all content removed. STORY: what the visitor understands, believes, and does. FIRST VIEWPORT: the exact composition, what is where and at what scale, and where the primary action sits. FORM: the chosen form, its position on your ordered list, and the seed key the script printed. Close the comment with one more line, FINISH: the run's exit condition, verbatim "unreviewed and undocumented is unfinished; this build ends with the finish review, the verdict, and DESIGN.md". The comment tops the artifact you re-open on every edit, the one reminder that survives a long build: a page that looks complete with the FINISH line undischarged is not done, it is abandoned at the finish line. If a block reads like a mood, the direction is not decided yet; the finishing review audits the render against this contract.
+Before code, state the chosen direction as a contract in the artifact's opening comment, five short blocks, 150 words at most, in a form that survives the production build: an HTML comment in the emitted markup, never only a templating-frontmatter comment, placed as the first child of the document's body in the root layout, never inside a slotted or child component (some compilers, Astro among them, strip a slot's leading comment while keeping deeper ones). After the first production build, grep the built output for the seed key; a contract the build erased is a contract nobody can audit. THESIS: the one idea this surface owns and the category-default arrangement it refuses. OWN-WORLD: the palette and component language, specific enough to be recognizable with all content removed. STORY: what the visitor understands, believes, and does. FIRST VIEWPORT: the exact composition, what is where and at what scale, and where the primary action sits. FORM: the chosen form, its position on your ordered list, and the seed key the script printed. Close with one more line, FINISH: the run's exit condition, verbatim "unreviewed and undocumented is unfinished; this build ends with the finish review, the verdict, DESIGN.md, and every shipping raster carrying its provenance". The comment tops the artifact you re-open on every edit, the one reminder that survives a long build: a page that looks complete with the FINISH line undischarged is not done, it is abandoned at the finish line. If a block reads like a mood, the direction is not decided yet; the finishing review audits the render against this contract.
 
 On a new or replacement world, DESIGN.md is written at finish, from the built world, by the shipped documenter (section 7); a rulebook written before the build gets defended against reality instead of describing it, and hands the design-system detector an unstable target. A new world shipped with no DESIGN.md is still an incomplete run. An ordinary extension does not rewrite DESIGN.md.
 
@@ -80,18 +80,18 @@ If the work establishes durable strategy for a route or artifact, read its exist
 
 Keep the brief small: scope and visitor mode; audience, job, action/task, proof/content, and constraints; chosen direction and memorable moment; unresolved decisions. Do not copy global product truth or DESIGN.md tokens into it.
 
-On a comp-led build, whenever any image generation is available, a harness-native tool or the API fallback context.mjs reports, the locked direction is visualized before it is built, never skipped: load [visualize.md](visualize.md) and follow it, three compositional options put before the user for approval, the chosen card's decision comp plus two variations. This step is proven to produce the most compositional and ambitious work. On a code-led build the comp round is skipped by contract, never by drift: the ambition it would have carried lives in the direction contract's FIRST VIEWPORT block and named signature interaction, and the finish reviewer audits those promises in behavior.
+On a comp-led build, whenever any image generation is available (a harness-native tool or the API fallback context.mjs reports), the locked direction is visualized before it is built, never skipped: load [visualize.md](visualize.md) and follow it, three compositional options put before the user for approval, the chosen card's decision comp plus two variations. This step is proven to produce the most compositional and ambitious work. On a code-led build the comp round is skipped by contract, never by drift: the ambition it would have carried lives in the direction contract's FIRST VIEWPORT block and named signature interaction, and the finish reviewer audits those promises in behavior.
 
 For `shape`, return the selected direction to [shape.md](shape.md) and stop before persistence or implementation.
 
 ## 6. Build with full commitment
 
-When an approved comp exists, the comp is king, and the build happens in phases. Phase one is reproduction: rebuild the comp at its own breakpoint until a screenshot at the comp's width and height overlaps it near pixel-perfectly, materials, components, elevation, assets, and implied design language included. Exactly three concessions exist: fonts (the closest obtainable face), icons (exact match unless the user already chose an icon library), and genuine defects in the generated comp such as spelling errors. Everything else must match, and models systematically believe their HTML, CSS, and SVG recreation succeeded when it did not, so the overlap comparison is the authority, never your conviction: set the screenshot beside the freshly reopened comp image at identical dimensions after every region, never beside your memory of it, and when a region keeps losing that comparison, stop recreating it in code and produce it as a rendered asset composited into the page. The comp also outranks every written record of it: when the recorded brief or inventory commits to less than the comp shows, a softer texture, a sparser field, a sculpted plate reduced to flat CSS, correct the record upward to the comp; qualifiers like subtle, restrained, and low-contrast, and counts rounded down to a comfortable fraction, are how approved materials die between approval and build. A produced material must then survive to the screen: a texture buried under a nearly opaque color wash ships the wash, not the material, so judge every material by the screenshot beside the comp, never by the stylesheet. Only when reproduction holds does phase two begin: static regions that should live become animated or interactive, reveals and motion are added, then responsiveness across the surface's devices. Where the comp does not cover the whole surface, continue building the remainder inside the comp's recorded world and design language; a component the comp never shows inherits the recorded system's corner language, line weights, and materials, and may not introduce container styles, border weights, or chrome the comp never uses.
+When an approved comp exists, the comp is king, and the build happens in phases. The comp is a spatial contract, not a mood board: only the user can downgrade its authority, in explicit words, and difficulty never infers a downgrade. Phase one is reproduction: rebuild the comp at its own breakpoint until a screenshot at the comp's width and height overlaps it near pixel-perfectly, materials, components, elevation, assets, and implied design language included. Exactly three concessions exist: fonts (the closest obtainable face), icons (exact match unless the user already chose an icon library), and genuine defects in the generated comp such as spelling errors. Everything else must match, and models systematically believe their HTML, CSS, and SVG recreation succeeded when it did not, so the overlap comparison is the authority, never your conviction: set the screenshot beside the freshly reopened comp image at identical dimensions after every region, never beside your memory of it, and when a region keeps losing that comparison, stop recreating it in code and produce it as a rendered asset composited into the page. The comp also outranks every written record of it: when the recorded brief or inventory commits to less than the comp shows, a softer texture, a sparser field, a sculpted plate reduced to flat CSS, correct the record upward to the comp; qualifiers like subtle, restrained, and low-contrast, and counts rounded down to a comfortable fraction, are how approved materials die between approval and build. A produced material must then survive to the screen: a texture buried under a nearly opaque color wash ships the wash, not the material, so judge every material by the screenshot beside the comp, never by the stylesheet. Every color the brief records gets that comparison by number, not by eye: sample the build screenshot's ground, dominant fields, and accents the same way each record was taken (an interior patch average where the record is an average, both end colors where the record is a gradient) and set each value against its recorded counterpart (sampled from the comp itself when the brief lacks one), and when a texture or tile paints over a base token, measure the net on-screen value, because the eye files a drifted color under the same color word and the number is what catches it. Judge the gap like a colorist, not a diff tool: a difference with a color name (warmer, grayer, darker than the record) is drift to fix, while a few digits of render and compression noise are the same color. Only when reproduction holds does phase two begin: static regions that should live become animated or interactive, reveals and motion are added, then responsiveness across the surface's devices. Where the comp does not cover the whole surface, continue building the remainder inside the comp's recorded world and design language; a component the comp never shows inherits the recorded system's corner language, line weights, and materials, and may not introduce container styles, border weights, or chrome the comp never uses.
 
 Build the assigned direction, not a safer interpretation of it. The form supplies structure, reading order, component conventions, and native motion; the product supplies every fact. Commit every atom: nav, buttons, inputs, and links are rebuilt in the form's vocabulary, and a stock component inside a committed form is a lapse. Land the first build fully committed; committing is the hard part, and the passes that follow exist to make the committed thing clear and effective, never to dilute it. In unattended work, the safe rendition is the known risk.
 
 - **The first viewport is a thesis, not a header.** Demonstrate the mechanism immediately, at the scale the form has in life; do not trap the concept inside a standard hero or card shell. The memory test: if someone left after one viewport, what would they describe an hour later? If the honest answer is a mood, the concept has not committed yet.
-- **Prove the hero before building past it.** When an approved comp exists, render the first viewport, capture it, and set it beside the comp's first viewport before any later section: the hero carries the run's ambition, and every following section inherits its shortfall. Judge scale and density as quantities, a field at a tenth of the comp's coverage or type at half its weight is a different design, and a five-minute retry here is what a rebuild verdict at the finish costs when this check is skipped.
+- **Prove the hero before building past it.** When an approved comp exists, render the first viewport, capture it at the comp's own pixel dimensions, and set it beside the comp's first viewport before any later section: the hero carries the run's ambition, and every following section inherits its shortfall. Save that capture as `.impeccable/review/hero-repro.png` (create the directory); the finish reviewer verifies it exists, so a skipped checkpoint is a visible checkpoint. Judge scale and density as quantities, a field at a tenth of the comp's coverage or type at half its weight is a different design, and a five-minute retry here is what a rebuild verdict at the finish costs when this check is skipped.
 - **Prove, don't claim.** Show the subject doing its job: the interface at work, the mechanism dramatized, specifics a competitor could not copy-paste. Sections that restate a claim in different words add length, not substance. Demonstration data is design material: author it at full fidelity and label it synthetic; claims stay uninventable.
 - **Author the assets; never substitute chrome.** Great surfaces live on carefully made content: names, entries, copy, covers, thumbnails, textures. In greenfield work every blank the ask round left open is yours to author at production fidelity; content is authorable, claims are labelable, no section is omittable. An unanswered commercial claim ships as a clearly marked placeholder on the user's replacement list. When image generation exists, producing the design's imagery is part of building, at the scale the composition needs: a viewport that wants atmosphere gets a full-bleed layered scene, and a library of small centered subjects standardized for tidiness forecloses it. Gradients, glass, and generic icon tiles where an authored asset belongs are the gap wearing chrome; icons drawn in the world's own grammar are the remedy, not the target.
 - **Build the form's web leverage.** When the chosen world names a technique (canvas, WebGL, view transitions, generative motion), build the technique itself, not a static imitation of it; the graceful fallback serves constrained clients, it is not the default experience.
@@ -103,8 +103,18 @@ Preserve semantics, accessibility, performance, responsiveness, project conventi
 
 ## 7. Inspect and finish
 
-Inspect the surface's target sizes in one batched screenshot round: desktop and mobile on the web; on a native platform (`ios` / `android` / `adaptive`), the shipped device classes per OS, captured from the simulator or emulator the way the platform reference's Verifying the build section describes. Critique the render against the user's request and the direction contract, fix material gaps, and confirm with one final round; two rounds is the ceiling, and fixes batch between them rather than earning per-tweak screenshots. When an approved comp exists, the critique is a side-by-side: view the comp region and the build region together, the hero and each section as its own crop at legible scale, never one full-page thumbnail, which hides exactly the failures that matter, crude controls, wrong lettering character, flattened material, behind a superficially similar section order. On a Persuade surface, verify the mode did its job: a first-time visitor should know what this is, why it matters, and what to do within seconds, in the form's own vocabulary.
+Inspect the surface's target sizes in one batched screenshot round: desktop and mobile on the web; on a native platform (`ios` / `android` / `adaptive`), the shipped device classes per OS, captured from the simulator or emulator the way the platform reference's Verifying the build section describes. When the harness reports the user's actual viewport (an in-app browser's size, a named resolution), add that width to the set: the width that breaks is the one the user sees first. Critique the render against the user's request and the direction contract, fix material gaps, and confirm with one final round; two rounds is the ceiling, and fixes batch between them rather than earning per-tweak screenshots. When an approved comp exists, the critique is a side-by-side: view the comp region and the build region together, the hero and each section as its own crop at legible scale, never one full-page thumbnail, which hides exactly the failures that matter, crude controls, wrong lettering character, flattened material, behind a superficially similar section order. On a Persuade surface, verify the mode did its job: a first-time visitor should know what this is, why it matters, and what to do within seconds, in the form's own vocabulary.
 
-After the second inspection round the build thread's polishing is over: no further defect hunts, micro-edit scripts, or rebuilds here; whatever remains ships through the handoffs, where a fresh context does the finding better and cheaper. On the web, where this harness runs no design hook, run `node .agent/skills/impeccable/scripts/detect.mjs --json` on the changed targets once here, fix what is mechanical, and pass the remaining findings to the reviewer; a hookless web build that skips this ships every tell the hook exists to catch. A native platform skips the detector entirely: it reads HTML and CSS and has no verdict on native code, so the reviewer's floor check is the only slop gate and the input packet says so. Capture the screenshots into `.impeccable/review/`, one file per captured viewport (on the web, `desktop.png` and `mobile.png`; on native, one per device class, such as `phone.png` and `tablet.png`, suffixed per OS on adaptive), creating that directory when the harness does not; the paths you pass the reviewer are its spec, and that directory is where it looks when a passed path is missing. Then spawn the shipped finish reviewer, `impeccable-finish-reviewer` (`impeccable_finish_reviewer` in codex; `/impeccable-finish-reviewer` in Cursor; on GitHub Copilot say "Use the impeccable-finish-reviewer agent"), with the original request, confirmed answers, the artifact path, the screenshot paths, its direction contract, existing hook findings, the QUALITY BAR card and approved comp paths (on a code-led build there is no approved comp; the chosen decision comp rides in that slot as the critique reference, named as such), the craft-floor reference path, and on a native platform the platform reference path(s), [ios.md](ios.md) / [android.md](android.md), both on adaptive, plus one line saying no detector ran, so the reviewer judges in the platform's conventions rather than the web's. The reviewer has no browser; screenshots you fail to pass are checks it cannot run. Never read the shipped agents' definition files before spawning; the harness loads them at spawn, and you owe only the input packet. Wait on any agent with one long timeout rather than a loop of short polls, and spend the wait on the next independent step. Verify its return carries the five contract sections; on an empty or thrashed return, respawn once with the same inputs before doing anything else. This review never runs inside the build thread and never inherits it: spawn the reviewer fresh, with no forked conversation history (`fork_turns: 0` in codex); a reviewer that inherits your transcript inherits your framing, your optimism, and your abstractions, and everything it needs travels in the inputs above. Only a harness whose tool surface has no subagent capability at all substitutes a fresh in-thread pass after stepping fully out of the build context, run from [degraded/finish-reviewer.md](degraded/finish-reviewer.md), and a substituted or failed-and-replaced review is disclosed in one line at finish, never silently. When the reviewer's first material fix is a rebuild directive, fidelity failed wholesale rather than in patches, so skip the fix batch and execute the rebuild immediately: re-derive the named regions, produce the named assets, and send the result back for a verdict, telling the user what is happening rather than asking permission to fix a failure. The user is consulted only when a second rebuild directive arrives, both verdicts on the table, or when rebuilding would discard content the user approved. Otherwise apply the material fixes in one batch, rebuild once, and recapture the same viewports over the same files. A recapture measures positions, loading, and overflow; it cannot measure whether a fix reached the quality the finding named, so send the recaptured screenshots back to the same reviewer for a verdict scoring every material fix resolved, partial, or unresolved (through the harness's agent continuation; without one, run the scoring fresh from [degraded/finish-reviewer.md](degraded/finish-reviewer.md)'s Verdict Pass). Fixes scored partial or unresolved get another batch, recapture, and verdict. Two rounds is the budget an unattended run ends at; an attended session's ceiling belongs to the user, so when the second verdict still lists open items, put the table in front of them and let them choose between shipping as it stands and funding another round. Whoever is deciding, stop the moment a round resolves nothing, and the reviewer's findings are the only list you work from, never your own re-opened hunt. Report the final verdict table to the user as it stands, open items included, under the reviewer's own disposition word: a table with open material findings is never announced as a pass, and never under a softer label than the reviewer wrote. Do not run a second detector.
+A capture is evidence only when it is valid, and you validate before you send. Settle or disable entrance motion first: an element hidden by animation timing reads as a missing element and gets fixed into a regression. Capture full-page shots from the document top. Capture the comp comparison at the comp's own pixel dimensions. Then open every file once and confirm it shows what its name claims: no black or blank regions, no wrong section behind a right filename, no half-loaded state. A malformed capture sent onward costs the whole round; the reviewer answers it with `disposition: recapture` and nothing it reviewed binds.
 
-Then spawn the shipped documenter, `impeccable-documenter` (`impeccable_documenter` in codex), with the project root, the artifact path, the direction contract, PRODUCT.md, the [document.md](document.md) reference path, and the boundary to write at; it records DESIGN.md and the sidecar from the built world, ground truth over intention; without subagents the pass runs from [degraded/documenter.md](degraded/documenter.md). A clean detector pass is not finished; finished is the contract kept, the comp honored, the review closed, and the system recorded.
+After the second inspection round the build thread's polishing is over: no further defect hunts, micro-edit scripts, or rebuilds here; whatever remains ships through the handoffs, where a fresh context does the finding better and cheaper. On the web, where this harness runs no design hook, run `node .agent/skills/impeccable/scripts/detect.mjs --json` on the changed targets once here, fix what is mechanical, and pass the remaining findings to the reviewer; a hookless web build that skips this ships every tell the hook exists to catch. A native platform skips the detector entirely: it reads HTML and CSS and has no verdict on native code, so the reviewer's floor check is the only slop gate and the input packet says so. Capture the screenshots into `.impeccable/review/`, one file per captured viewport (on the web, `desktop.png` and `mobile.png`, plus `user-<width>.png` whenever the user's viewport joined the inspected set; on native, one per device class, such as `phone.png` and `tablet.png`, suffixed per OS on adaptive), creating that directory when the harness does not; the paths you pass the reviewer are its spec, every viewport you inspected is named required in the packet, and that directory is where it looks when a passed path is missing.
+
+Then spawn the shipped finish reviewer, `impeccable-finish-reviewer` (`impeccable_finish_reviewer` in codex; `/impeccable-finish-reviewer` in Cursor; on GitHub Copilot say "Use the impeccable-finish-reviewer agent"), with the original request, confirmed answers, the artifact path, the screenshot paths, the direction contract, existing hook findings, the QUALITY BAR card and approved comp paths (a code-led build has no approved comp; the chosen decision comp rides in that slot as the critique reference, named as such), the craft-floor reference path, and on a native platform the platform reference path(s), [ios.md](ios.md) / [android.md](android.md), both on adaptive, plus one line saying no detector ran, so the reviewer judges in the platform's conventions rather than the web's. The reviewer has no browser; screenshots you fail to pass are checks it cannot run. Never read the shipped agents' definition files before spawning; the harness loads them at spawn, and you owe only the input packet. Wait on any agent with one long timeout rather than a loop of short polls, and spend the wait on the next independent step. Verify the return carries the five contract sections (a recapture return carries one, its recapture list); on an empty or thrashed return, respawn once with the same inputs. This review never runs inside the build thread and never inherits it: spawn the reviewer fresh, with no forked conversation history (`fork_turns: 0` in codex); a reviewer that inherits your transcript inherits your framing, your optimism, and your abstractions, and everything it needs travels in the inputs above. Only a harness with no subagent capability at all substitutes a fresh in-thread pass after stepping fully out of the build context, run from [degraded/finish-reviewer.md](degraded/finish-reviewer.md), and a substituted or failed-and-replaced review is disclosed in one line at finish, never silently.
+
+Act on the disposition word; there are exactly four. **recapture**: the evidence failed, not the build. Recapture what the return names under the capture-validity rules, then run a full review over the new evidence. A review conducted on invalid evidence binds nothing, and a verdict pass may never follow it. **rebuild**: fidelity failed wholesale, not in patches. Skip the fix batch and execute the rebuild immediately: re-derive the named regions, produce the named assets, and send the result back for a fresh full review, never a verdict pass; a rebuild replaces regions wholesale, so the whole matrix runs again over the recaptures. Tell the user what is happening rather than asking permission to fix a failure. Consult the user only on a second rebuild directive, both verdicts on the table, or when rebuilding would discard content the user approved. **ship**: nothing is owed; report the verdict at its scope and continue to the documenter. **fix**: apply the material fixes in one batch, rebuild once, and recapture the same viewports over the same files. A recapture measures positions, loading, and overflow; it cannot measure whether a fix reached the quality the finding named, so send the recaptured screenshots back to the same reviewer for a verdict scoring every material fix resolved, partial, or unresolved (through the harness's agent continuation; without one, run the scoring fresh from [degraded/finish-reviewer.md](degraded/finish-reviewer.md)'s Verdict Pass). Fixes scored partial or unresolved get another batch, recapture, and verdict. Two rounds is the budget an unattended run ends at; an attended session's ceiling belongs to the user, so when the second verdict still lists open items, put the table in front of them and let them choose between shipping as it stands and funding another round. Whoever decides, stop the moment a round resolves nothing, and the reviewer's findings are the only list you work from, never your own re-opened hunt. Do not run a second detector.
+
+A rebuild and a fix round share one asset rule: a raster either round creates or replaces is still asset work under [visualize.md](visualize.md)'s Produce section and keeps its **provenance** like every build raster, and a raster the round abandons is deleted in the same batch. Before either round's result goes back for review or verdict, run `node .agent/skills/impeccable/scripts/embed-prompt.mjs --scan <asset-dir...>` over the directories the artifact's rasters ship from and clear every file it reports by embedding what it is missing: the exact generation prompt for a produced raster, the origin for a sourced, stock, or pre-existing one. The scan only reads; deletion is reserved for rasters the round abandoned, never for a file the scan flagged.
+
+Report the final verdict under the reviewer's own disposition word and at its actual scope. A verdict pass scores the listed fixes and nothing else: "the reviewer scored all three fixes resolved" is a claim it supports, "no material issues remain" is not. A table with open material findings is never announced as a pass, never softened, and never dressed as whole-surface approval when only a fix list was scored. When the user answers a ship with evidence against it, their own screenshot, a named mismatch with the comp, that evidence outranks every capture you made: put their material in the packet and spawn a fresh reviewer for a new full review. Patching inline and self-certifying is how a rejected page ships twice.
+
+Then spawn the shipped documenter, `impeccable-documenter` (`impeccable_documenter` in codex), with the project root, the artifact path, the direction contract, PRODUCT.md, the [document.md](document.md) reference path, and the boundary to write at; it records DESIGN.md and the sidecar from the built world, ground truth over intention; without subagents the pass runs from [degraded/documenter.md](degraded/documenter.md). The documenter runs after the last correction lands: when any fix round follows the documentation, re-run the documenter over the changed surface, because a DESIGN.md describing a layout that no longer exists turns defects into system guidance. A clean detector pass is not finished; finished is the contract kept, the comp honored, the review closed, and the system recorded.

.agent/skills/impeccable/reference/visualize.md

@@ -1,51 +1,55 @@
 # Visualize: Direction Comps & Asset Production
 
-Load this from [new-work.md](new-work.md) on a comp-led build, when image generation is available (a harness-native tool or the API fallback context.mjs reports). A code-led execution contract skips this file by design, not by drift: its ambition lives in the written direction contract and is audited in behavior, so do not load it for a code-led round. PRODUCT.md and DESIGN.md are preconditions. New-work has already resolved the visual world; this file must not reopen it. A surface-scope structure round that already put three visualized cards before the user (new-work.md, established world) has discharged this round: the locked card’s comp is the approved comp, so record the approval and continue at After approval; generate nothing new.
+Load this from [new-work.md](new-work.md) on a comp-led build, when image generation is available (a harness-native tool or the API fallback context.mjs reports). A code-led contract skips this file by design, not by drift; do not load it then. PRODUCT.md and DESIGN.md are preconditions. New-work has already resolved the visual world; this file must not reopen it. A surface-scope structure round that already put three visualized cards before the user (new-work.md, established world) has discharged this round: the locked card's comp is the approved comp, so record the approval and continue at After approval; generate nothing new.
 
-The purpose of a probe is to test composition, narrative, hierarchy, density, focal moment, signature use, and image requirements. It is not a second identity workshop. Keep DESIGN.md's palette, typography direction, material language, component character, imagery stance, and motion grammar fixed.
+A probe tests composition, narrative, hierarchy, density, focal moment, signature use, and image requirements. It is not a second identity workshop. Keep DESIGN.md's palette, typography direction, material language, component character, imagery stance, and motion grammar fixed.
 
 ## Generate three compositional options
 
-Render three distinct high-fidelity north-star comps of the requested surface, with whatever generation capability exists, saved under `.impeccable/mocks/` so they survive the session. Comp at the surface's own viewport: portrait at device size for a native app or mobile-first surface, desktop landscape otherwise; a phone screen comped landscape misstates the composition before anything gets built against it. Comps are the build thread's own work, never delegated: the thread that writes the comp prompts holds the direction's full context, and it has already seen every comp when the build starts. Open every image you produce or reference by its workspace-relative path, never an absolute one: sandboxed viewers reject absolute paths, and everything under the project root has a relative path. Base them on the real content and the surface concepts already developed with the user. On an established world, anchor every comp on the real identity: capture a screenshot of a representative existing page and pass it as a reference image (the harness image tool’s input image, or `generate-image.mjs --ref`); the prompt then leads with the new surface’s structure while the reference carries palette, type, and component character, because DESIGN.md words alone drift where a pixel reference does not. Name what the reference contributes and what it must not: chrome, palette, type, and component character carry over; the reference page’s own content does not, so a banner, hero, or card lifted verbatim from the reference is the reference leaking, not fidelity. Three is the number: one comp invites rubber-stamping, and the spread between three is what surfaces the composition worth building. The chosen card's decision comp is the first of the three: it already renders this direction at full fidelity under this file's discipline, so this round generates two more that vary what the first held fixed, and all three go to the approval point together. Only a round that arrives with no decision comp, a degraded roll, an identity-mode page, a direction pinned without the decision round, renders all three here.
+Render three distinct high-fidelity north-star comps of the requested surface, saved under `.impeccable/mocks/` so they survive the session. Comp at the surface's own viewport: portrait at device size for a native app or mobile-first surface, desktop landscape otherwise; a phone screen comped landscape misstates the composition before anything is built against it. Comps are the build thread's own work, never delegated: the thread that writes the prompts holds the direction's full context and has seen every comp when the build starts. Open every image by its workspace-relative path; sandboxed viewers reject absolute paths, and everything under the project root has a relative one. Base the comps on real content and the surface concepts already developed with the user. On an established world, anchor every comp on the real identity: capture a screenshot of a representative existing page and pass it as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`); the prompt leads with the new surface's structure while the reference carries palette, type, and component character, because DESIGN.md words alone drift where a pixel reference does not. Name what the reference contributes and what it must not: chrome, palette, type, and component character carry over; the reference page's own content does not, and a banner, hero, or card lifted verbatim is the reference leaking, not fidelity. Three is the number: one comp invites rubber-stamping; the spread between three surfaces the composition worth building. The chosen card's decision comp is the first of the three: it already renders this direction at full fidelity under this discipline, so generate two more that vary what the first held fixed, and send all three to the approval point together. Only a round arriving with no decision comp (a degraded roll, an identity-mode page, a direction pinned without the decision round) renders all three here.
 
-- A comp is a designed surface, not a picture of the subject. Lead the generation prompt with the surface's own structure, whatever regions this design actually has, named in order with their scale relationships; a page with no navigation states that instead of inventing one, and an unconventional surface states its unconventional skeleton. A prompt that leads with the world's atmosphere gets a vignette back: the model paints the fish market instead of the fish market's website. Self-check every render: if it could hang as a poster, or reads as a photograph or scene with some text on it, it is not a comp; regenerate with the layout scaffold stated more literally.
-- The inverse is also a failure: a surface with none of its subject in it. The subject appears as the content the regions exist to hold; the world dresses the frame and never displaces what the frame exists to show. The deletion usually rides in on the prompt's exclusion list, so exclusions bind invented claims, and a medium ban belongs to the committed imagery stance, never to caution. Before accepting a render, point at the subject: a render that depicts everything about the world and nothing of the subject fails however faithful its atmosphere, so regenerate with the subject's content named region by region.
-- A comp is judged as the shipped screen: the visitor's job must be readable from the image alone. Name the surface's mode from the render with no caption; a render whose mode cannot be read back is art direction without a surface, so regenerate with the visitor's job as the prompt's spine.
-- Commitment is depth, not coverage. The world enters through one dominant move plus the material, type, and spacing that support it, and the remaining regions hold still so that move can be read; a region that simply does its job in the world's own grammar carries the direction further than a region performing the concept. The check cuts competition, never content: a quieted region keeps its information and stops performing. Where the direction names a focal moment, a second element competing with it at the same scale means the comp is shouting; where it names none, several regions performing the concept at once is the same shout. Regenerate keeping the strongest move and quieting the rest. Busy is louder, not bolder.
+- A comp is a designed surface, not a picture of the subject. Lead the prompt with the surface's own structure: the regions this design has, named in order with their scale relationships; a page with no navigation says so instead of inventing one, and an unconventional surface states its unconventional skeleton. A prompt that leads with atmosphere gets a vignette back: the model paints the fish market instead of the fish market's website. Self-check every render: if it could hang as a poster, or reads as a photograph with some text on it, it is not a comp; regenerate with the layout scaffold stated more literally.
+- The inverse is also a failure: a surface with none of its subject in it. The subject appears as the content the regions hold; the world dresses the frame and never displaces what the frame shows. The deletion usually rides in on the prompt's exclusion list, so exclusions bind invented claims, and a medium ban belongs to the committed imagery stance, never to caution. Before accepting a render, point at the subject; a render that depicts everything about the world and nothing of the subject fails however faithful its atmosphere. Regenerate with the subject's content named region by region.
+- Judge a comp as the shipped screen: the visitor's job must be readable from the image alone. Name the surface's mode from the render with no caption; a render whose mode cannot be read back is art direction without a surface. Regenerate with the visitor's job as the prompt's spine.
+- Commitment is depth, not coverage. The world enters through one dominant move plus the material, type, and spacing that support it; the remaining regions hold still so that move can be read. A region that simply does its job in the world's grammar carries the direction further than a region performing the concept. The check cuts competition, never content: a quieted region keeps its information and stops performing. A second element competing with the named focal moment at the same scale means the comp is shouting; with no named focal moment, several regions performing the concept at once is the same shout. Regenerate keeping the strongest move and quieting the rest. Busy is louder, not bolder.
 - When the user shortlisted multiple concepts, spread the three across them.
 - When one direction is committed, vary the structural uncertainty an image can resolve: topology, sequence, density, hierarchy, focal composition, or interaction framing.
-- Show enough beyond the opening moment to prove the concept can govern the whole requested surface.
+- Show enough beyond the opening moment to prove the concept can govern the whole surface.
 - Do not generate a palette artifact, ask new atmosphere questions, introduce a different type voice, or invent a new motif. If the committed world cannot support the concept, return to the concept shortlist rather than changing the world.
 
-Treat each comp as a direction test, not a screenshot specification. Core UI text, responsive behavior, accessibility, semantics, and interaction states remain implementation responsibilities.
+Each comp is a direction test, not a screenshot specification. Core UI text, responsive behavior, accessibility, semantics, and interaction states remain implementation responsibilities.
 
 ## One approval point
 
-Show the three together on the decision page (`serve-question.mjs`, one option per comp with the comp as its hero), or in the harness only when it renders images inline; a text-only surface does not count as display. Ask what should carry forward, what feels false to the world, and whether the selected surface concept should be approved, combined, revised, or rejected. Then stop and wait. A structured simulated user counts as attended and receives the same question.
+Show the three together on the decision page (`serve-question.mjs`, one option per comp with the comp as its hero), or in the harness only when it renders images inline; a text-only surface does not count as display. Ask what should carry forward, what feels false to the world, and whether the selected concept should be approved, combined, revised, or rejected. Then stop and wait. A structured simulated user counts as attended and receives the same question.
 
 Do not begin code until the user approves a direction or explicitly delegates the choice. If they delegate, choose using the task brief, PRODUCT.md, and DESIGN.md, and state the evidence. Approval refines the task concept; it does not modify DESIGN.md.
 
-This approval point has no substitute and no skip condition. When the structured question tool errors, fall back to the decision page; only after both fail may you treat the choice as delegated, and a delegated pick is still recorded exactly as an approval is and disclosed in your first reply, not your last. The finish reviewer treats a build whose comp round produced comps with no recorded approval as carrying a material finding; decision comps under `.impeccable/mocks/decision/` are the direction round's hand, not comp-round output, and imply no approval on their own.
+This approval point has no substitute and no skip condition. When the structured question tool errors, fall back to the decision page; only after both fail may you treat the choice as delegated, and a delegated pick is recorded exactly as an approval is and disclosed in your first reply, not your last. The finish reviewer treats comp-round comps with no recorded approval as a material finding; decision comps under `.impeccable/mocks/decision/` are the direction round's hand, not comp-round output, and imply no approval on their own.
 
-After approval, record the choice where tools can find it: the approved comp's path goes in the surface brief, and the approved comp's `.json` prompt sidecar gains `"approved": true` (every comp generated through `generate-image.mjs` has one; create it if a native tool didn't). The sidecar travels with the mocks folder, so the approval survives sessions and machines that never see the brief. Then summarize the composition and the parts of the comp that must not be literalized, return to new-work.md, record the direction contract from the approved surface concept, and build.
+After approval, record the choice where tools can find it: the approved comp's path goes in the surface brief, and its `.json` prompt sidecar gains `"approved": true` (every comp generated through `generate-image.mjs` has one; create it if a native tool didn't). The sidecar travels with the mocks folder, so the approval survives sessions and machines that never see the brief. Summarize the composition and the parts of the comp that must not be literalized, return to new-work.md, record the direction contract from the approved concept, and build.
 
 ## Inventory implementation fidelity
 
-Before building, read the approved comp as a design system and record it in the brief: component grammar, corner language, line weights, elevation treatment, and the type ramp, because everything the comp does not show gets built from this record, and without it the fallback is the model's stock kit of square boxes, 1px grids, bento cells, and hard shadows. Then inventory the comp's major visible ingredients in writing (a short table in the surface brief or working notes; the finish reviewer audits shipped assets against it) and choose an implementation medium for each: semantic HTML/CSS/SVG, existing project asset, generated raster, sourced raster, icon library, canvas/WebGL, or accepted omission. The same written inventory names the comp's compositional commitments: navigation items and icons, headline levels and their scale relationship, signature geometry such as seams, masks, and overlaps, and each section's arrangement and density. The primary action gets its own row with its own medium: when the comp dissolves, stamps, erodes, or otherwise physically works the main CTA, that treatment is signature material on the page's most important element, and shrinking it to a border trick or a few decorative pixels is the compliance-token version of commitment. An element never written down is the element the build silently drops, and the direction contract's 150 words cannot carry this list, so this inventory is where it lives.
+Before building, read the approved comp as a design system and record it in the brief: component grammar, corner language, line weights, elevation treatment, and the type ramp. Everything the comp does not show gets built from this record; without it the fallback is the model's stock kit of square boxes, 1px grids, bento cells, and hard shadows. Then inventory the comp's major visible ingredients in writing (a short table in the surface brief or working notes; the finish reviewer audits shipped assets against it) and choose an implementation medium for each: semantic HTML/CSS/SVG, existing project asset, generated raster, sourced raster, icon library, canvas/WebGL, or accepted omission. The same inventory names the comp's compositional commitments: navigation items and icons, headline levels and their scale relationship, signature geometry such as seams, masks, and overlaps, and each section's arrangement and density. The primary action gets its own row with its own medium: when the comp dissolves, stamps, erodes, or otherwise physically works the main CTA, that treatment is signature material on the page's most important element, and shrinking it to a border trick is the compliance-token version of commitment. An element never written down is the element the build silently drops; the direction contract's 150 words cannot carry this list, so it lives here.
 
-The medium column is where an approved design most often dies, so it obeys a gate: the medium is decided by what the comp region shows, never by what feels buildable in the current stack. A human figure, a product object, machinery, or any material with lighting and depth is raster whatever the stack, and so is any texture by that name alone: woven cloth, paper grain, fabric, leather, brushed metal need no depth argument, because a CSS gradient or layered background is not a texture medium and "layered CSS textures" is not a medium at all. Writing "silhouette" for a photographic figure, or "CSS" for a sculpted panel's finish or a cotton field's weave, is not a medium choice, it is the quiet deletion of the approved design, and it is how a comp full of physical material becomes a flat page with the same section order. Style does not move this boundary: a comp region with perspective, shading, figure drawing, or dense mechanical detail is illustration however line-drawn it looks, and no build session can author illustration as vectors, so it regenerates as raster like any photograph. Authored SVG covers what a session can specify exactly, diagrams with countable elements, controls, flat shape systems, and it ends where drawing skill begins; an instruction-manual world does not convert its illustrations into diagrams, it makes them line-art illustrations. Produce such regions by regenerating them cleanly, with the approved comp and its embedded prompt as the reference for a fresh render at asset resolution; never crop pixels out of the comp itself, whose effective resolution sits far below asset grade. Dropping an image-native region instead of producing it is a scope decision the user makes at the approval point, never a silent flattening after it. Generated imagery is a material, not a claim: evidence rules bind assertions, specs, testimonials, and photographs presented as real, never render fidelity, so "no photography on hand" forbids fake proof, not an illustrated hero.
+The record is sampled, never estimated: read the comp's page **ground**, each dominant field, and each accent's actual hex from its pixels (ImageMagick, Python with PIL, any pixel-reading tool on the machine) and write the values into the same record. Take a flat field from any interior pixel, a textured or grainy one as the average of an interior patch (crop a swatch, scale it to one pixel), and a gradient as its two end colors; never sample an edge, where antialiasing blends neighbors into colors the design never chose. An adjective is a direction, not a record: cream covers everything from near-white to beige, charcoal a third of the value scale, and wherever no number pins a color, the rendition prior picks the spot. Sampled values supersede the palette chips on the decision and composition cards: those were authored before this comp existed, and a chip that disagrees with the comp's pixels is a draft the approval retired.
 
-The gate runs both ways: precise geometry, hard-edged shape systems, diagrams, expressive motion, shaders, and anything interactive are vector and GPU territory (SVG, canvas, WebGL), where a raster flattens what should move, scale, and respond, and code executed safely and professionally remains first-class there. A field or texture built from many small elements carries a quantity commitment either way: write down its approximate density and coverage ("thousands of glyphs over two-thirds of the fold, dense at the top fading into the path"), because a field rebuilt at a tenth of its density passes every checklist and still is not the design. TYPE rows carry the same discipline: name the face's compression class, and render one headline word against the comp before building on it; a visibly wider or lighter silhouette means the face is wrong, and every section built on it inherits the miss. Raster is for what the world paints; code is for what the world draws, animates, or reacts with, and choosing code there is ambition, not economy. Every `produce` entry is produced before the build ships, through the asset producer or in the current thread; an inventory with unproduced entries is an unfinished build, and this gate is where imagery-free pages come from when it is skipped.
+The medium column is where an approved design most often dies, so it obeys a gate: the medium is decided by what the comp region shows, never by what feels buildable in the current stack. A human figure, a product object, machinery, or any material with lighting and depth is raster whatever the stack; so is any texture by name alone: woven cloth, paper grain, fabric, leather, brushed metal need no depth argument, because a CSS gradient is not a texture medium and "layered CSS textures" is not a medium at all. Writing "silhouette" for a photographic figure, or "CSS" for a sculpted panel's finish, is not a medium choice; it is the quiet deletion of the approved design, and it is how a comp full of physical material becomes a flat page with the same section order. Style does not move this boundary: a comp region with perspective, shading, figure drawing, or dense mechanical detail is illustration however line-drawn it looks, and no build session can author illustration as vectors, so it regenerates as raster like any photograph. Authored SVG covers what a session can specify exactly (diagrams with countable elements, controls, flat shape systems) and ends where drawing skill begins; an instruction-manual world keeps its illustrations as line-art illustrations, not diagrams. Produce such regions by regenerating them cleanly, with the approved comp and its embedded prompt as the reference for a fresh render at asset resolution; never crop pixels out of the comp itself, whose effective resolution sits far below asset grade. Dropping an image-native region is a scope decision the user makes at the approval point, never a silent flattening after it. Generated imagery is a material, not a claim: evidence rules bind assertions, specs, testimonials, and photographs presented as real, never render fidelity; "no photography on hand" forbids fake proof, not an illustrated hero.
+
+The gate runs both ways: precise geometry, hard-edged shape systems, diagrams, expressive motion, shaders, and anything interactive are vector and GPU territory (SVG, canvas, WebGL), where a raster flattens what should move, scale, and respond. A field or texture built from many small elements carries a quantity commitment either way: write down its approximate density and coverage ("thousands of glyphs over two-thirds of the fold, dense at the top fading into the path"), because a field rebuilt at a tenth of its density passes every checklist and still is not the design. TYPE rows carry the same discipline: name the face's compression class, and render one headline word against the comp before building on it; a visibly wider or lighter silhouette means the face is wrong, and every section built on it inherits the miss. Raster is for what the world paints; code is for what the world draws, animates, or reacts with, and choosing code there is ambition, not economy. Every `produce` entry is produced before the build ships, through the asset producer or in the current thread; an inventory with unproduced entries is an unfinished build, and this gate is where imagery-free pages come from when it is skipped.
 
 Pay special attention to the dominant composition, signature use, image-native content, second-fold system, and any interaction the still image only implies.
 
-Treat the comp as a north star, not something to trace, and know what that allows: translation into semantic, responsive, accessible code, never recomposition. Keeping the palette and mood while redrawing the topology is a second art direction, not an adaptation. Do not rasterize core UI text or controls. Do not substitute a different visual driver after approval without asking.
+The comp is a north star, not something to trace, and know what that allows: translation into semantic, responsive, accessible code, never recomposition. Keeping the palette and mood while redrawing the topology is a second art direction, not an adaptation. Do not rasterize core UI text or controls. Do not substitute a different visual driver after approval without asking.
 
 ## Produce only the assets the build needs
 
-Generation context is part of the asset: a build composed by a thread that never saw the prompts places assets it does not understand. So prefer generating build-critical imagery in the build thread when the budget allows, and when a subagent produces assets instead, every asset must carry its prompt, and the builder reads those prompts before composing a single one of them. The carrier is uniform across harnesses: after generating any image with any tool, native or `generate-image.mjs` (which does it automatically), run `node .agent/skills/impeccable/scripts/embed-prompt.mjs <image> --prompt "<the prompt used>"` so the intent lives inside the file itself and survives copies between machines and harnesses; `--read` recovers it from any impeccable-generated image.
+Generation context is part of the asset: a build composed by a thread that never saw the prompts places assets it does not understand. Prefer generating build-critical imagery in the build thread when the budget allows; when a subagent produces assets instead, every asset carries its prompt, and the builder reads those prompts before composing. The carrier is uniform across harnesses: after generating any image with any tool, native or `generate-image.mjs` (which does it automatically), run `node .agent/skills/impeccable/scripts/embed-prompt.mjs <image> --prompt "<prompt>"` with the exact string the generation tool received, pasted whole, so the intent lives inside the file and survives copies between machines and harnesses; a summary reconstructed from memory records an asset that was never made. `--read` recovers the prompt from any impeccable-generated image, and `--scan <dir>` lists every raster in a directory still missing one. The embedded prompt plus the asset's row in the written inventory is the raster's **provenance**, and every raster the artifact references carries it; a sourced, stock, or pre-existing raster with no generation prompt embeds its origin instead.
 
-When the harness runs subagents, spawn the shipped asset producer every time, even when the inventory's produce bucket looks empty: its manifest is the independent second opinion on your media, and runs that skipped the spawn are the runs whose cotton became CSS. An honestly empty manifest costs one cheap spawn; a wrongly empty produce bucket costs the build its materials. Use the producer, `impeccable-asset-producer` (`impeccable_asset_producer` in codex; `/impeccable-asset-producer` in Cursor; on GitHub Copilot say "Use the impeccable-asset-producer agent"): give it the approved comp, output paths, required dimensions and formats, transparency needs, crop notes, and what must remain semantic code. Otherwise produce the minimum required assets in the current thread by the book: load [degraded/asset-producer.md](degraded/asset-producer.md) and follow it inline, with whatever generation exists, the native tool or generate-image.mjs.
+Provenance is owed for the run, not the build phase: a raster created or replaced later, in a fix batch or a reviewer's rebuild, is produced under this same section, prompt embedded and inventory row added, because the inventory is how the next thread knows what ships. A raster a fix abandons or supersedes is deleted from the assets directory in the same batch; an unreferenced raster with no record is a provenance leak, not a spare.
+
+When the harness runs subagents, spawn the shipped asset producer every time, even when the inventory's produce bucket looks empty: its manifest is the independent second opinion on your media, and the runs that skipped the spawn are the runs whose cotton became CSS. An honestly empty manifest costs one cheap spawn; a wrongly empty produce bucket costs the build its materials. Use `impeccable-asset-producer` (`impeccable_asset_producer` in codex; `/impeccable-asset-producer` in Cursor; on GitHub Copilot say "Use the impeccable-asset-producer agent"): give it the approved comp, output paths, required dimensions and formats, transparency needs, crop notes, and what must remain semantic code. Without subagents, produce the minimum required assets in the current thread by the book: load [degraded/asset-producer.md](degraded/asset-producer.md) and follow it inline, with whatever generation exists.
 
 Convert images with a converter context.mjs reported at boot (the IMAGE_TOOLS line); probe only when it reported none, at most once per session, never per image.
 

.agent/skills/impeccable/scripts/context.mjs

@@ -1013,6 +1013,27 @@ async function fetchLatestSkillVersion() {
   }
 }
 
+// Destroy fetch's global undici dispatcher before process.exit(): a live
+// keep-alive socket trips a libuv assertion on Windows/Node 24 after a
+// successful boot (nodejs/node#56645, issue #573).
+async function destroyFetchDispatcher() {
+  const dispatcher = globalThis[Symbol.for('undici.globalDispatcher.1')];
+  if (dispatcher && typeof dispatcher.destroy === 'function') {
+    try { await dispatcher.destroy(); } catch { /* exit regardless */ }
+  }
+}
+
+// Drain the boot payload before process.exit(): a live pipe that has not
+// flushed yet is truncated when Node tears down (issue #573 review). Then
+// close fetch so Windows teardown does not abort on the keep-alive socket.
+async function finishCli(output) {
+  await new Promise((resolve) => {
+    process.stdout.write(output, () => resolve());
+  });
+  await destroyFetchDispatcher();
+  process.exit(0);
+}
+
 // Two instructions used to sit in one directive: ask, and "if they agree, run
 // it". Nothing gated the second on an answer, and the same sentence said to
 // continue without waiting, so a run that could never establish agreement was
@@ -1159,8 +1180,7 @@ async function cli() {
     appendImageToolsDirective(parts);
     appendStalenessDirective(parts, ctx, cliOptions);
     if (updateDirective) parts.push(updateDirective);
-    process.stdout.write(parts.join('\n\n---\n\n') + '\n');
-    process.exit(0);
+    await finishCli(parts.join('\n\n---\n\n') + '\n');
   }
   const parts = [`# PRODUCT.md\n\n${ctx.product.trim()}`];
   if (ctx.hasDesign) {
@@ -1206,7 +1226,7 @@ async function cli() {
     }
   }
   if (updateDirective) parts.push(updateDirective);
-  process.stdout.write(parts.join('\n\n---\n\n') + '\n');
+  await finishCli(parts.join('\n\n---\n\n') + '\n');
 }
 
 function parseCliOptions(args) {

.agent/skills/impeccable/scripts/detector/browser/injected/index.mjs

@@ -626,7 +626,7 @@ if (IS_BROWSER) {
       if (currentStyle.filter && currentStyle.filter !== 'none') reasons.add('filter');
       if (currentStyle.backdropFilter && currentStyle.backdropFilter !== 'none') reasons.add('backdrop filter');
 
-      const solidBg = parseRgb(currentStyle.backgroundColor);
+      const solidBg = parseRgb(currentStyle.backgroundColor) || parseAnyColor(currentStyle.backgroundColor);
       if (solidBg && solidBg.a >= 0.95 && (!bgImage || bgImage === 'none')) break;
       current = current.parentElement;
     }
@@ -688,7 +688,7 @@ if (IS_BROWSER) {
       // starve the url()-backed texts this mode exists to sample.
       if (options.imageOnly && !reasons.includes('image background')) continue;
 
-      const textColor = parseRgb(style.color);
+      const textColor = parseRgb(style.color) || parseAnyColor(style.color);
       const fontSize = parseFloat(style.fontSize) || 16;
       const fontWeight = parseInt(style.fontWeight) || 400;
       const isLargeText = fontSize >= WCAG_LARGE_TEXT_PX || (fontSize >= WCAG_LARGE_BOLD_TEXT_PX && fontWeight >= 700);
@@ -985,7 +985,7 @@ if (IS_BROWSER) {
         return sample;
       }
     }
-    const bg = parseRgb(style.backgroundColor);
+    const bg = parseRgb(style.backgroundColor) || parseAnyColor(style.backgroundColor);
     if (bg && bg.a > 0.05) return { status: 'sampled', color: bg, method: 'solid-background' };
     return { status: 'unresolved', reason: 'no readable background' };
   }
@@ -1115,7 +1115,7 @@ if (IS_BROWSER) {
     }
 
     const style = getComputedStyle(el);
-    const textColor = parseRgb(style.color) || candidate.textColor;
+    const textColor = parseRgb(style.color) || parseAnyColor(style.color) || candidate.textColor;
     if (!textColor) return { ...candidate, status: 'unresolved', confidence: 'none', reason: 'unreadable text color' };
 
     const rect = getDirectTextRect(el) || el.getBoundingClientRect();

.agent/skills/impeccable/scripts/detector/cli/main.mjs

@@ -105,6 +105,13 @@ function formatFindings(findings, jsonMode) {
 // `optionsFor` maps a local path to scan options carrying that path's own
 // project design system (or base options when null). Falls back to a plain
 // object so direct/legacy callers still work.
+async function detectLocalFile(filePath, options) {
+  if (HTML_EXTENSIONS.has(path.extname(filePath).toLowerCase())) {
+    return detectHtml(filePath, options);
+  }
+  return detectText(fs.readFileSync(filePath, 'utf-8'), filePath, options);
+}
+
 async function handleStdin(optionsFor = () => ({})) {
   const resolve = typeof optionsFor === 'function' ? optionsFor : () => optionsFor;
   const chunks = [];
@@ -114,9 +121,7 @@ async function handleStdin(optionsFor = () => ({})) {
     const parsed = JSON.parse(input);
     const fp = parsed?.tool_input?.file_path;
     if (fp && fs.existsSync(fp)) {
-      const options = resolve(fp);
-      return HTML_EXTENSIONS.has(path.extname(fp).toLowerCase())
-        ? detectHtml(fp, options) : detectText(fs.readFileSync(fp, 'utf-8'), fp, options);
+      return detectLocalFile(fp, resolve(fp));
     }
   } catch { /* not JSON */ }
   return detectText(input, '<stdin>', resolve(null));
@@ -374,16 +379,10 @@ async function detectCli() {
           }
 
           for (const file of files) {
-            const ext = path.extname(file).toLowerCase();
             // Each file resolves its own project design system (cached by root),
             // so a scan spanning sibling projects applies the right rules per file.
             const fileOptions = scanOptionsFor(file);
-            let fileFindings;
-            if (HTML_EXTENSIONS.has(ext)) {
-              fileFindings = await detectHtml(file, fileOptions);
-            } else {
-              fileFindings = detectText(fs.readFileSync(file, 'utf-8'), file, fileOptions);
-            }
+            const fileFindings = await detectLocalFile(file, fileOptions);
             // Annotate findings with import context
             const importers = importedByMap.get(file);
             if (importers && importers.size > 0) {
@@ -396,13 +395,8 @@ async function detectCli() {
           }
         } else if (stat.isFile()) {
           if (shouldIgnoreDetectionFile(resolved, process.cwd(), detectionConfig)) continue;
-          const ext = path.extname(resolved).toLowerCase();
           const fileOptions = scanOptionsFor(resolved);
-          if (HTML_EXTENSIONS.has(ext)) {
-            allFindings.push(...await detectHtml(resolved, fileOptions));
-          } else {
-            allFindings.push(...detectText(fs.readFileSync(resolved, 'utf-8'), resolved, fileOptions));
-          }
+          allFindings.push(...await detectLocalFile(resolved, fileOptions));
         }
       }
     } finally {

.agent/skills/impeccable/scripts/detector/design-system.mjs

@@ -13,6 +13,11 @@ const FALLBACK_DIRS = ['.agents/context', 'docs'];
 // CLI can't import (separate tree). `.git` and `package.json` are the common
 // boundaries; `.impeccable` is our own project marker.
 const PROJECT_ROOT_MARKERS = ['.git', 'package.json', '.impeccable'];
+// Monorepo-root recognition, mirroring context.mjs's isMonorepoRoot: declared
+// workspace globs (package.json `workspaces`, pnpm-workspace.yaml `packages:`)
+// or a marker file beside apps/ or packages/ children.
+const MONOREPO_MARKER_FILES = ['pnpm-workspace.yaml', 'turbo.json', 'nx.json', 'lerna.json'];
+const MONOREPO_FALLBACK_PROJECT_DIRS = ['apps', 'packages'];
 const COLOR_CHANNEL_TOLERANCE = 6;
 // Shadow blacks at different alphas are different tokens (0.28 vs 0.55 is the
 // difference between a documented shadow and drift), so shadow matching cannot
@@ -575,14 +580,179 @@ function designSystemStartDir(targetPath, cwd = process.cwd()) {
   }
 }
 
+// Same two groups as context.mjs's readProjectPatternGroups: Impeccable
+// projectRoots govern any path they match (positive or negated); package-manager
+// globs only apply to paths the Impeccable group does not match.
+function readWorkspacePatternGroups(dir) {
+  const impeccable = [];
+  for (const name of ['config.json', 'config.local.json']) {
+    const roots = safeReadJson(path.join(dir, '.impeccable', name))?.projectRoots;
+    if (Array.isArray(roots)) {
+      impeccable.push(...roots.filter(entry => typeof entry === 'string' && entry.trim()).map(entry => entry.trim()));
+    }
+  }
+  const pkg = [];
+  const workspaces = safeReadJson(path.join(dir, 'package.json'))?.workspaces;
+  if (Array.isArray(workspaces)) pkg.push(...workspaces);
+  else if (Array.isArray(workspaces?.packages)) pkg.push(...workspaces.packages);
+  const lernaPackages = safeReadJson(path.join(dir, 'lerna.json'))?.packages;
+  if (Array.isArray(lernaPackages)) pkg.push(...lernaPackages);
+  try {
+    let inPackages = false;
+    for (const line of fs.readFileSync(path.join(dir, 'pnpm-workspace.yaml'), 'utf-8').split(/\r?\n/)) {
+      const trimmed = stripInlineYamlComment(line).trim();
+      if (!trimmed || trimmed.startsWith('#')) continue;
+      const flow = trimmed.match(/^packages:\s*\[(.*)\]\s*$/);
+      if (flow) {
+        pkg.push(...flow[1].split(',').map(entry => entry.trim().replace(/^['"]|['"]$/g, '')).filter(Boolean));
+        break;
+      }
+      if (/^packages:\s*$/.test(trimmed)) { inPackages = true; continue; }
+      if (!inPackages) continue;
+      const item = trimmed.match(/^-\s*(.+)$/);
+      if (item) pkg.push(item[1].trim().replace(/^['"]|['"]$/g, ''));
+      else if (/^[A-Za-z0-9_-]+:\s*/.test(trimmed)) break;
+    }
+  } catch { /* no pnpm-workspace.yaml */ }
+  return [impeccable, pkg];
+}
+
+function readWorkspacePatterns(dir) {
+  return readWorkspacePatternGroups(dir).flat();
+}
+
+function isMonorepoRoot(dir) {
+  if (readWorkspacePatterns(dir).some(pattern => !String(pattern).trim().startsWith('!'))) return true;
+  if (!MONOREPO_MARKER_FILES.some(file => fs.existsSync(path.join(dir, file)))) return false;
+  return MONOREPO_FALLBACK_PROJECT_DIRS.some(name => {
+    try {
+      return fs.readdirSync(path.join(dir, name), { withFileTypes: true }).some(entry => entry.isDirectory());
+    } catch {
+      return false;
+    }
+  });
+}
+
+function monorepoOwnsPath(root, boundaryDir) {
+  const rel = path.relative(root, boundaryDir);
+  if (!rel || rel.startsWith('..') || path.isAbsolute(rel)) return false;
+  const relSegments = rel.split(path.sep).filter(Boolean);
+
+  function normalizeWorkspacePattern(pattern) {
+    return String(pattern || '')
+      .trim()
+      .replace(/^['"]|['"]$/g, '')
+      .replace(/^\.\//, '')
+      .replace(/\/+$/, '');
+  }
+
+  function escapeRegExp(s) {
+    return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+  }
+
+  function segmentMatches(patternSegment, relSegment) {
+    if (patternSegment === '*') return true;
+    if (!patternSegment.includes('*')) return patternSegment === relSegment;
+    const re = new RegExp(`^${escapeRegExp(patternSegment).replace(/\\\*/g, '[^/]*')}$`);
+    return re.test(relSegment);
+  }
+
+  function matchGlobSegments(patternSegments, relSegments) {
+    function rec(pi, ri) {
+      if (pi === patternSegments.length) return ri === relSegments.length;
+      if (patternSegments[pi] === '**') {
+        if (pi === patternSegments.length - 1) return true;
+        for (let k = ri; k <= relSegments.length; k++) {
+          if (rec(pi + 1, k)) return true;
+        }
+        return false;
+      }
+      if (ri >= relSegments.length) return false;
+      if (!segmentMatches(patternSegments[pi], relSegments[ri])) return false;
+      return rec(pi + 1, ri + 1);
+    }
+    return rec(0, 0);
+  }
+
+  // Negations like !packages/excluded must also cover nested dirs under that path.
+  function matchesNegation(pattern) {
+    const patternSegments = normalizeWorkspacePattern(pattern).split('/').filter(Boolean);
+    if (!patternSegments.length) return false;
+    if (patternSegments.includes('**')) return matchGlobSegments(patternSegments, relSegments);
+    if (relSegments.length < patternSegments.length) return false;
+    for (let i = 0; i < patternSegments.length; i++) {
+      if (!segmentMatches(patternSegments[i], relSegments[i])) return false;
+    }
+    return true;
+  }
+
+  // Positive globs identify workspace packages at exact depth (`*` is a direct
+  // child). A nested package.json under that package is still owned: the
+  // ancestor directory of glob length must itself be a package.
+  function positiveOwns(pattern) {
+    const patternSegments = normalizeWorkspacePattern(pattern).split('/').filter(Boolean);
+    if (!patternSegments.length) return false;
+    if (patternSegments.includes('**')) return matchGlobSegments(patternSegments, relSegments);
+    if (relSegments.length < patternSegments.length) return false;
+    for (let i = 0; i < patternSegments.length; i++) {
+      if (!segmentMatches(patternSegments[i], relSegments[i])) return false;
+    }
+    if (relSegments.length === patternSegments.length) return true;
+    const ancestorDir = path.join(root, ...relSegments.slice(0, patternSegments.length));
+    return fs.existsSync(path.join(ancestorDir, 'package.json'));
+  }
+
+  function groupOwns(rawPatterns) {
+    const patterns = rawPatterns.map(normalizeWorkspacePattern).filter(Boolean);
+    if (!patterns.length) return null;
+    const excluded = patterns.some((pattern) => (
+      pattern.startsWith('!') && matchesNegation(pattern.slice(1))
+    ));
+    const included = patterns.filter((pattern) => !pattern.startsWith('!')).some(positiveOwns);
+    if (!excluded && !included) return null;
+    if (excluded) return false;
+    return true;
+  }
+
+  const [impeccable, pkg] = readWorkspacePatternGroups(root);
+  const fromImpeccable = groupOwns(impeccable);
+  if (fromImpeccable !== null) return fromImpeccable;
+  const fromPkg = groupOwns(pkg);
+  if (fromPkg !== null) return fromPkg;
+  if ([...impeccable, ...pkg].some((pattern) => !normalizeWorkspacePattern(pattern).startsWith('!'))) {
+    return false;
+  }
+  return relSegments.length >= 2 && MONOREPO_FALLBACK_PROJECT_DIRS.includes(relSegments[0]);
+}
+
+// Both forms of the home directory. The walk compares path strings, and a
+// symlinked home (e.g. /home -> /var/home) never string-matches the physical
+// paths a cwd-resolved target produces, which would let the post-boundary walk
+// sail through $HOME and inherit from it.
+function homeDirForms() {
+  const homeDir = path.resolve(os.homedir());
+  const forms = new Set([homeDir]);
+  try {
+    forms.add(fs.realpathSync(homeDir));
+  } catch { /* keep the logical form only */ }
+  return forms;
+}
+
 // Walk up from `startDir` to the directory that governs the target's design
 // system, mirroring skill/scripts/context.mjs's project-boundary semantics:
 //
 //   - A directory carrying a DESIGN.md (directly or in a fallback dir) IS the
 //     design root — that's where the rules live.
 //   - A directory carrying a project marker (.git / package.json / .impeccable)
-//     but no DESIGN.md is a project BOUNDARY: the walk stops with no design
-//     system, so a sibling project never inherits a parent's or cwd's rules.
+//     but no DESIGN.md is a project BOUNDARY. A nested package.json inherits
+//     the ancestor DESIGN.md only when that ancestor's workspace declarations
+//     include the path (negations win; a nested package under a matched
+//     workspace still inherits). Marker-only roots (turbo/nx/lerna/pnpm
+//     with no globs) still own apps/<name> and packages/<name>. A stray nested
+//     package that matches no glob does not inherit. This is detect's
+//     contamination contract, not skill-context's repoRoot fallback for
+//     excluded paths. A nested separate repository (.git with no workspace
+//     declaration) still inherits nothing (issue #570).
 //   - Reaching the home directory / filesystem root with neither means no
 //     design system at all — never process.cwd()'s.
 //
@@ -590,15 +760,33 @@ function designSystemStartDir(targetPath, cwd = process.cwd()) {
 // runs out. This is the fix for cross-project contamination.
 export function findDesignRoot(startDir) {
   let dir = path.resolve(startDir);
-  const homeDir = path.resolve(os.homedir());
+  const homeDirs = homeDirForms();
+  let boundary = null;
   while (true) {
-    if (resolveDesignMdPath(dir)) return { dir, hasDesign: true };
-    if (PROJECT_ROOT_MARKERS.some((marker) => fs.existsSync(path.join(dir, marker)))) {
-      return { dir, hasDesign: false };
+    if (!boundary && resolveDesignMdPath(dir)) return { dir, hasDesign: true };
+    if (boundary) {
+      // Past the boundary the walk only looks for the monorepo root that owns
+      // the workspace path (workspace globs including negations, or marker-only
+      // apps/packages fallback). Monorepo-root before .git, same order as
+      // context.mjs: a workspace root carrying its own .git is still recognized,
+      // while a .git that declares no workspaces is a separate repository and
+      // stops the walk with nothing inherited. The home directory is never an
+      // owning root, same as context.mjs's findMonorepoRoot, which stops at
+      // homeDir before its monorepo check.
+      if (!homeDirs.has(dir) && isMonorepoRoot(dir)) {
+        if (monorepoOwnsPath(dir, boundary.dir)) return { dir, hasDesign: !!resolveDesignMdPath(dir) };
+        return boundary;
+      }
+      if (fs.existsSync(path.join(dir, '.git'))) return boundary;
+    } else if (PROJECT_ROOT_MARKERS.some((marker) => fs.existsSync(path.join(dir, marker)))) {
+      boundary = { dir, hasDesign: false };
+      // A boundary that is itself a monorepo root, or a separate repository
+      // with its own .git, inherits nothing from above.
+      if (isMonorepoRoot(dir) || fs.existsSync(path.join(dir, '.git'))) return boundary;
     }
-    if (dir === homeDir) return null;
+    if (homeDirs.has(dir)) return boundary;
     const parent = path.dirname(dir);
-    if (parent === dir) return null;
+    if (parent === dir) return boundary;
     dir = parent;
   }
 }

.agent/skills/impeccable/scripts/detector/detect-antipatterns-browser.js

@@ -773,14 +773,22 @@ function extractColorFunctionTokens(value) {
 function parseGradientColors(bgImage) {
   if (!bgImage || !bgImage.includes('gradient')) return [];
   const colors = [];
+  const tokenSpans = [];
+  let from = 0;
   // Stops arrive in whatever syntax the author wrote and the browser kept.
   // A dark ground painted as `linear-gradient(oklch(...), oklch(...))` used
   // to read as a gradient with no stops at all.
   for (const token of extractColorFunctionTokens(bgImage)) {
+    const start = bgImage.indexOf(token, from);
+    if (start < 0) break;
+    tokenSpans.push({ start, end: start + token.length });
+    from = start + token.length;
     const c = parseAnyColor(token);
     if (c) colors.push(c);
   }
   for (const m of bgImage.matchAll(/#([0-9a-f]{6}|[0-9a-f]{3})\b/gi)) {
+    // Nested hex inside color-mix is an ingredient, not a stop (issue #578).
+    if (tokenSpans.some(s => m.index >= s.start && m.index < s.end)) continue;
     const h = m[1];
     if (h.length === 6) {
       colors.push({ r: parseInt(h.slice(0,2),16), g: parseInt(h.slice(2,4),16), b: parseInt(h.slice(4,6),16), a: 1 });
@@ -1955,20 +1963,19 @@ function scanCssTextForGlow(content) {
   return results;
 }
 
-// Decorative grid or line-field backgrounds drawn with hairline
+// Decorative two-axis grid backgrounds drawn with hairline
 // linear-gradient layers tiled by a fixed pixel cell. Shared by the HTML
 // pattern pass and the regex source engine so standalone CSS, component
 // styles, and inline styles receive the same coverage. Both signals must
 // co-occur in one declaration block; unrelated rules must not add up across
-// the file. Returns [{ index, snippet }], capped at one finding per source to
-// match the page-level HTML check's existing behavior.
+// the file. A single hairline is a line, divider, or rail, not a grid, even
+// when tiled by a 2D px cell. Returns [{ index, snippet }], capped at one
+// finding per source to match the page-level HTML check's existing behavior.
 function scanCssTextForGridBackground(content) {
   const hairlineRe = /\b\d{1,3}px\s*,\s*transparent\s+\d{1,3}px/gi;
   const invertedHairlineRe = /transparent\s+calc\(100%\s*-\s*\d{1,3}px\)/gi;
   const sizeDeclPxRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\b/i;
-  const sizeDeclPxPairRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\s+\d{1,3}px/i;
   const shorthandPxAnyRe = /\/\s*\d{1,3}px\b/;
-  const shorthandPxPairRe = /\/\s*\d{1,3}px\s+\d{1,3}px/;
   const bgDeclRe = /\bbackground(?:-image)?\s*:\s*([^;{}"']*)/gi;
   const blockRe = /\{([^{}]*)\}|style\s*=\s*"([^"]*)"|style\s*=\s*'([^']*)'/gi;
   let blk;
@@ -1985,13 +1992,10 @@ function scanCssTextForGridBackground(content) {
     }
     if (hairlineCount === 0) continue;
     const hasPxCell = sizeDeclPxRe.test(block) || shorthandPxAnyRe.test(bgJoined);
-    const hasPxPairCell = sizeDeclPxPairRe.test(block) || shorthandPxPairRe.test(bgJoined);
-    if ((hairlineCount >= 2 && hasPxCell) || hasPxPairCell) {
+    if (hairlineCount >= 2 && hasPxCell) {
       return [{
         index: blk.index,
-        snippet: hairlineCount >= 2
-          ? 'two-axis grid-line gradient background'
-          : 'px-tiled hairline line-field background',
+        snippet: 'two-axis grid-line gradient background',
       }];
     }
   }
@@ -3986,7 +3990,7 @@ function checkElementAIPaletteDOM(el) {
   }
 
   // Check for neon text (vivid cyan/purple color on dark background)
-  const textColor = parseRgb(style.color);
+  const textColor = parseRgb(style.color) || parseAnyColor(style.color);
   if (textColor && hasChroma(textColor, 80)) {
     const hue = getHue(textColor);
     const isAIPalette = (hue >= 160 && hue <= 200) || (hue >= 260 && hue <= 310);
@@ -7281,7 +7285,7 @@ if (IS_BROWSER) {
       if (currentStyle.filter && currentStyle.filter !== 'none') reasons.add('filter');
       if (currentStyle.backdropFilter && currentStyle.backdropFilter !== 'none') reasons.add('backdrop filter');
 
-      const solidBg = parseRgb(currentStyle.backgroundColor);
+      const solidBg = parseRgb(currentStyle.backgroundColor) || parseAnyColor(currentStyle.backgroundColor);
       if (solidBg && solidBg.a >= 0.95 && (!bgImage || bgImage === 'none')) break;
       current = current.parentElement;
     }
@@ -7343,7 +7347,7 @@ if (IS_BROWSER) {
       // starve the url()-backed texts this mode exists to sample.
       if (options.imageOnly && !reasons.includes('image background')) continue;
 
-      const textColor = parseRgb(style.color);
+      const textColor = parseRgb(style.color) || parseAnyColor(style.color);
       const fontSize = parseFloat(style.fontSize) || 16;
       const fontWeight = parseInt(style.fontWeight) || 400;
       const isLargeText = fontSize >= WCAG_LARGE_TEXT_PX || (fontSize >= WCAG_LARGE_BOLD_TEXT_PX && fontWeight >= 700);
@@ -7640,7 +7644,7 @@ if (IS_BROWSER) {
         return sample;
       }
     }
-    const bg = parseRgb(style.backgroundColor);
+    const bg = parseRgb(style.backgroundColor) || parseAnyColor(style.backgroundColor);
     if (bg && bg.a > 0.05) return { status: 'sampled', color: bg, method: 'solid-background' };
     return { status: 'unresolved', reason: 'no readable background' };
   }
@@ -7770,7 +7774,7 @@ if (IS_BROWSER) {
     }
 
     const style = getComputedStyle(el);
-    const textColor = parseRgb(style.color) || candidate.textColor;
+    const textColor = parseRgb(style.color) || parseAnyColor(style.color) || candidate.textColor;
     if (!textColor) return { ...candidate, status: 'unresolved', confidence: 'none', reason: 'unreadable text color' };
 
     const rect = getDirectTextRect(el) || el.getBoundingClientRect();

.agent/skills/impeccable/scripts/detector/engines/regex/detect-text.mjs

@@ -42,6 +42,7 @@ function shouldRunPageAnalyzers(content, filePath) {
 }
 
 const JS_SOURCE_EXTS = new Set(['.js', '.jsx', '.ts', '.tsx', '.mjs', '.cjs']);
+const STYLESHEET_EXTS = new Set(['.css', '.scss', '.sass', '.less']);
 const REGEX_PREFIX_KEYWORDS = new Set(['await', 'case', 'default', 'delete', 'do', 'else', 'in', 'instanceof', 'new', 'of', 'return', 'throw', 'typeof', 'void', 'yield']);
 const BLOCK_BRACE_PREFIX_KEYWORDS = new Set(['do', 'else', 'finally', 'try']);
 
@@ -256,6 +257,153 @@ function stripCssComments(content) {
   return content.replace(/\/\*[\s\S]*?\*\//g, comment => comment.replace(/[^\n]/g, ' '));
 }
 
+function blankHtmlComments(text) {
+  return text.replace(/<!--[\s\S]*?-->/g, comment => comment.replace(/[^\n]/g, ' '));
+}
+
+function blankCssLineCommentsInStyleBlocks(text) {
+  const re = /<style\b[^>]*>([\s\S]*?)<\/style>/gi;
+  let output = '';
+  let lastIndex = 0;
+  let match;
+  while ((match = re.exec(text)) !== null) {
+    const inner = match[1];
+    const openLength = match[0].length - inner.length - '</style>'.length;
+    output += text.slice(lastIndex, match.index);
+    output += match[0].slice(0, openLength);
+    output += blankCssLineComments(inner);
+    output += match[0].slice(openLength + inner.length);
+    lastIndex = re.lastIndex;
+  }
+  return output + text.slice(lastIndex);
+}
+
+function blankHtmlAndCssCommentsOutsideScripts(text) {
+  const re = /<script\b[^>]*>[\s\S]*?<\/script>/gi;
+  let output = '';
+  let lastIndex = 0;
+  let match;
+  while ((match = re.exec(text)) !== null) {
+    output += blankCssLineCommentsInStyleBlocks(stripCssComments(blankHtmlComments(text.slice(lastIndex, match.index))));
+    output += match[0];
+    lastIndex = re.lastIndex;
+  }
+  return output + blankCssLineCommentsInStyleBlocks(stripCssComments(blankHtmlComments(text.slice(lastIndex))));
+}
+
+function blankCssLineComments(text) {
+  let output = '';
+  let state = 'code';
+  let urlDepth = 0;
+  for (let i = 0; i < text.length; i++) {
+    const char = text[i];
+    const next = text[i + 1];
+    if (state === 'line') {
+      if (char === '\n') {
+        output += '\n';
+        state = 'code';
+      } else {
+        output += ' ';
+      }
+      continue;
+    }
+    if (state === 'single' || state === 'double') {
+      output += char;
+      if (char === '\\' && next) {
+        output += next;
+        i++;
+      } else if ((state === 'single' && char === "'") || (state === 'double' && char === '"')) {
+        state = 'code';
+      }
+      continue;
+    }
+    const prev = output.length ? output[output.length - 1] : '';
+    if (char === '/' && next === '/' && urlDepth === 0 && prev !== ':' && prev !== '(' && prev !== '\\') {
+      output += '  ';
+      i++;
+      state = 'line';
+      continue;
+    }
+    if (char === "'") state = 'single';
+    else if (char === '"') state = 'double';
+    if (char === '(') {
+      const behind = output.replace(/\s+$/, '');
+      if (urlDepth > 0 || /url$/i.test(behind)) urlDepth++;
+    } else if (char === ')' && urlDepth) {
+      urlDepth--;
+    }
+    output += char;
+  }
+  return output;
+}
+
+function findAstroFrontmatterClose(text) {
+  if (!text.startsWith('---')) return -1;
+  let cursor = text.indexOf('\n');
+  if (cursor === -1) return -1;
+  cursor += 1;
+  while (cursor < text.length) {
+    if (text[cursor - 1] === '\n' && text.startsWith('---', cursor)) {
+      let end = cursor + 3;
+      while (text[end] === ' ' || text[end] === '\t') end++;
+      if (end >= text.length || text[end] === '\n' || text[end] === '\r') return cursor - 1;
+    }
+    const char = text[cursor];
+    const next = text[cursor + 1];
+    if (char === "'" || char === '"') {
+      const close = findQuotedStringEnd(text, cursor, char);
+      if (close === -1) return -1;
+      cursor = close + 1;
+      continue;
+    }
+    if (char === '`') {
+      const close = findTemplateLiteralEnd(text, cursor);
+      if (close === -1) return -1;
+      cursor = close + 1;
+      continue;
+    }
+    if (char === '/' && next === '/') {
+      const lineEnd = text.indexOf('\n', cursor);
+      if (lineEnd === -1) return -1;
+      cursor = lineEnd;
+      continue;
+    }
+    if (char === '/' && next === '*') {
+      const commentEnd = text.indexOf('*/', cursor + 2);
+      if (commentEnd === -1) return -1;
+      cursor = commentEnd + 2;
+      continue;
+    }
+    if (char === '/' && next !== '/' && next !== '*') {
+      const close = findRegexLiteralEnd(text, cursor);
+      if (close !== -1) {
+        cursor = close + 1;
+        continue;
+      }
+    }
+    cursor++;
+  }
+  return -1;
+}
+
+function blankAstroFrontmatterComments(text) {
+  const close = findAstroFrontmatterClose(text);
+  if (close === -1) return text;
+  return stripJsComments(text.slice(0, close)) + text.slice(close);
+}
+
+function blankCommentsForMatchers(text, ext) {
+  if (PAGE_ANALYZER_EXTS.has(ext)) {
+    const withFrontmatter = ext === '.astro' ? blankAstroFrontmatterComments(text) : text;
+    return blankHtmlAndCssCommentsOutsideScripts(withFrontmatter);
+  }
+  if (STYLESHEET_EXTS.has(ext)) {
+    const withoutBlocks = stripCssComments(text);
+    return ext === '.css' ? withoutBlocks : blankCssLineComments(withoutBlocks);
+  }
+  return text;
+}
+
 function firstOverusedGoogleFont(text) {
   return extractGoogleFontFamilies(text).find(f => OVERUSED_FONTS.has(f)) || '';
 }
@@ -1028,14 +1176,13 @@ function detectText(content, filePath, options = {}) {
   const ext = extFromFilePath(filePath);
   const commentStrippedSource = JS_SOURCE_EXTS.has(ext) ? stripJsComments(content, {
     jsx: ext === '.js' || ext === '.jsx' || ext === '.tsx',
-  }) : content;
+  }) : blankCommentsForMatchers(content, ext);
   const source = stripCssInJsComments(commentStrippedSource, ext);
   const lines = source.split('\n');
 
   // Run regex matchers on the full file content (catches Tailwind classes, inline styles)
   // Enable block context for CSS files where related properties span multiple lines
-  const cssLike = new Set(['.css', '.scss', '.sass', '.less']);
-  findings.push(...runRegexMatchers(lines, filePath, 0, cssLike.has(ext) || null, {
+  findings.push(...runRegexMatchers(lines, filePath, 0, STYLESHEET_EXTS.has(ext) || null, {
     profile,
     phase: 'source',
   }));
@@ -1050,7 +1197,7 @@ function detectText(content, filePath, options = {}) {
     scanCssTextForPseudoStripe(text).map(hit =>
       finding(hit.id, filePath, hit.snippet, lineOffset + text.slice(0, hit.index).split('\n').length));
 
-  if (cssLike.has(ext)) {
+  if (STYLESHEET_EXTS.has(ext)) {
     findings.push(...scanInsetStripeCss(content, filePath));
     findings.push(...pseudoStripeFindings(content, 0));
   }
@@ -1078,7 +1225,8 @@ function detectText(content, filePath, options = {}) {
     }, () => extractStyleBlocks(content, ext))
     : extractStyleBlocks(content, ext);
   for (const block of styleBlocks) {
-    const blockLines = block.content.split('\n');
+    const blockContent = blankCssLineComments(stripCssComments(block.content));
+    const blockLines = blockContent.split('\n');
     findings.push(...runRegexMatchers(blockLines, filePath, block.startLine - 1, true, {
       profile,
       phase: 'style-block',
@@ -1089,8 +1237,8 @@ function detectText(content, filePath, options = {}) {
     // 1-based, so the offset is startLine - 2; startLine - 1 double-counted and
     // reported every selector one line low. runRegexMatchers keeps startLine - 1
     // because it indexes its split lines from zero.
-    findings.push(...scanInsetStripeCss(block.content, filePath, block.startLine - 2));
-    findings.push(...pseudoStripeFindings(block.content, block.startLine - 2));
+    findings.push(...scanInsetStripeCss(blockContent, filePath, block.startLine - 2));
+    findings.push(...pseudoStripeFindings(blockContent, block.startLine - 2));
   }
 
   // Extract and scan CSS-in-JS template literals

.agent/skills/impeccable/scripts/detector/engines/static-html/css-cascade.mjs

@@ -835,10 +835,11 @@ class StaticElement {
     }
   }
   closest(selector) {
+    const matcher = this._doc.matcherFor(selector);
     let cur = this.node;
     while (cur && cur.type === 'tag') {
       try {
-        if (this._doc.is(cur, selector)) return this._doc.wrap(cur);
+        if (matcher(cur)) return this._doc.wrap(cur);
       } catch {
         return null;
       }
@@ -862,9 +863,10 @@ class StaticDocument {
     this.root = root;
     this.selectAll = modules.selectAll;
     this.selectOne = modules.selectOne;
-    this.is = modules.is;
+    this.compile = modules.compile;
     this.domutils = modules.domutils;
     this._wrappers = new WeakMap();
+    this._compiledSelectors = new Map();
     this._styleMap = new WeakMap();
     this._hoverStyleMap = new WeakMap();
     this._accentDashPseudo = new WeakSet();
@@ -882,6 +884,20 @@ class StaticDocument {
     }
     return wrapped;
   }
+  matcherFor(selector) {
+    let matcher = this._compiledSelectors.get(selector);
+    if (!matcher) {
+      try {
+        matcher = this.compile(selector);
+      } catch (err) {
+        // Cache the failure as a rethrower so a bad selector still reaches
+        // closest()'s catch on every call, first and repeat alike.
+        matcher = () => { throw err; };
+      }
+      this._compiledSelectors.set(selector, matcher);
+    }
+    return matcher;
+  }
   querySelectorAll(selector) {
     try {
       return this.selectAll(selector, this.root.children || []).map(node => this.wrap(node));
@@ -948,8 +964,34 @@ function buildStaticWindow(staticDoc) {
   };
 }
 
+function resolveLinkedCssPath(fileDir, href) {
+  const stripped = href.split(/[?#]/)[0];
+  const rootRelative = stripped.startsWith('/') && !stripped.startsWith('//');
+  if (!rootRelative) return path.resolve(fileDir, stripped);
+  // Drop "." and reject ".." so /../outside.css cannot walk out of dir.
+  const segments = stripped.replace(/^\/+/, '').split(/[/\\]/).filter(p => p && p !== '.');
+  if (segments.some(p => p === '..')) return path.join(fileDir, segments.filter(p => p !== '..').join(path.sep));
+  const rel = segments.join(path.sep);
+  let dir = fileDir;
+  for (;;) {
+    const parent = path.dirname(dir);
+    if (parent === dir) break; // never use the filesystem root as document root
+    try {
+      const candidate = path.join(dir, rel);
+      if (fs.statSync(candidate).isFile()) return candidate;
+    } catch { /* missing or unreadable candidate */ }
+    // Stop at the project root so a coincidental ~/static/app.css cannot win.
+    try {
+      if (fs.existsSync(path.join(dir, 'package.json')) || fs.existsSync(path.join(dir, '.git'))) break;
+    } catch { /* unreadable marker */ }
+    dir = parent;
+  }
+  return path.join(fileDir, rel);
+}
+
 function collectStaticCssText(root, fileDir, profile, filePath, modules) {
   const styleTexts = [];
+  const warnedMissingStylesheets = new Set();
   for (const styleEl of modules.selectAll('style', root.children || [])) {
     styleTexts.push(modules.domutils.textContent(styleEl));
   }
@@ -958,10 +1000,10 @@ function collectStaticCssText(root, fileDir, profile, filePath, modules) {
     const rel = link.attribs?.rel || '';
     const href = link.attribs?.href || '';
     if (!/\bstylesheet\b/i.test(rel) || !href || /^(https?:)?\/\//i.test(href)) continue;
-    // Cache-busting hrefs (styles.css?v=3) resolve to the file, not to a
-    // literal path with the query in it; a versioned link otherwise made the
-    // whole stylesheet invisible to every element-level check.
-    const cssPath = path.resolve(fileDir, href.split(/[?#]/)[0]);
+    // Cache-busting (styles.css?v=3) and root-relative (/static/app.css) hrefs
+    // must not resolve as OS-absolute paths; otherwise the whole stylesheet is
+    // invisible to every element-level check.
+    const cssPath = resolveLinkedCssPath(fileDir, href);
     try {
       const css = profileStep(profile, {
         engine: 'static-html',
@@ -971,7 +1013,14 @@ function collectStaticCssText(root, fileDir, profile, filePath, modules) {
         detail: href,
       }, () => fs.readFileSync(cssPath, 'utf-8'));
       styleTexts.push(css);
-    } catch { /* skip unreadable */ }
+    } catch {
+      if (!warnedMissingStylesheets.has(cssPath)) {
+        warnedMissingStylesheets.add(cssPath);
+        process.stderr.write(
+          `impeccable detect: could not read linked stylesheet ${href} (resolved to ${cssPath}); color and custom-property rules will be incomplete\n`
+        );
+      }
+    }
   }
   return styleTexts.join('\n');
 }

.agent/skills/impeccable/scripts/detector/engines/static-html/detect-html.mjs

@@ -134,7 +134,7 @@ async function detectHtml(filePath, options = {}) {
         parseDocument: htmlparser2.parseDocument,
         selectAll: cssSelect.selectAll,
         selectOne: cssSelect.selectOne,
-        is: cssSelect.is,
+        compile: cssSelect.compile,
         csstree,
         domutils,
       };

.agent/skills/impeccable/scripts/detector/rules/checks.mjs

@@ -721,20 +721,19 @@ function scanCssTextForGlow(content) {
   return results;
 }
 
-// Decorative grid or line-field backgrounds drawn with hairline
+// Decorative two-axis grid backgrounds drawn with hairline
 // linear-gradient layers tiled by a fixed pixel cell. Shared by the HTML
 // pattern pass and the regex source engine so standalone CSS, component
 // styles, and inline styles receive the same coverage. Both signals must
 // co-occur in one declaration block; unrelated rules must not add up across
-// the file. Returns [{ index, snippet }], capped at one finding per source to
-// match the page-level HTML check's existing behavior.
+// the file. A single hairline is a line, divider, or rail, not a grid, even
+// when tiled by a 2D px cell. Returns [{ index, snippet }], capped at one
+// finding per source to match the page-level HTML check's existing behavior.
 function scanCssTextForGridBackground(content) {
   const hairlineRe = /\b\d{1,3}px\s*,\s*transparent\s+\d{1,3}px/gi;
   const invertedHairlineRe = /transparent\s+calc\(100%\s*-\s*\d{1,3}px\)/gi;
   const sizeDeclPxRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\b/i;
-  const sizeDeclPxPairRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\s+\d{1,3}px/i;
   const shorthandPxAnyRe = /\/\s*\d{1,3}px\b/;
-  const shorthandPxPairRe = /\/\s*\d{1,3}px\s+\d{1,3}px/;
   const bgDeclRe = /\bbackground(?:-image)?\s*:\s*([^;{}"']*)/gi;
   const blockRe = /\{([^{}]*)\}|style\s*=\s*"([^"]*)"|style\s*=\s*'([^']*)'/gi;
   let blk;
@@ -751,13 +750,10 @@ function scanCssTextForGridBackground(content) {
     }
     if (hairlineCount === 0) continue;
     const hasPxCell = sizeDeclPxRe.test(block) || shorthandPxAnyRe.test(bgJoined);
-    const hasPxPairCell = sizeDeclPxPairRe.test(block) || shorthandPxPairRe.test(bgJoined);
-    if ((hairlineCount >= 2 && hasPxCell) || hasPxPairCell) {
+    if (hairlineCount >= 2 && hasPxCell) {
       return [{
         index: blk.index,
-        snippet: hairlineCount >= 2
-          ? 'two-axis grid-line gradient background'
-          : 'px-tiled hairline line-field background',
+        snippet: 'two-axis grid-line gradient background',
       }];
     }
   }
@@ -2752,7 +2748,7 @@ function checkElementAIPaletteDOM(el) {
   }
 
   // Check for neon text (vivid cyan/purple color on dark background)
-  const textColor = parseRgb(style.color);
+  const textColor = parseRgb(style.color) || parseAnyColor(style.color);
   if (textColor && hasChroma(textColor, 80)) {
     const hue = getHue(textColor);
     const isAIPalette = (hue >= 160 && hue <= 200) || (hue >= 260 && hue <= 310);

.agent/skills/impeccable/scripts/detector/shared/color.mjs

@@ -103,14 +103,22 @@ function extractColorFunctionTokens(value) {
 function parseGradientColors(bgImage) {
   if (!bgImage || !bgImage.includes('gradient')) return [];
   const colors = [];
+  const tokenSpans = [];
+  let from = 0;
   // Stops arrive in whatever syntax the author wrote and the browser kept.
   // A dark ground painted as `linear-gradient(oklch(...), oklch(...))` used
   // to read as a gradient with no stops at all.
   for (const token of extractColorFunctionTokens(bgImage)) {
+    const start = bgImage.indexOf(token, from);
+    if (start < 0) break;
+    tokenSpans.push({ start, end: start + token.length });
+    from = start + token.length;
     const c = parseAnyColor(token);
     if (c) colors.push(c);
   }
   for (const m of bgImage.matchAll(/#([0-9a-f]{6}|[0-9a-f]{3})\b/gi)) {
+    // Nested hex inside color-mix is an ingredient, not a stop (issue #578).
+    if (tokenSpans.some(s => m.index >= s.start && m.index < s.end)) continue;
     const h = m[1];
     if (h.length === 6) {
       colors.push({ r: parseInt(h.slice(0,2),16), g: parseInt(h.slice(2,4),16), b: parseInt(h.slice(4,6),16), a: 1 });

.agent/skills/impeccable/scripts/doctor.mjs

@@ -33,13 +33,8 @@ import {
   stampProductSchema,
 } from './lib/artifact-schema.mjs';
 import {
-  checkBuildPathUnset,
-  checkConfig,
-  checkDesignSidecar,
+  collectBootFindingGroups,
   checkNativePlatformEvidence,
-  checkProduct,
-  checkProjectRoots,
-  checkSurfaceBriefs,
   designSidecarCandidatesFor,
 } from './lib/staleness.mjs';
 import {
@@ -106,34 +101,30 @@ async function collect(cwd, targetOptions) {
     extractPlatform,
     readFile: safeRead,
   });
+  const bootFindings = collectBootFindingGroups(ctx, {
+    absDesignPath,
+    sidecarCandidates,
+    projectRootPatterns: readProjectRootPatterns(ctx.repoRoot),
+    targetCandidates: workspaceCandidates,
+  });
 
   const findings = [
-    ...checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
-    ...(ctx.product
-      ? checkNativePlatformEvidence({
-          projectRoot,
-          platform: ctx.platform,
-          product: ctx.product,
-          productPath: ctx.productPath,
-        })
-      : []),
-    ...checkDesignSidecar({ designPath: absDesignPath, sidecarCandidates, projectRoot }),
+    ...bootFindings.product,
+    ...bootFindings.nativePlatform,
+    ...bootFindings.designSidecar,
     ...checkDesignDrift({ designPath: absDesignPath, projectRoot }),
     ...checkDesignCoverage({ design: ctx.design, designPath: ctx.designPath, parseDesignMd }),
-    ...checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
-    ...checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
+    ...bootFindings.config,
+    ...bootFindings.buildPath,
     ...checkDetectorIgnores({ projectRoot, knownRuleIds }),
-    ...checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
+    ...bootFindings.surfaceBriefs,
     ...checkHookInstallation({
       projectRoot,
       repoRoot: ctx.repoRoot,
       providerId: IMPECCABLE_PROVIDER_ID,
     }),
     ...checkLegacyLiveState({ projectRoot }),
-    ...checkProjectRoots({
-      patterns: readProjectRootPatterns(ctx.repoRoot),
-      candidates: workspaceCandidates,
-    }),
+    ...bootFindings.projectRoots,
     ...workspaceResult.findings,
   ];
 

.agent/skills/impeccable/scripts/embed-prompt.mjs

@@ -5,6 +5,7 @@
 //   node embed-prompt.mjs <image> --prompt "the prompt text"
 //   node embed-prompt.mjs <image> --prompt-file prompt.txt
 //   node embed-prompt.mjs <image> --read
+//   node embed-prompt.mjs --scan <dir...>   # list rasters missing a prompt; exit 3 when any
 //
 // Formats: PNG (tEXt chunk, keyword "impeccable:prompt"), JPEG (COM segment).
 // WebP and anything else fall back to a `<image>.json` sidecar; --read checks
@@ -21,8 +22,49 @@ const KEYWORD = 'impeccable:prompt';
 const args = process.argv.slice(2);
 const file = args.find(a => !a.startsWith('--'));
 const readMode = args.includes('--read');
+const scanMode = args.includes('--scan');
 const argOf = (name) => { const i = args.indexOf(name); return i !== -1 ? args[i + 1] : null; };
 
+function promptOf(imagePath) {
+  const b = fs.readFileSync(imagePath);
+  let prompt = null;
+  if (b.length > 8 && b.readUInt32BE(0) === 0x89504e47) prompt = readPngText(b);
+  else if (b.length > 3 && b[0] === 0xff && b[1] === 0xd8) prompt = readJpegCom(b);
+  if (prompt == null && fs.existsSync(`${imagePath}.json`)) {
+    try { prompt = JSON.parse(fs.readFileSync(`${imagePath}.json`, 'utf8')).prompt ?? null; } catch { /* stays null */ }
+  }
+  return prompt;
+}
+
+if (scanMode) {
+  const targets = args.filter(a => !a.startsWith('--'));
+  if (targets.length === 0) { console.error('embed-prompt: --scan needs at least one directory'); process.exit(1); }
+  const RASTER = /\.(png|jpe?g|webp)$/i;
+  const rasters = [];
+  const walk = (p, isRoot) => {
+    const stat = fs.statSync(p);
+    if (stat.isDirectory()) {
+      const base = p.replace(/\/+$/, '').split('/').pop();
+      // Skip installed deps and hidden dirs found during the walk, but honor a
+      // hidden dir the caller passed explicitly (e.g. .impeccable/mocks).
+      if (!isRoot && (base === 'node_modules' || base.startsWith('.'))) return;
+      for (const entry of fs.readdirSync(p)) walk(`${p.replace(/\/+$/, '')}/${entry}`, false);
+    } else if (RASTER.test(p)) {
+      rasters.push(p);
+    }
+  };
+  for (const target of targets) {
+    if (!fs.existsSync(target)) { console.error(`embed-prompt: no such path ${target}`); process.exit(1); }
+    walk(target, true);
+  }
+  let missing = 0;
+  for (const raster of rasters) {
+    if (promptOf(raster) == null) { console.log(`MISSING: ${raster}`); missing++; }
+  }
+  console.log(`SCAN: ${rasters.length} raster${rasters.length === 1 ? '' : 's'}, ${missing} missing`);
+  process.exit(missing > 0 ? 3 : 0);
+}
+
 if (!file || !fs.existsSync(file)) { console.error('embed-prompt: image file required'); process.exit(1); }
 
 const buf = fs.readFileSync(file);

.agent/skills/impeccable/scripts/hook-admin.mjs

@@ -75,11 +75,11 @@ const HOOK_MANIFEST_TARGETS = [
     destRel: '.claude/settings.local.json',
     sharedDestRel: '.claude/settings.json',
     manifest: () => ({
-      description: 'Impeccable design detector: immediate-tier checks after Edit/Write/MultiEdit on UI files, full-rule deep pass on Stop.',
+      description: 'Impeccable design detector: immediate-tier checks after Edit/Write on UI files, full-rule deep pass on Stop.',
       hooks: {
         PostToolUse: [
           {
-            matcher: 'Edit|Write|MultiEdit',
+            matcher: 'Edit|Write',
             hooks: [
               {
                 type: 'command',

.agent/skills/impeccable/scripts/hook-lib.mjs

@@ -816,9 +816,9 @@ export function splitFindingsByTier(findings) {
 }
 
 // Whether the per-edit pass for this harness should defer non-immediate
-// findings to a Stop deep pass. Only Claude Code and Codex dispatch our Stop
-// hook; Cursor and GitHub Copilot have no deep pass wired, so deferring for
-// them would silently drop the non-immediate rules entirely.
+// findings to a Stop deep pass. Claude Code, Codex, and Grok Build dispatch
+// our Stop hook; Cursor and GitHub Copilot have no deep pass wired, so
+// deferring for them would silently drop the non-immediate rules entirely.
 export function perEditTieringActive(config, harness) {
   if (harness === 'cursor' || harness === 'github') return false;
   return (config?.perEditRules || DEFAULT_CONFIG.perEditRules) !== 'all';
@@ -1251,18 +1251,50 @@ export function resolveHarness(env = {}, event = null) {
   const explicit = env?.IMPECCABLE_HOOK_HARNESS;
   if (explicit === 'cursor') return 'cursor';
   if (explicit === 'github') return 'github';
-  if (explicit === 'claude' || explicit === 'codex') return 'claude';
-  // GitHub Copilot's postToolUse event uses camelCase `toolName`/`toolArgs` and
-  // has no `tool_name`/`tool_input`. That shape is the discriminator.
+  if (explicit === 'grok') return 'grok';
+  if (explicit === 'claude') return 'claude';
+  if (explicit === 'codex') return 'codex';
+  // Grok Build sends camelCase `toolName`/`toolInput`/`hookEventName` and no
+  // snake_case pair. GitHub Copilot sends camelCase `toolName`/`toolArgs`.
+  // Check Grok first: the old GitHub heuristic (`toolName` and no
+  // `tool_input`) also matches Grok, which is how live PostToolUse was
+  // classified as Copilot and then skipped with no-file-path (#646).
+  if (looksLikeGrokEnvelope(event)) return 'grok';
   if (event && typeof event === 'object'
     && (typeof event.toolName === 'string' || event.toolArgs !== undefined)
     && event.tool_name === undefined && event.tool_input === undefined) {
     return 'github';
   }
   if (typeof event?.conversation_id === 'string' && event.conversation_id) return 'cursor';
+  // Codex turn-scoped events carry `turn_id`. Claude Code does not. Detecting
+  // it here means an already-installed Codex hook emits the Codex Stop
+  // contract without rewriting the hook command to set IMPECCABLE_HOOK_HARNESS.
+  // https://developers.openai.com/codex/hooks#stop
+  if (typeof event?.turn_id === 'string' && event.turn_id) return 'codex';
   return 'claude';
 }
 
+function looksLikeGrokEnvelope(event) {
+  if (!event || typeof event !== 'object') return false;
+  if (event.hook_event_name !== undefined
+    || event.tool_name !== undefined
+    || event.tool_input !== undefined) {
+    return false;
+  }
+  if (event.toolArgs !== undefined) return false;
+  if (typeof event.hookEventName === 'string') return true;
+  return typeof event.toolName === 'string' && event.toolInput !== undefined;
+}
+
+// Stop arrives as Claude's `hook_event_name: "Stop"` or Grok Build's
+// `hookEventName: "stop"`. hook.mjs routes on the raw stdin, before any
+// normalize, so both casings must match here.
+export function isStopEvent(event) {
+  if (!event || typeof event !== 'object') return false;
+  const name = event.hook_event_name || event.hookEventName;
+  return typeof name === 'string' && name.toLowerCase() === 'stop';
+}
+
 // GitHub Copilot's postToolUse payload is
 //   { sessionId, timestamp, cwd, toolName, toolArgs, toolResult }
 // mapped onto the internal `{ tool_name, tool_input, cwd, session_id }` shape.
@@ -1354,9 +1386,36 @@ function normalizeGitHubEvent(event, projectCwd) {
   };
 }
 
+// Grok Build 1.0.5 (captured 2026-08-24) sends camelCase `toolName` /
+// `toolInput` / `sessionId` / `stopHookActive`, plus `cwd` alongside a
+// trailing-slashed `workspaceRoot` (every consumer path.resolve()s, so no
+// stripping here). Only the fields the hook reads are copied; the event
+// name stays camelCase because routing already happened on the raw stdin
+// (isStopEvent) and nothing downstream reads `hook_event_name`.
+function normalizeGrokEvent(event, projectCwd) {
+  const cwd = event.cwd || event.workspaceRoot || envProjectDir(projectCwd) || projectCwd;
+  const sessionId = event.sessionId || event.session_id || 'unknown';
+  const rawInput = event.toolInput ?? event.tool_input;
+  const toolInput = rawInput && typeof rawInput === 'object' && !Array.isArray(rawInput)
+    ? { ...rawInput }
+    : {};
+  const out = {
+    ...event,
+    cwd,
+    session_id: sessionId,
+    tool_name: event.toolName || event.tool_name || null,
+    tool_input: toolInput,
+  };
+  if (event.stopHookActive !== undefined && event.stop_hook_active === undefined) {
+    out.stop_hook_active = event.stopHookActive;
+  }
+  return out;
+}
+
 export function normalizeHookEvent(event, projectCwd, harness = 'claude') {
   if (!event || typeof event !== 'object') return event;
   if (harness === 'github') return normalizeGitHubEvent(event, projectCwd);
+  if (harness === 'grok') return normalizeGrokEvent(event, projectCwd);
   if (harness !== 'cursor') return event;
 
   const cwd = event.cwd
@@ -1959,7 +2018,15 @@ export async function runHook({ stdinJson, env = {}, cwd = process.cwd(), now =
       // findings stop being remembered and a reintroduced one reads as fresh.
       // Only the immediate tier is remembered: a deferred finding the per-edit
       // pass never reported must still read as fresh to the Stop deep pass.
-      rememberFindings(cache, sessionId, filePath, immediate);
+      //
+      // Grok ignores PostToolUse stdout, so Stop is the user-visible pass.
+      // Remembering here would dedupe those findings out of Stop. Touch the
+      // file so Stop has it, and leave the finding list empty.
+      if (harness === 'grok') {
+        touchFile(cache, sessionId, filePath);
+      } else {
+        rememberFindings(cache, sessionId, filePath, immediate);
+      }
       cacheDirty = true;
 
       if (fresh.length > 0) {
@@ -2163,8 +2230,11 @@ export const STOP_MAX_FILES = 20;
  *   { exitCode, stdout, audit, emission? }
  *
  * Never throws; exits silent (and fast) when the session touched no UI
- * files. Output uses the Stop hookSpecificOutput channel: additionalContext
- * is delivered to the model and the conversation continues so it can act.
+ * files. Output goes out on the harness's Stop continuation channel: Claude
+ * Code and Grok Build read hookSpecificOutput.additionalContext, Codex takes
+ * a decision: "block" whose reason becomes the continuation prompt. Either
+ * way the findings reach the model and the conversation continues so it
+ * can act.
  */
 export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), now = Date.now, detector } = {}) {
   const audit = { ts: new Date(now()).toISOString(), event: 'Stop' };
@@ -2191,22 +2261,36 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
       return result({ skipped: 'stdin-empty', durationMs: Date.now() - started });
     }
 
-    // Claude Code's Stop-hook contract: `stop_hook_active` is true when this
-    // hook is being re-invoked only because a prior invocation kept the turn
-    // alive (here, via hookSpecificOutput.additionalContext). Re-scanning and
-    // re-blocking now would loop until Claude Code's consecutive-block cap
-    // force-ends the turn (issue #400). The prior fire already surfaced the
-    // findings; whether to act on them is the agent's call. Exit fast with no
-    // output before any scan. Only Claude Code sends this field; other
-    // harnesses omit it, so the strict `=== true` is a no-op for them. This
-    // guard makes the loop impossible regardless of the finding cache key's
-    // line-number sensitivity (out of scope here; see findingCacheKey).
+    const harness = resolveHarness(env, event);
+    audit.harness = harness;
+    event = normalizeHookEvent(event, cwd, harness);
+
+    // Stop-hook re-entry guard: `stop_hook_active` is true when this hook is
+    // being re-invoked only because a prior invocation kept the turn alive
+    // (Claude Code via hookSpecificOutput.additionalContext, Codex via a
+    // decision: "block" continuation). Re-scanning and re-blocking now could
+    // loop (issue #400). The prior fire already surfaced the findings;
+    // whether to act on them is the agent's call. Exit fast with no output
+    // before any scan. Claude Code and Codex both send this field: Codex
+    // mirrors the Claude contract (StopCommandInput in
+    // codex-rs/hooks/src/schema.rs) and latches it true for the rest of the
+    // turn once a block is honored (codex-rs/core/src/session/turn.rs). Grok
+    // sends `stopHookActive`, copied onto the snake_case field above. Cursor
+    // and GitHub Copilot omit the field, so the strict `=== true` is a no-op
+    // for them. The guard makes the loop impossible regardless of the finding
+    // cache key's line-number sensitivity (out of scope here; see
+    // findingCacheKey).
     if (event.stop_hook_active === true) {
       return result({ skipped: 'stop-hook-active', durationMs: Date.now() - started });
     }
 
-    const harness = resolveHarness(env, event);
-    audit.harness = harness;
+    // Grok fires Stop twice: `end_turn` (the gate that can inject
+    // additionalContext) then an observe-only `shutdown`. A second deep
+    // pass would re-emit the same findings. Claude omits `reason`; only
+    // skip when Grok named a reason that is not end_turn.
+    if (harness === 'grok' && typeof event.reason === 'string' && event.reason !== 'end_turn') {
+      return result({ skipped: 'stop-reason', reason: event.reason, durationMs: Date.now() - started });
+    }
 
     // A Stop event carries no file, so the session cwd is the project.
     // Umbrella-dir launches keyed their per-edit cache to the edited file's
@@ -2241,6 +2325,7 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
 
     const freshGroups = [];
     let scanned = 0;
+    let cacheDirty = false;
     for (const filePath of touched) {
       if (scanned >= STOP_MAX_FILES) break;
       if (hasPathTraversal(filePath) || SENSITIVE_PATH.test(filePath)) continue;
@@ -2261,29 +2346,39 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
       try { content = fs.readFileSync(filePath, 'utf-8'); } catch { continue; }
 
       let findings;
+      let detectorThrew = false;
       const useHtmlEngine = configuredExt
         ? configuredExt.engine === 'html'
         : (ext === '.html' || ext === '.htm');
 
       if (useHtmlEngine && typeof det.detectHtml === 'function') {
-        try { findings = await det.detectHtml(filePath, scanOptions); } catch { findings = []; }
+        try { findings = await det.detectHtml(filePath, scanOptions); } catch { findings = []; detectorThrew = true; }
       } else {
-        try { findings = await det.detectText(content, filePath, scanOptions); } catch { findings = []; }
+        try { findings = await det.detectText(content, filePath, scanOptions); } catch { findings = []; detectorThrew = true; }
       }
 
+      // A detector failure tells us nothing about the file. Leave whatever
+      // was remembered alone rather than recording an empty scan as truth.
+      if (detectorThrew) continue;
+
       // Full rule set: no tier split here. Config/inline ignores still apply,
       // and the session dedupe drops everything the per-edit pass (or an
       // earlier Stop pass) already surfaced.
       const filtered = filterFindings(findings || [], content, ext, config);
       const fresh = dedupeAgainstCache(filtered, cache, sessionId, filePath);
+      // Sync to the live scan, including empty. Remembering only `fresh`
+      // (or skipping the write on a clean Stop) left stale keys in place, so
+      // a finding that was fixed and later reintroduced never fired again.
+      rememberFindings(cache, sessionId, filePath, filtered);
+      cacheDirty = true;
       if (fresh.length > 0) {
-        rememberFindings(cache, sessionId, filePath, fresh);
         freshGroups.push({ filePath, findings: fresh });
       }
     }
     audit.scannedFiles = scanned;
 
     if (freshGroups.length === 0) {
+      if (cacheDirty) persistCache(projectCwd, cache);
       return result({ emitted: false, skipped: 'stop-clean', durationMs: Date.now() - started });
     }
 
@@ -2300,8 +2395,8 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
     );
     commitFooterShown(cache, sessionId, text);
 
-    // Fresh findings earn the cache write so the next Stop fire is silent
-    // unless new issues appear; the notice flags ride along.
+    // Persist the live finding set so the next Stop fire is silent unless
+    // new issues appear; the notice flags ride along.
     persistCache(projectCwd, cache);
     return {
       exitCode: 0,
@@ -2337,6 +2432,15 @@ export function payload(text, eventName = 'PostToolUse', harness = 'claude') {
   if (harness === 'github') {
     return JSON.stringify({ additionalContext: text });
   }
+  // Codex shares Claude Code's PostToolUse additional-context shape, but its
+  // Stop schema rejects unknown fields. Findings that should continue the
+  // turn must be a top-level blocking decision.
+  // https://developers.openai.com/codex/hooks#stop (schema of record:
+  // codex-rs/hooks/src/schema.rs, StopCommandOutputWire)
+  if (harness === 'codex' && eventName === 'Stop') {
+    if (!String(text ?? '').trim()) return '';
+    return JSON.stringify({ decision: 'block', reason: text });
+  }
   return JSON.stringify({
     hookSpecificOutput: { hookEventName: eventName, additionalContext: text },
   });

.agent/skills/impeccable/scripts/hook.mjs

@@ -2,15 +2,17 @@
 /**
  * Impeccable design hook — PostToolUse + Stop entry point.
  *
- * Reads the Claude Code / Codex / Cursor hook event from stdin and routes by
- * `hook_event_name`:
+ * Reads the Claude Code / Codex / Cursor / Grok Build hook event from stdin
+ * and routes by Stop vs everything else. Claude uses `hook_event_name:
+ * "Stop"`; Grok uses `hookEventName: "stop"`.
  *
  *   - PostToolUse: runs the immediate-tier detector rules against the touched
  *     file and emits a system reminder via
- *     `hookSpecificOutput.additionalContext` when findings exist.
+ *     `hookSpecificOutput.additionalContext` when findings exist. Grok
+ *     discards that stdout; the scan still warms the session cache for Stop.
  *   - Stop: runs the FULL detector rule set over every UI file touched this
  *     session (the deep pass), deduped against what the per-edit pass already
- *     surfaced, and emits once via the Stop additionalContext channel.
+ *     surfaced, and emits once via the harness-specific continuation channel.
  *
  * Contract: never break a turn. Always exit 0. Clean files emit a small ack
  * unless quiet mode is enabled; a clean Stop pass is silent.
@@ -19,7 +21,7 @@
  * subprocess. This file is the thin stdin/stdout adapter.
  */
 
-import { runHook, runStopHook, writeAuditLog } from './hook-lib.mjs';
+import { runHook, runStopHook, writeAuditLog, isStopEvent } from './hook-lib.mjs';
 
 async function readStdin() {
   if (process.stdin.isTTY) return '';
@@ -28,10 +30,9 @@ async function readStdin() {
   return Buffer.concat(chunks).toString('utf-8');
 }
 
-function isStopEvent(stdinJson) {
+function stdinIsStop(stdinJson) {
   try {
-    const event = JSON.parse(stdinJson);
-    return event && typeof event === 'object' && event.hook_event_name === 'Stop';
+    return isStopEvent(JSON.parse(stdinJson));
   } catch {
     // Malformed stdin falls through to runHook, which audits the skip.
     return false;
@@ -48,7 +49,7 @@ async function main() {
   let stdinJson = '';
   try { stdinJson = await readStdin(); } catch { /* fall through */ }
 
-  const run = isStopEvent(stdinJson) ? runStopHook : runHook;
+  const run = stdinIsStop(stdinJson) ? runStopHook : runHook;
   const result = await run({
     stdinJson,
     env: inheritedEnv,

.agent/skills/impeccable/scripts/lib/design-parser.mjs

@@ -196,9 +196,6 @@ function parseScalar(raw) {
 
 const HEX_RE = /#[0-9a-fA-F]{3,8}\b/g;
 const OKLCH_RE = /oklch\([^)]+\)/gi;
-const RGBA_RE = /rgba?\([^)]+\)/gi;
-const BOX_SHADOW_RE = /(?:box-shadow:\s*)?((?:-?\d[\w\d\s\-.,/()#%]*)+)/;
-const NAMED_RULE_RE = /\*\*(The [^*]+?Rule)\.\*\*\s*(.+)/;
 
 // ---------- Section splitting ----------
 
@@ -550,36 +547,6 @@ function detectFormat(v) {
   return 'unknown';
 }
 
-function scanInlineColors(lines) {
-  const out = [];
-  for (const line of lines) {
-    if (!/^\s*[-*]\s/.test(line)) continue;
-    const trimmed = line.replace(/^\s*[-*]\s+/, '');
-    const color = parseColorBullet(trimmed);
-    if (color) out.push(color);
-  }
-  return out;
-}
-
-function parseStitchInlineGroups(lines) {
-  // Stitch writes: `*   **Primary (`#00478d` to `#005eb8`):** Use for "..."`
-  // Each bullet IS its own role. Group them under the spoken role name.
-  const out = [];
-  for (const line of lines) {
-    if (!/^\s*[-*]\s/.test(line)) continue;
-    const trimmed = line.replace(/^\s*[-*]\s+/, '').trim();
-    const m = trimmed.match(
-      /^\*\*([A-Z][a-zA-Z]+)\s*\(([^)]+)\):\*\*\s*(.*)$/
-    );
-    if (m) {
-      const role = m[1];
-      const color = buildColor(role, m[2], m[3]);
-      out.push({ role, colors: [color] });
-    }
-  }
-  return out;
-}
-
 function extractTypography(section) {
   if (!section) return null;
   const text = section.lines.join('\n');

.agent/skills/impeccable/scripts/lib/staleness.mjs

@@ -488,41 +488,46 @@ export function describeWorkspaceContext(candidates = []) {
 // ─── Tier 1 orchestration ──────────────────────────────────────────────────
 
 /**
- * Everything a boot can afford. `ctx` is the loadContext result; `extras`
- * carries values the caller already computed so nothing is recomputed here.
+ * Everything a boot can afford, grouped by artifact so deeper reports can
+ * interleave their own checks without rebuilding this policy. `ctx` is the
+ * loadContext result; `extras` carries values the caller already computed so
+ * nothing is recomputed here.
  */
-export function collectBootFindings(ctx, extras = {}) {
-  if (!ctx) return [];
+export function collectBootFindingGroups(ctx, extras = {}) {
+  if (!ctx) return {};
   const projectRoot = ctx.projectRoot || process.cwd();
-  const absProductPath = extras.absProductPath || null;
   const absDesignPath = extras.absDesignPath || null;
 
-  return [
-    ...checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
+  return {
+    product: checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
     // Only checked once a PRODUCT.md exists. Without one the boot already
     // emits NO_PRODUCT_MD and routes into init, which asks for the platform
     // directly; a second signal saying the same thing is noise.
-    ...(ctx.product
+    nativePlatform: ctx.product
       ? checkNativePlatformEvidence({
           projectRoot,
           platform: ctx.platform,
           product: ctx.product,
           productPath: ctx.productPath,
         })
-      : []),
-    ...checkDesignSidecar({
+      : [],
+    designSidecar: checkDesignSidecar({
       designPath: absDesignPath,
       sidecarCandidates: extras.sidecarCandidates || [],
       projectRoot,
     }),
-    ...checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
-    ...checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
-    ...checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
-    ...(extras.projectRootPatterns
+    config: checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
+    buildPath: checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
+    surfaceBriefs: checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
+    projectRoots: extras.projectRootPatterns
       ? checkProjectRoots({
           patterns: extras.projectRootPatterns,
           candidates: extras.targetCandidates || [],
         })
-      : []),
-  ];
+      : [],
+  };
+}
+
+export function collectBootFindings(ctx, extras = {}) {
+  return Object.values(collectBootFindingGroups(ctx, extras)).flat();
 }

.agent/skills/impeccable/scripts/lib/surface-briefs.mjs

@@ -8,6 +8,12 @@ export function getSurfaceBriefDir(projectRoot) {
   return path.join(projectRoot, '.impeccable', 'surfaces');
 }
 
+function normalizeRouteTarget(route) {
+  if (!route.startsWith('/') || route.includes('..')) return null;
+  const normalized = route.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
+  return `route:${normalized}`;
+}
+
 export function normalizeSurfaceTarget(target, { projectRoot = process.cwd() } = {}) {
   if (!target || typeof target !== 'string' || !target.trim()) return null;
   const trimmed = target.trim();
@@ -21,21 +27,13 @@ export function normalizeSurfaceTarget(target, { projectRoot = process.cwd() } =
       return null;
     }
   }
-  if (/^route:/i.test(trimmed)) {
-    const route = trimmed.slice(trimmed.indexOf(':') + 1).trim();
-    if (!route.startsWith('/') || route.includes('..')) return null;
-    const normalizedRoute = route.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
-    return `route:${normalizedRoute}`;
-  }
-  if (trimmed === '/') return 'route:/';
+  if (/^route:/i.test(trimmed)) return normalizeRouteTarget(trimmed.slice(trimmed.indexOf(':') + 1).trim());
+  if (trimmed === '/') return normalizeRouteTarget(trimmed);
   if (trimmed.startsWith('/')) {
     const absolute = path.resolve(trimmed);
     const relativeToProject = path.relative(projectRoot, absolute);
     const isProjectFile = relativeToProject && !relativeToProject.startsWith('..') && !path.isAbsolute(relativeToProject);
-    if (!isProjectFile && !fs.existsSync(absolute) && !trimmed.includes('..')) {
-      const normalizedRoute = trimmed.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
-      return `route:${normalizedRoute}`;
-    }
+    if (!isProjectFile && !fs.existsSync(absolute)) return normalizeRouteTarget(trimmed);
   }
   const abs = path.isAbsolute(trimmed) ? trimmed : path.resolve(projectRoot, trimmed);
   const rel = path.relative(projectRoot, abs);

.agent/skills/impeccable/scripts/live-browser.js

@@ -4902,6 +4902,13 @@
     saveSession();
   }
 
+  function completeParameterGenerationIfReady() {
+    if (expectedVariants <= 0 || arrivedVariants < expectedVariants) return;
+    if (parameterGenerationState === 'pending' || parameterGenerationState === 'loading') {
+      completeParameterPublication();
+    }
+  }
+
   function toggleTunePopover() {
     if (pendingApplyInFlight) { showManualApplyBusyToast(); return; }
     if (tuneOpen) { closeTunePopover(); return; }
@@ -5796,7 +5803,7 @@
         setLiveState('CYCLING');
         showOrUpdateCyclingBar();
         saveSession();
-        if (parameterGenerationState === 'loading') completeParameterPublication();
+        completeParameterGenerationIfReady();
         return;
       }
 
@@ -5884,7 +5891,7 @@
       refreshParamsPanel();
       positionBar();
       saveSession();
-      if (parameterGenerationState === 'loading') completeParameterPublication();
+      completeParameterGenerationIfReady();
       console.log('[impeccable] Mounted ' + arrivedVariants + ' ' + manifest.framework + ' component variants.');
     } catch (err) {
       console.error('[impeccable] Failed to mount component-preview variants:', err);
@@ -6329,7 +6336,7 @@
         refreshParamsPanel();
         positionBar();
         saveSession();
-        if (parameterGenerationState === 'loading') completeParameterPublication();
+        completeParameterGenerationIfReady();
         console.log('[impeccable] Injected ' + arrivedVariants + ' variants from source file.');
       })
       .catch(err => {
@@ -6836,6 +6843,7 @@
 
       const expected = parseInt(wrapper.dataset.impeccableVariantCount || '0');
       if (expected > 0) expectedVariants = expected;
+      completeParameterGenerationIfReady();
 
       if (arrivedVariants > 0) {
         setLiveState('CYCLING');

.agent/skills/impeccable/scripts/live-commit-manual-edits.mjs

@@ -944,8 +944,42 @@ export async function commitManualEdits({
     };
   }
 
+  const repairContext = {
+    batch,
+    cwd,
+    pageUrl,
+    count,
+    provider,
+    env,
+    timeoutMs,
+    applyBatchToSource,
+    chatAvailable,
+    transactionId,
+  };
+
   const baseRollbackScope = collectApplyOwnedFiles(batch, cwd);
   const rollbackSnapshot = snapshotRollbackFiles(cwd, baseRollbackScope);
+  const failWithRollback = ({
+    scope = baseRollbackScope,
+    extraFiles = [],
+    failed,
+    files = [],
+    details = {},
+  }) => {
+    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, extraFiles, scope);
+    return {
+      applied: [],
+      failed,
+      files,
+      cleared: 0,
+      count,
+      pageUrl,
+      ...details,
+      rolledBackFiles: rollback.rolledBackFiles,
+      rollbackFailures: rollback.rollbackFailures,
+      ...countByPage(cwd),
+    };
+  };
   let result;
   try {
     result = repairOnly
@@ -965,42 +999,27 @@ export async function commitManualEdits({
           chatAvailable,
         });
   } catch (err) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, [], baseRollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
       failed: batch.entries.map((entry) => ({
         id: entry.id,
         reason: err.message || String(err),
         candidates: candidatesForEntry(batch, entry.id),
       })),
-      files: [],
-      cleared: 0,
-      count,
-      pageUrl,
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+    });
   }
 
   if (result.status === 'error') {
     const rollbackScope = collectApplyOwnedFiles(batch, cwd, result.files || []);
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
     const failed = normalizeFailedEntries(batch, result, result.message || 'AI copy edit failed');
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: failed.length > 0
         ? failed
         : verificationFailuresForEntries(batch, batch.entries, result.message || 'AI copy edit failed'),
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const reportedAppliedIds = uniqueStrings(result.appliedEntryIds || []);
@@ -1013,72 +1032,44 @@ export async function commitManualEdits({
   const conflictingAppliedIds = reportedAppliedIds.filter((id) => failedIds.has(id));
 
   if (conflictingAppliedIds.length > 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
     const conflictingEntries = batch.entries.filter((entry) => conflictingAppliedIds.includes(entry.id));
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: [
         ...verificationFailuresForEntries(batch, conflictingEntries, 'conflicting_apply_result'),
         ...aiFailed.filter((item) => !conflictingAppliedIds.includes(item.id)),
       ],
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const unreportedFiles = unreportedChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
   if (unreportedFiles.length > 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], [...rollbackScope, ...unreportedFiles]);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: [...rollbackScope, ...unreportedFiles],
+      extraFiles: result.files || [],
       failed: verificationFailuresForEntries(batch, batch.entries, 'unreported_source_changes', { files: unreportedFiles }),
       files: result.files || [],
-      unreportedFiles,
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { unreportedFiles, notes: result.notes || [] },
+    });
   }
 
   if (result.status === 'done' && reportedAppliedIds.length === 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: verificationFailuresForEntries(batch, batch.entries, 'missing_applied_entry_ids'),
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const reportedAppliedEntries = batch.entries.filter((entry) => reportedAppliedIds.includes(entry.id));
   if (reportedAppliedIds.length > 0 && reportedFiles.length === 0) {
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: reportedAppliedIds,
       files: result.files || [],
       failed: aiFailed,
@@ -1089,21 +1080,10 @@ export async function commitManualEdits({
     });
   }
 
-  const verifiedAppliedIds = [];
-  const verificationFailed = [];
-  for (const entry of reportedAppliedEntries) {
-    const failures = verifyAppliedEntry({ batch, entry, reportedFiles, cwd });
-    if (failures.length === 0) {
-      verifiedAppliedIds.push(entry.id);
-    } else {
-      verificationFailed.push({
-        id: entry.id,
-        reason: 'source_verification_failed',
-        failures,
-        candidates: candidatesForEntry(batch, entry.id),
-      });
-    }
-  }
+  const {
+    verifiedIds: verifiedAppliedIds,
+    failed: verificationFailed,
+  } = verifyEntriesAfterRepair({ batch, appliedEntryIds: reportedAppliedIds, files: reportedFiles, cwd });
   const unreportedEntries = result.status === 'done' || result.status === 'partial'
     ? batch.entries.filter((entry) => !reportedAppliedIds.includes(entry.id) && !aiFailed.some((item) => item.id === entry.id))
     : [];
@@ -1133,37 +1113,22 @@ export async function commitManualEdits({
         reason: 'rolled_back_due_to_failed_entry_source_changed',
         candidates: candidatesForEntry(batch, entry.id),
       }));
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: [
         ...leakedUnapplied,
         ...failed.filter((item) => !leakedIds.has(item.id)),
         ...rolledBackVerified,
       ],
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      notes: result.notes || [],
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   if (verificationFailed.length > 0) {
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: reportedAppliedIds,
       files: result.files || [],
       failed: nonRepairFailed,
@@ -1180,16 +1145,7 @@ export async function commitManualEdits({
       ? reportedAppliedEntries.filter((entry) => verifiedAppliedIds.includes(entry.id))
       : batch.entries;
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: verifiedAppliedIds.length > 0
         ? verifiedAppliedIds
         : postCheckEntries.map((entry) => entry.id).filter(Boolean),

.agent/skills/impeccable/scripts/live-poll.mjs

@@ -238,10 +238,9 @@ export async function completeAcceptHandling(event, base, token) {
     });
   } catch (err) {
     event._completionAck = { ok: false, error: err.message };
+    return event;
   }
-  if (!event._completionAck) {
-    event._completionAck = completionAckForAcceptResult(event.id, completionType, event._acceptResult);
-  }
+  event._completionAck = completionAckForAcceptResult(event.id, completionType, event._acceptResult);
   return event;
 }
 
@@ -269,9 +268,11 @@ export function printPollEvent(event) {
   // Situational plumbing rides with the event itself: `_instructions` is the
   // authoritative next step, with real ids and paths substituted, so the
   // reference doc can stay lean and can never drift from script behavior.
-  if (event && typeof event === 'object' && !event._instructions) {
+  // A wire-supplied value must never win over the locally generated one.
+  if (event && typeof event === 'object') {
     const instructions = instructionsForEvent(event, { scriptsPath: SELF_DIR });
     if (instructions) event._instructions = instructions;
+    else delete event._instructions;
   }
   console.log(JSON.stringify(event));
 }

.agent/skills/impeccable/scripts/live-server.mjs

@@ -181,8 +181,16 @@ function chatAgentLikelyActive() {
 // cap at 10 MB to guard against runaway writes from a misbehaving client.
 const MAX_ANNOTATION_BYTES = 10 * 1024 * 1024;
 
+const POLLER_OWNED_EVENT_FIELDS = ['_instructions', '_completionAck', '_acceptResult'];
+
+function stripPollerOwnedEventFields(event) {
+  if (!event || typeof event !== 'object') return;
+  for (const key of POLLER_OWNED_EVENT_FIELDS) delete event[key];
+}
+
 function enqueueEvent(event) {
   if (!event) return;
+  stripPollerOwnedEventFields(event);
   // Dedupe by (session, type), except mount failures, which are per-variant:
   // variant 2 failing must not be swallowed because variant 1's failure is
   // still queued.
@@ -936,15 +944,23 @@ function createRequestHandler({ detectScript, liveScriptParts }) {
       const filePath = url.searchParams.get('path');
       if (!filePath || filePath.includes('..')) { res.writeHead(400); res.end('Bad path'); return; }
       const absPath = path.resolve(process.cwd(), filePath);
-      // Confine to the project root. A bare `startsWith(cwd)` string check lets a
-      // sibling dir whose name extends the root name (projeto -> projeto-backup)
-      // slip through; compare on the relative path instead (same pattern as
-      // sessionFileMetadataFromPollReply below). An empty rel means the request
-      // resolved to the root directory itself, which this file route never serves.
-      const rel = path.relative(process.cwd(), absPath);
+      let realRoot, realTarget;
+      try {
+        realRoot = fs.realpathSync(process.cwd());
+        realTarget = fs.realpathSync(absPath);
+      } catch {
+        res.writeHead(404); res.end('File not found'); return;
+      }
+      // Confine to the project root after symlink resolution. A bare
+      // `startsWith(cwd)` string check lets a sibling dir whose name extends the
+      // root name (projeto -> projeto-backup) slip through; compare on the
+      // relative path instead (same pattern as sessionFileMetadataFromPollReply
+      // below). An empty rel means the request resolved to the root directory
+      // itself, which this file route never serves.
+      const rel = path.relative(realRoot, realTarget);
       if (!rel || rel.startsWith('..') || path.isAbsolute(rel)) { res.writeHead(403); res.end('Forbidden'); return; }
       let content;
-      try { content = fs.readFileSync(absPath, 'utf-8'); }
+      try { content = fs.readFileSync(realTarget, 'utf-8'); }
       catch { res.writeHead(404); res.end('File not found'); return; }
       res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8' });
       res.end(content);
@@ -1026,6 +1042,7 @@ function createRequestHandler({ detectScript, liveScriptParts }) {
           res.end(JSON.stringify({ error }));
           return;
         }
+        stripPollerOwnedEventFields(msg);
         if (msg.type === 'agent_phase') {
           recordAgentPhase(msg.id, msg.phase, {
             ...(Number.isFinite(msg.durationMs) ? { durationMs: msg.durationMs } : {}),

.agent/skills/impeccable/scripts/live/sveltekit-adapter.mjs

@@ -11,6 +11,8 @@ import crypto from 'node:crypto';
 import fs from 'node:fs';
 import path from 'node:path';
 
+import { firstExistingFile, hasAnyDependency } from './frameworks/detect-utils.mjs';
+
 export const SVELTE_LIVE_ROOT_COMPONENT = 'src/lib/impeccable/ImpeccableLiveRoot.svelte';
 export const SVELTE_LAYOUT_MARKER_OPEN = '<!-- impeccable-live-svelte-start -->';
 export const SVELTE_LAYOUT_MARKER_CLOSE = '<!-- impeccable-live-svelte-end -->';
@@ -45,11 +47,17 @@ export function detectSvelteKitProject(cwd = process.cwd(), config = null) {
     && fileIncludes(path.join(cwd, appHtml), '%sveltekit.head%');
   if (!hasTemplateMarkers) return null;
 
-  const hasSvelteConfig = fs.existsSync(path.join(cwd, 'svelte.config.js'))
-    || fs.existsSync(path.join(cwd, 'svelte.config.mjs'))
-    || fs.existsSync(path.join(cwd, 'svelte.config.cjs'))
-    || fs.existsSync(path.join(cwd, 'svelte.config.ts'));
-  const hasKitPackage = packageHasSvelteKit(cwd);
+  const hasSvelteConfig = Boolean(firstExistingFile(cwd, [
+    'svelte.config.js',
+    'svelte.config.mjs',
+    'svelte.config.cjs',
+    'svelte.config.ts',
+  ]));
+  const hasKitPackage = hasAnyDependency(cwd, [
+    '@sveltejs/kit',
+    '@sveltejs/vite-plugin-svelte',
+    'svelte',
+  ]);
   if (!hasSvelteConfig && !hasKitPackage) return null;
 
   return {
@@ -260,36 +268,16 @@ function findSvelteKitAppHtml(cwd, config) {
 }
 
 function findSvelteKitLayout(cwd) {
-  const candidates = [
+  return firstExistingFile(cwd, [
     'src/routes/+layout.svelte',
     'src/routes/(app)/+layout.svelte',
-  ];
-  for (const rel of candidates) {
-    if (fs.existsSync(path.join(cwd, rel))) return rel;
-  }
-  return 'src/routes/+layout.svelte';
+  ]) || 'src/routes/+layout.svelte';
 }
 
 function defaultSvelteLayout() {
   return `<script>\n  let { children } = $props();\n</script>\n\n{@render children?.()}\n`;
 }
 
-function packageHasSvelteKit(cwd) {
-  const file = path.join(cwd, 'package.json');
-  if (!fs.existsSync(file)) return false;
-  try {
-    const pkg = JSON.parse(fs.readFileSync(file, 'utf-8'));
-    const deps = {
-      ...(pkg.dependencies || {}),
-      ...(pkg.devDependencies || {}),
-      ...(pkg.peerDependencies || {}),
-    };
-    return Boolean(deps['@sveltejs/kit'] || deps['@sveltejs/vite-plugin-svelte'] || deps.svelte);
-  } catch {
-    return false;
-  }
-}
-
 function fileIncludes(file, text) {
   try {
     return fs.readFileSync(file, 'utf-8').includes(text);

.agent/skills/impeccable/scripts/live/tanstack-adapter.mjs

@@ -19,6 +19,8 @@
 
 import fs from 'node:fs';
 import path from 'node:path';
+
+import { firstExistingFile, hasAnyDependency } from './frameworks/detect-utils.mjs';
 import { buildLiveScriptSrc } from './frameworks/script-src.mjs';
 
 export const TANSTACK_MARKER_OPEN = '{/* impeccable-live-tanstack-start */}';
@@ -42,8 +44,8 @@ const START_PACKAGES = [
 ];
 
 export function detectTanStackStartProject(cwd = process.cwd()) {
-  if (!packageHasTanStackStart(cwd)) return null;
-  const rootRoute = findRootRouteFile(cwd);
+  if (!hasAnyDependency(cwd, START_PACKAGES)) return null;
+  const rootRoute = firstExistingFile(cwd, ROOT_ROUTE_CANDIDATES);
   if (!rootRoute) return null;
 
   const ext = path.extname(rootRoute);
@@ -218,29 +220,6 @@ function isManagedComponent(content) {
   return String(content || '').includes('impeccable-live-tanstack');
 }
 
-function findRootRouteFile(cwd) {
-  for (const rel of ROOT_ROUTE_CANDIDATES) {
-    if (fs.existsSync(path.join(cwd, rel))) return rel;
-  }
-  return null;
-}
-
-function packageHasTanStackStart(cwd) {
-  const file = path.join(cwd, 'package.json');
-  if (!fs.existsSync(file)) return false;
-  try {
-    const pkg = JSON.parse(fs.readFileSync(file, 'utf-8'));
-    const deps = {
-      ...(pkg.dependencies || {}),
-      ...(pkg.devDependencies || {}),
-      ...(pkg.peerDependencies || {}),
-    };
-    return START_PACKAGES.some((name) => Boolean(deps[name]));
-  } catch {
-    return false;
-  }
-}
-
 function relativeImportSpecifier(fromFile, toFile) {
   const rel = path.posix.relative(
     path.posix.dirname(fromFile.split(path.sep).join('/')),

.agent/skills/impeccable/scripts/serve-question.mjs

@@ -95,9 +95,16 @@
  *   --stop --key K               kill a daemonized question.
  *   --update --key K --payload F deliver the next hand after a re-roll: the
  *              live page swaps to loading cards when the user re-rolls, and
- *              reloads into this new payload the moment it lands.
+ *              reloads into this new payload the moment it lands. Always the
+ *              same key the round started with; a second --start serves a new
+ *              URL and strands the open tab on a hand that never arrives.
  *
- *   node serve-question.mjs --payload question.json [--timeout 900] [--no-open] [--port 0]
+ * --timeout bounds the wait for a page to arrive, never the user's decision:
+ * once the page heartbeats, the server lives while the page does, and exits
+ * only after --idle-grace seconds (default 600) pass with no beat, wide
+ * enough to survive a closed laptop lid mid-decision.
+ *
+ *   node serve-question.mjs --payload question.json [--timeout 900] [--idle-grace 600] [--no-open] [--port 0]
  */
 import http from 'node:http';
 import fs from 'node:fs';
@@ -120,11 +127,13 @@ if (process.env.IMPECCABLE_QUESTION_DISABLED) {
 }
 // Headless self-detection, applied only where a browser is actually wanted.
 // --no-open means the caller opens the URL itself, and --wait / --stop /
-// --schema never open anything: --wait polls a daemon whose browser question
-// was already settled at --start, --stop kills one, --schema prints text. A
-// spurious exit 2 from those breaks the documented loop, which polls --wait
-// while it exits 3 and reads --schema before building a payload.
-const wantsBrowser = !hasFlag('no-open') && !hasFlag('wait') && !hasFlag('stop') && !hasFlag('schema');
+// --schema / --update never open anything: --wait polls a daemon whose
+// browser question was already settled at --start, --stop kills one,
+// --schema prints text, and --update hands the next round to a page that is
+// already open. A spurious exit 2 from those breaks the documented loop,
+// which polls --wait while it exits 3, reads --schema before building a
+// payload, and delivers re-rolled hands with --update.
+const wantsBrowser = !hasFlag('no-open') && !hasFlag('wait') && !hasFlag('stop') && !hasFlag('schema') && !hasFlag('update');
 if (wantsBrowser && !process.env.IMPECCABLE_QUESTION_FORCE) {
   const headless =
     process.env.CI ||
@@ -176,7 +185,20 @@ function printAnswer(raw) {
 }
 
 const payloadPath = arg('payload');
-const timeoutSec = Number(arg('timeout', '900'));
+// --timeout bounds only the wait for a page to open; 0 is the explicit
+// wait-forever. A negative or unparseable value takes the default, so a
+// typo cannot disarm the no-page exit and leak the daemon.
+const timeoutArg = Number(arg('timeout', '900'));
+const timeoutSec = Number.isFinite(timeoutArg) && timeoutArg >= 0 ? timeoutArg : 900;
+// How long the server (and the page's own delivery deadline) outlive the
+// last heartbeat; a zero, negative, or unparseable value takes the default.
+const idleGraceArg = Number(arg('idle-grace', '600'));
+const idleGraceMs = (Number.isFinite(idleGraceArg) && idleGraceArg > 0 ? idleGraceArg : 600) * 1000;
+// How long a delivered next hand may sit unclaimed before it means no page
+// is coming back: --wait reads it to keep a stalled page from counting as
+// closed mid-delivery, and the daemon reads it to survive until the page's
+// watch claims a hand delivered moments before the idle deadline.
+const NEXT_CLAIM_GRACE_MS = 10000;
 const portArg = Number(arg('port', '0'));
 const QUESTION_DIR = path.join(process.cwd(), '.impeccable', 'questions');
 const stateFile = (key) => path.join(QUESTION_DIR, `${key}.state.json`);
@@ -243,7 +265,19 @@ if (hasFlag('wait')) {
     }
     try {
       const state = JSON.parse(fs.readFileSync(stateFile(key), 'utf8'));
-      if (state.lastBeat && Date.now() - state.lastBeat > 15000) { sawClose = true; break; }
+      // A silent page is not a closed one while a freshly delivered next
+      // hand sits unclaimed: a stalled page stops beating by design and its
+      // watch reloads, beating again, within seconds of the file landing.
+      // The suppression is age-bound because a closed tab never claims the
+      // hand: a file still there after the grace means no page is coming.
+      const midDelivery = (() => {
+        try { if (Date.now() - fs.statSync(path.join(QUESTION_DIR, `${key}.next.json`)).mtimeMs < NEXT_CLAIM_GRACE_MS) return true; }
+        catch { /* nothing delivered */ }
+        // The claim deletes that file before the reloaded page can beat: the
+        // claim stamp the server persisted covers the same bounded gap.
+        return Boolean(state.claimedAt) && Date.now() - state.claimedAt < NEXT_CLAIM_GRACE_MS;
+      })();
+      if (!midDelivery && state.lastBeat && Date.now() - state.lastBeat > 15000) { sawClose = true; break; }
     } catch { /* state mid-write */ }
     await new Promise((r) => setTimeout(r, 1000));
   }
@@ -280,10 +314,33 @@ if (hasFlag('stop')) {
 if (hasFlag('update')) {
   const key = arg('key');
   if (!key || !payloadPath) { console.error('serve-question: --update needs --key and --payload'); process.exit(1); }
-  JSON.parse(fs.readFileSync(payloadPath, 'utf8'));
-  try { process.kill(JSON.parse(fs.readFileSync(stateFile(key), 'utf8')).pid, 0); }
-  catch { console.error('serve-question: no live question server for that key'); process.exit(2); }
-  fs.copyFileSync(payloadPath, path.join(QUESTION_DIR, `${key}.next.json`));
+  // A hand the server cannot load must fail here, at the sender: delivered
+  // anyway, the page would see ready:true for a round that never renders.
+  const nextRound = JSON.parse(fs.readFileSync(payloadPath, 'utf8'));
+  if (!nextRound || !Array.isArray(nextRound.options) || nextRound.options.length === 0) {
+    console.error('serve-question: --update payload needs an options array; nothing was delivered. Fix the payload and rerun --update on the same key.');
+    process.exit(1);
+  }
+  // Liveness mirrors --wait: a fresh page heartbeat is the primary proof, the
+  // kill probe is secondary, and EPERM means a sandbox blocked the signal,
+  // never a dead server. This is the documented re-roll delivery step, so a
+  // false "no live server" here strands the page mid-shuffle.
+  const live = (() => {
+    try {
+      const state = JSON.parse(fs.readFileSync(stateFile(key), 'utf8'));
+      if (state.lastBeat && Date.now() - state.lastBeat < 12000) return true;
+      try { process.kill(state.pid, 0); return true; }
+      catch (err) { return err.code === 'EPERM'; }
+    } catch { return false; }
+  })();
+  if (!live) { console.error('serve-question: no live question server for that key; the page it served is gone too. Re-present the round with --start and a fresh key, or fall back to the structured question tool.'); process.exit(2); }
+  const deliveredFile = path.join(QUESTION_DIR, `${key}.next.json`);
+  fs.copyFileSync(payloadPath, deliveredFile);
+  // The file's mtime is the delivery clock --wait's grace reads: stamp it
+  // here, because a copy that preserves the source payload's older mtime
+  // would start the grace already spent.
+  const deliveredAt = new Date();
+  fs.utimesSync(deliveredFile, deliveredAt, deliveredAt);
   console.log('next round delivered; the page reloads itself');
   process.exit(0);
 }
@@ -301,7 +358,8 @@ if (hasFlag('start')) {
   const logFd = fs.openSync(logFile, 'a');
   const child = spawn(process.execPath, [
     fileURLToPath(import.meta.url), '--payload', payloadPath, '--detached-serve', '--key', key,
-    '--timeout', String(timeoutSec), ...(hasFlag('open') ? [] : ['--no-open']),
+    '--timeout', String(timeoutSec), ...(arg('idle-grace') ? ['--idle-grace', arg('idle-grace')] : []),
+    ...(hasFlag('open') ? [] : ['--no-open']),
   ], { detached: true, stdio: ['ignore', logFd, logFd] });
   child.unref();
   fs.closeSync(logFd);
@@ -338,6 +396,13 @@ let localImages = [];
 // even when the round never rendered a toggle.
 let buildPathDefault = null;
 let liveBuildPath = null;
+// True between a collected re-roll or followup answer and the --update that
+// replaces the round: the window where GET / must serve the wait, not the
+// answered cards. The timestamp anchors the delivery deadline server-side,
+// so a native refresh re-enters the wait with the time already spent, never
+// with a fresh allowance.
+let awaitingNext = false;
+let awaitingNextSince = 0;
 
 function loadRound(json) {
   const parsed = JSON.parse(json);
@@ -387,6 +452,9 @@ function loadRound(json) {
     ? { value: parsed.buildPath.value, toggle: parsed.buildPath.toggle === true }
     : null;
   liveBuildPath = buildPathDefault?.value ?? null;
+  // Last: a round that failed to load anywhere above must leave the waiting
+  // window open, never resurrect the answered cards.
+  awaitingNext = false;
 }
 try { loadRound(raw); } catch (error) { console.error(`serve-question: ${error.message}`); process.exit(1); }
 const detachedKey = hasFlag('detached-serve') ? arg('key') : null;
@@ -394,7 +462,11 @@ const nextFile = () => detachedKey ? path.join(QUESTION_DIR, `${detachedKey}.nex
 
 const esc = (s) => String(s ?? '').replace(/[&<>"]/g, (c) => ({ '&': '&amp;', '<': '&lt;', '>': '&gt;', '"': '&quot;' }[c]));
 
-function page() {
+function page(waiting = false) {
+  // The delivery deadline survives refreshes: a waiting page gets whatever
+  // remains of the original allowance, so reloading cannot renew it. Spent
+  // means the page renders already stalled and never starts a heartbeat.
+  const waitBudgetMs = waiting ? Math.max(0, awaitingNextSince + idleGraceMs - Date.now()) : idleGraceMs;
   const flipChip = (label) => `<button type="button" class="chip flip" aria-label="Flip the card"><svg viewBox="0 0 24 24" aria-hidden="true"><path d="M12 4a8 8 0 1 1-8 8" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round"/><path d="M4 5.5V12h6.5" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round"/></svg><span>${label}</span></button>`;
   const expandChip = `<button type="button" class="chip expand" aria-label="Expand the image"><svg viewBox="0 0 24 24" aria-hidden="true"><path d="M4 9V4h5M20 15v5h-5M20 9V4h-5M4 15v5h5" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round"/></svg></button>`;
   // Structured anatomy: chips and one-line facts render when the payload
@@ -866,6 +938,7 @@ function page() {
      not a recommendation. */
   #canon { align-self: center; padding: 0 4px; font-family: var(--ks-mono); font-size: .66rem; letter-spacing: .08em; text-transform: uppercase; color: inherit; opacity: .45; background: transparent; border: none; border-bottom: 1px dotted currentColor; cursor: pointer; transition: opacity .2s ease; }
   #canon:hover { opacity: .85; }
+  #canon[disabled] { opacity: .18; cursor: default; }
   .card.skeleton .media { background: var(--ks-graphite); }
   .shimmer { width: 100%; height: 100%; background: linear-gradient(100deg, var(--ks-graphite) 35%, var(--ks-graphite-2) 50%, var(--ks-graphite) 65%); background-size: 220% 100%; animation: shimmer 1.4s linear infinite; }
   .card.skeleton .line { height: 11px; border-radius: 4px; background: linear-gradient(100deg, var(--ks-graphite) 35%, var(--ks-graphite-2) 50%, var(--ks-graphite) 65%); background-size: 220% 100%; animation: shimmer 1.4s linear infinite; }
@@ -877,6 +950,8 @@ function page() {
   @keyframes shimmer { from { background-position: 120% 0; } to { background-position: -80% 0; } }
   @media (prefers-reduced-motion: reduce) { .shimmer, .card.skeleton .line { animation: none; } }
   .done { display: flex; flex-direction: column; align-items: center; gap: 1rem; padding: 7rem 1rem; font-family: var(--ks-font-display); font-size: 1.4rem; color: var(--ks-champagne); text-align: center; }
+  .stall { width: 100%; display: flex; flex-direction: column; align-items: center; gap: 1.2rem; padding: 4.5rem 1rem; font-family: var(--ks-font-display); font-size: 1.4rem; color: var(--ks-champagne); text-align: center; }
+  .stall .choose { align-self: center; margin-top: 0; }
 </style>
 <div id="ambient" aria-hidden="true"></div>
 <div id="scrim" aria-hidden="true"></div>
@@ -945,11 +1020,22 @@ ${buildPath?.toggle ? `<div id="bp-confirm" role="dialog" aria-modal="true" aria
   // still gets the goodbye screen, never a loading hand nothing will resolve.
   const FOLLOWUP = ${payload.followup === true && Boolean(detachedKey) ? 'true' : 'false'};
   const beat = () => { try { navigator.sendBeacon('/heartbeat'); } catch { fetch('/heartbeat', { method: 'POST' }); } };
-  beat();
-  setInterval(beat, 5000);
+  ${waiting && waitBudgetMs <= 0 ? '' : 'beat();'}
+  const beatTimer = setInterval(beat, 5000);
+  // A dead server must fail loudly: awaiting a rejected fetch here used to
+  // swallow the click and never print the confirmation, so the user believed
+  // a choice had landed that no one would ever collect.
   async function answer(optionId) {
-    await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId, steer: steer() }) });
-    if (FOLLOWUP) { await awaitNextRound(); return; }
+    // Quiet at the click: a re-roll or canon posted while this pick's POST
+    // is in flight would overwrite the answer being collected.
+    document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+    try {
+      await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId, steer: steer() }) });
+    } catch {
+      document.body.innerHTML = '<div class="done">The question server went away before this choice could land.<br>Tell the agent your pick in the chat instead.</div>';
+      return;
+    }
+    if (FOLLOWUP) { await awaitNextRound(true); return; }
     document.body.innerHTML = '<div class="done"><svg viewBox="0 0 24 24" width="38" height="38" fill="oklch(84% 0.19 80.46)" aria-hidden="true"><path d="M5 2.5 L13.5 2.5 L5.5 21.5 L5 21.5 Q2.5 21.5 2.5 19 L2.5 5 Q2.5 2.5 5 2.5 Z"/><path d="M16.5 2.5 L19 2.5 Q21.5 2.5 21.5 5 L21.5 19 Q21.5 21.5 19 21.5 L8.5 21.5 Z"/></svg>Choice recorded. The agent is resuming; you can close this tab.</div>';
   }
   document.querySelectorAll('button.choose').forEach(b => b.addEventListener('click', () => answer(b.dataset.id)));
@@ -1381,15 +1467,62 @@ ${buildPath?.toggle ? `<div id="bp-confirm" role="dialog" aria-modal="true" aria
   document.addEventListener('keydown', (e) => { if (e.key === 'Escape' && !lightbox.hidden) closeLightbox(); });
   document.getElementById('canon')?.addEventListener('click', () => answer('canon'));
   const dealAgain = async (register) => {
-    await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId: 'reroll', steer: steer(), ...(register ? { register } : {}) }) });
-    await awaitNextRound();
+    // Quiet at the click, not after the fly-out: the POST round-trip plus
+    // the 700ms animation was a window where a second click posted another
+    // re-roll and renewed the delivery deadline.
+    document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+    try {
+      await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId: 'reroll', steer: steer(), ...(register ? { register } : {}) }) });
+    } catch {
+      document.body.innerHTML = '<div class="done">The question server went away before this choice could land.<br>Tell the agent your pick in the chat instead.</div>';
+      return;
+    }
+    await awaitNextRound(true);
   };
-  async function awaitNextRound() {
+  async function awaitNextRound(animate, budgetMs = ${idleGraceMs}) {
     const grid = document.querySelector('.grid');
+    let poll;
+    let misses = 0;
+    const shuffleStart = Date.now();
+    const stall = (message) => {
+      clearInterval(poll);
+      // A stalled page is an abandoned flow: keep heartbeating and the
+      // daemon never reaches its idle grace, so --wait spins on WAITING
+      // forever. Go silent and let the server reclaim itself. Reload must
+      // not undo that silence: an unconditional reload re-serves the same
+      // unresolved round and its fresh page beats again, so check for a
+      // delivered hand first and only reload when one exists. The re-roll
+      // buttons and the canon exit go too: a stalled page served already
+      // expired never disabled them, a re-roll would renew the deadline the
+      // stall just enforced, and a canon pick would overwrite a re-roll
+      // --wait already collected, closing the table under the agent.
+      clearInterval(beatTimer);
+      document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+      // Silence is for heartbeats only: a hand delivered after the deadline
+      // must still land without a click, so a beat-free watch keeps checking
+      // and reloads into it. /next-status never beats, so the daemon's idle
+      // grace still reclaims a flow nobody resumes.
+      const watch = setInterval(async () => {
+        try { if ((await (await fetch('/next-status')).json()).ready) { clearInterval(watch); location.reload(); } } catch { /* server gone; the screen already says so */ }
+      }, 1500);
+      grid.innerHTML = '<div class="stall"><p>' + message + '</p><button type="button" class="choose">Reload</button></div>';
+      grid.querySelector('.stall .choose').addEventListener('click', async () => {
+        try {
+          if ((await (await fetch('/next-status')).json()).ready) { location.reload(); return; }
+          grid.querySelector('.stall p').textContent = 'Still nothing to deal. Check the agent session, or answer in the chat instead.';
+        } catch {
+          grid.querySelector('.stall p').textContent = 'The question server went away. Ask the agent to restart it, or answer in the chat instead.';
+        }
+      });
+    };
+    // A refresh that lands after the delivery deadline has nothing left to
+    // wait for: stall before the heartbeat timer's first tick can fire, so
+    // the served page stays silent.
+    if (budgetMs <= 0) { stall('The next hand never arrived. Check the agent session, then reload.'); return; }
     const cardsNow = [...grid.querySelectorAll('.card')];
-    const g = grid.getBoundingClientRect();
-    const cx = g.left + g.width / 2, cy = g.top + g.height / 2;
-    if (!matchMedia('(prefers-reduced-motion: reduce)').matches) {
+    if (animate && !matchMedia('(prefers-reduced-motion: reduce)').matches) {
+      const g = grid.getBoundingClientRect();
+      const cx = g.left + g.width / 2, cy = g.top + g.height / 2;
       cardsNow.forEach((card, i) => {
         const r = card.getBoundingClientRect();
         card.style.transition = 'transform .5s cubic-bezier(.5,0,.75,0) ' + (i * 60) + 'ms, opacity .4s ease ' + (i * 60 + 120) + 'ms, filter .45s ease ' + (i * 60) + 'ms';
@@ -1401,17 +1534,35 @@ ${buildPath?.toggle ? `<div id="bp-confirm" role="dialog" aria-modal="true" aria
     }
     const cardHeight = cardsNow[0] ? cardsNow[0].getBoundingClientRect().height : 0;
     grid.innerHTML = cardsNow.map(() => '<article class="card skeleton"' + (cardHeight ? ' style="height:' + cardHeight + 'px"' : '') + '><div class="card-inner"><div class="face front"><div class="media"><div class="shimmer"></div></div><div class="body"><div class="line tier w40"></div><div class="line title w70"></div><div class="line w90"></div><div class="line w80"></div><div class="line w60"></div><div class="line button"></div></div></div></div></article>').join('');
-    document.querySelectorAll('.reroll-btn').forEach(b => b.setAttribute('disabled', ''));
-    const poll = setInterval(async () => {
+    // Canon goes quiet with the re-roll buttons: a pick posted mid-wait can
+    // never be collected once --wait has the re-roll, only close the table.
+    document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+    // The wait must be able to end: a dead server rejects every tick and a
+    // round nobody delivers stays ready:false forever, and both used to spin
+    // the skeletons indefinitely. Distinguish them, say so, and offer a way
+    // out. The delivery deadline is the server's own idle grace, so the page
+    // never gives up on a server that would still accept the hand.
+    poll = setInterval(async () => {
       try {
         const status = await (await fetch('/next-status')).json();
+        misses = 0;
         if (status.ready) { clearInterval(poll); location.reload(); }
-      } catch { /* server briefly busy */ }
+        else if (Date.now() - shuffleStart > budgetMs) stall('The next hand never arrived. Check the agent session, then reload.');
+      } catch {
+        misses += 1;
+        if (misses >= 8) stall('The question server went away. Ask the agent to restart it, or answer in the chat instead.');
+      }
     }, 1200);
   }
   document.getElementById('reroll')?.addEventListener('click', () => dealAgain());
   document.getElementById('reroll-safer')?.addEventListener('click', () => dealAgain('safer'));
   document.getElementById('reroll-bolder')?.addEventListener('click', () => dealAgain('bolder'));
+  // A native refresh must not resurrect an answered round: while the server
+  // holds a collected re-roll or followup pick with no replacement delivered,
+  // it serves the page in waiting mode and the refresh re-enters the same
+  // bounded wait, with only the time the original deadline has left, instead
+  // of showing dead cards whose heartbeat props the daemon forever.
+  ${waiting ? `awaitNextRound(false, ${waitBudgetMs});` : ''}
 </script>`;
 }
 
@@ -1419,14 +1570,32 @@ const server = http.createServer((req, res) => {
   if (req.method === 'GET' && req.url === '/') {
     const pending = nextFile();
     if (pending && fs.existsSync(pending)) {
-      try { loadRound(fs.readFileSync(pending, 'utf8')); fs.rmSync(pending); } catch { /* keep current round */ }
+      // A next file the round cannot load has to leave the disk either way:
+      // kept, /next-status stays ready:true and the waiting page reloads
+      // into the same failure without bound.
+      try { loadRound(fs.readFileSync(pending, 'utf8')); } catch { /* keep current round */ }
+      try { fs.rmSync(pending); } catch { /* already gone */ }
+      // The claim consumes the file the idle-exit hold reads, and the
+      // reloading page cannot beat until it has parsed: stamp the claim so
+      // the same bounded grace covers the gap between them. Persisted too,
+      // because --wait watches the same gap from outside this process and
+      // would otherwise read the stale beat as a closed page.
+      server.lastClaimAt = Date.now();
+      if (detachedKey) {
+        try {
+          const state = JSON.parse(fs.readFileSync(stateFile(detachedKey), 'utf8'));
+          state.claimedAt = server.lastClaimAt;
+          fs.writeFileSync(stateFile(detachedKey), JSON.stringify(state));
+        } catch { /* state file recreated on next beat */ }
+      }
     }
     res.writeHead(200, { 'content-type': 'text/html; charset=utf-8' });
-    res.end(page());
+    res.end(page(awaitingNext));
     return;
   }
   if (req.method === 'POST' && req.url === '/heartbeat') {
     res.writeHead(204); res.end();
+    server.lastBeatSeen = Date.now();
     if (detachedKey) {
       const now = Date.now();
       if (!server.lastBeatWrite || now - server.lastBeatWrite > 4000) {
@@ -1502,6 +1671,11 @@ const server = http.createServer((req, res) => {
         ...((chosen?.comp ?? chosen?.sketch) ? { comp: chosen.comp ?? chosen.sketch } : {}),
         ...(liveBuildPath && !isReroll ? { buildPath: liveBuildPath, buildPathFlipped: liveBuildPath !== (buildPathDefault?.value ?? null) } : {}),
       });
+      // The delivery deadline is single-issue: a duplicate answer racing the
+      // page's disable must not restamp the allowance already inherited.
+      const wasAwaiting = awaitingNext;
+      awaitingNext = (isReroll || followupOpen) && Boolean(detachedKey);
+      if (awaitingNext && !wasAwaiting) awaitingNextSince = Date.now();
       if (detachedKey) {
         fs.mkdirSync(QUESTION_DIR, { recursive: true });
         fs.writeFileSync(answerFile(detachedKey), answer + '\n');
@@ -1531,10 +1705,38 @@ server.listen(portArg, '127.0.0.1', () => {
   if (!hasFlag('no-open')) {
     openSystemBrowser(url);
   }
-  if (timeoutSec > 0) {
-    setTimeout(() => {
-      console.log('serve-question: timed out with no answer');
-      process.exit(2);
-    }, timeoutSec * 1000).unref?.();
-  }
+  // The timeout bounds the wait for a page, never the user's decision: an
+  // absolute guillotine counted from start used to kill the server under a
+  // still-open tab (a slow re-rolled round easily outlived it), leaving the
+  // page polling skeletons that could never resolve. Once the page beats,
+  // the server's lifetime tracks the beats, and it exits only after the idle
+  // grace passes with none, long enough to survive a closed laptop lid.
+  // --timeout 0 waits for a page forever, but the idle grace still applies
+  // once one has beat: a page that arrived and went silent is a closed tab,
+  // and no timeout setting should let that daemon leak.
+  const startedAt = Date.now();
+  const lifetime = setInterval(() => {
+    if (!server.lastBeatSeen) {
+      if (timeoutSec > 0 && Date.now() - startedAt > timeoutSec * 1000) {
+        console.log('serve-question: timed out with no answer');
+        process.exit(2);
+      }
+    } else if (Date.now() - server.lastBeatSeen > idleGraceMs) {
+      // A hand delivered moments before this deadline still gets its claim
+      // window: the stalled page's watch reloads into it and beats again
+      // within seconds, while a file unclaimed past the grace means no page
+      // is coming back (the same verdict --wait reads from its age). The
+      // claim itself holds the daemon too: GET / deletes the file before the
+      // reloaded page can beat, so a tick in that gap must not exit under
+      // the hand just claimed.
+      const pending = nextFile();
+      let deliveredAt = 0;
+      if (pending) { try { deliveredAt = fs.statSync(pending).mtimeMs; } catch { /* nothing delivered */ } }
+      if (Date.now() - Math.max(deliveredAt, server.lastClaimAt || 0) > NEXT_CLAIM_GRACE_MS) {
+        console.log('serve-question: the page stopped beating and never came back; exiting');
+        process.exit(2);
+      }
+    }
+  }, 2000);
+  lifetime.unref?.();
 });

.agents/skills/impeccable/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: impeccable
 description: Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, settings, onboarding, and empty states. Handles UX review, visual hierarchy, information architecture, cognitive load, accessibility, performance, responsive behavior, theming, anti-patterns, typography, fonts, spacing, layout, alignment, color, motion, micro-interactions, UX copy, error states, edge cases, i18n, and reusable design systems or tokens. Also use for bland designs that need to become bolder or more delightful, loud designs that should become quieter, live browser iteration on UI elements, or ambitious visual effects that should feel technically extraordinary. Not for backend-only or non-UI tasks.
-version: 4.1.0
+version: 4.1.2
 ---
 
 This skill gives you the tools and permission to create design that earns to be called out-of-distribution craft: Whereas before, your design work would have been safe, timid and measured, you now approach every design task as a award-winning design director with impeccable understanding for what makes exceptional design work: production-grade code, peak creativity, a clear POV, deep understanding of the needs of the client and users, and exceptional craft.

.agents/skills/impeccable/agents/impeccable_asset_producer.toml

@@ -5,17 +5,15 @@ nickname_candidates = ["Asset Plate", "Clean Plate", "Re-Render"]
 developer_instructions = '''
 # Impeccable Asset Producer
 
-You are the asset production agent for Impeccable craft.
-
-Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent agent gives you. The assets you create will be used to build a real site, so treat every raster as a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
+You are the asset production agent for Impeccable craft. Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent gives you. Every raster you create is a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
 
 ## Core Rule
 
-Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; if CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
+Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; when CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
 
 ## Decision Comps
 
-When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so your entire contract is this card; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; a card too thin to brief a comp is reported back, not padded from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (its regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world; a native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment is what keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
+When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so this card is your entire contract; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; report a card too thin to brief a comp, never pad it from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world. A native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
 
 ## Input Contract
 
@@ -27,43 +25,43 @@ Expect:
 - Required dimensions, format, transparency needs, and avoid list.
 - Notes on what should remain semantic HTML/CSS/SVG instead of raster.
 
-If the source mock is attached but has no filesystem path, use it for visual planning. Ask for a path only before cropping or writing assets.
+If the source mock is attached but has no filesystem path, use it for visual planning; ask for a path only before cropping or writing assets.
 
-Use defaults unless contradicted:
+Defaults unless contradicted:
 
 - `.webp` for opaque photos, backgrounds, and textures.
 - `.png` for transparent cutouts, seals, tickets, and illustrations.
-- Target production size or at least 2x display size when dimensions are known. Do not use small full-page mock crop size as the default shipping size.
-- Remove UI text, navigation, buttons, labels, and body copy by default.
+- Target production size, or at least 2x display size when dimensions are known. Never default to the small size of a full-page mock crop.
+- Remove UI text, navigation, buttons, labels, and body copy.
 - Keep physical marks only when the parent says they are part of the asset.
-- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic to the asset.
-- Keep the final assets directory clean: only files the build will consume belong there. Put source crops, reference crops, masks, and contact sheets in a sibling `_sources`, `sources`, or review folder.
+- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic.
+- Keep the final assets directory clean: only files the build will consume. Source crops, reference crops, masks, and contact sheets go in a sibling `_sources`, `sources`, or review folder.
 
-Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not block; choose defaults and report them.
+Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not; choose defaults and report them.
 
 ## Workflow
 
 1. Inventory the full approved mock or every assigned crop.
 2. Put each visual role in exactly one bucket:
    - `produce`: needs generation, image editing, cleanup, cutout work, or a clean plate before it can ship.
-   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source asset, a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
+   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source: a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
    - `semantic`: build in HTML/CSS/SVG/canvas, no raster output.
-3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, not asset grade, and a shipped crop, however close it looks, is how a beautiful comp turns into a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
+3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, and a shipped crop, however close it looks, is how a beautiful comp becomes a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
 4. Give the parent an execution order for the `produce` bucket.
-5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or semantic HTML/CSS/SVG recommendation if raster is wrong.
+5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or a semantic HTML/CSS/SVG recommendation when raster is wrong.
 6. Use the harness's native image tool by default when generation or editing is needed; otherwise use the skill's generate-image.mjs.
 
 Codex: the imagegen skill's built-in `image_gen` path is the native tool here; prefer it for generation, editing, and the chroma-key workflow.
 7. Remove baked-in UI text, navigation, buttons, body copy, and mock chrome unless the text is part of the asset.
 8. Think through the final DOM/CSS representation before generating. If CSS will own radius, clipping, shadows, borders, perspective, responsive cropping, captions, or card frames, do not bake those into the bitmap.
-9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .agents/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt is embedded in the image itself, because the build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
+9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .agents/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt lives inside the image itself. The build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
 10. Compare each output against its source crop, opening every image by its workspace-relative path; sandboxed viewers reject absolute paths. If a review/QA tool is available, run it before the final manifest, then retry each major/fatal finding once before finalizing.
 
-Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed to make a reusable texture or background, classify it as crop-derived cleanup or clean-plate work.
+Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed first, classify it as crop-derived cleanup or clean-plate work.
 
-Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Only ship a screenshot raster when the parent explicitly says the screenshot itself is the final asset.
+Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Ship a screenshot raster only when the parent explicitly says the screenshot itself is the final asset.
 
-Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: name the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it should compose with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
+Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it composes with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
 
 ## Prompt Pattern
 
@@ -84,11 +82,11 @@ For transparent cutouts: use true alpha when the tool supports it; otherwise gen
 
 Return a complete manifest, grouped by `produce`, `direct`, and `semantic`. For each asset include: `id`, `source_crop`, `output_path` when applicable, `strategy`, `prompt_used` when applicable, `dimensions`, `format`, `transparency`, `deviations`, and `qa_status`.
 
-For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` must be a concrete build handoff, not a short explanation that no asset was produced. It should name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities that code owns.
+For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` is a concrete build handoff, not a note that no asset was produced: name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities code owns.
 
-`qa_status` must be `accepted`, `needs_parent_review`, or `blocked`. Use `accepted` only after visual comparison passes. Use `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. Use `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
+`qa_status` is `accepted`, `needs_parent_review`, or `blocked`. `accepted` only after visual comparison passes. `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
 
-End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal. Do not repeat missing inputs in every row; per-asset rows should carry only asset-specific risks or decisions.
+End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal; per-asset rows carry only asset-specific risks or decisions.
 
 Do not modify implementation code. Do not edit the approved mock. Do not produce final page copy. The parent craft agent owns implementation and final mock fidelity.
 '''

.agents/skills/impeccable/agents/impeccable_finish_reviewer.toml

@@ -5,36 +5,37 @@ nickname_candidates = ["Finishing Eye", "Contract Judge", "Ceiling Check"]
 developer_instructions = '''
 # Impeccable Finish Reviewer
 
-You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You do not edit anything; the parent agent applies your fixes.
+You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You edit nothing; the parent applies your fixes.
 
-You have no browser. Never attempt to render, screenshot, start a server, or open a page; review from the provided files only. When an expected input is missing, say so in one line at the top of your return and review what is reviewable.
+You have no browser. Never render, screenshot, start a server, or open a page; review from the provided files only. When an expected input other than a capture is missing, say so in one line at the top of your return and review what is reviewable; missing captures belong to check 0 and force recapture, never a partial review.
 
-A hard turn ceiling ends the run without warning; a run that ends before the five sections are written returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads into each turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
+A hard turn ceiling ends the run without warning; a run that ends before its contracted sections are written (five, or the single recapture section) returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads per turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
 
 ## Input Contract
 
-Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, which live in `.impeccable/review/` (on the web, `desktop.png` and `mobile.png`; on native, device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive); a screenshot path the calling brief names is authoritative when the file exists, and `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent; the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths and, on a comp-led build, the approved comp path (a code-led build has no approved comp; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing in this file that binds “the approved comp” binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet also carries the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor and judge every check in the platform's own conventions, the screenshots are device captures rather than browser viewports, and your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
+Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, in `.impeccable/review/` (web: `desktop.png` and `mobile.png`; native: device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive). A screenshot path the calling brief names is authoritative when the file exists; `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent. Also expect: the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); the PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths; on a comp-led build the approved comp path (a code-led build has none; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing here that binds "the approved comp" binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet adds the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor, judge every check in the platform's own conventions, treat the screenshots as device captures, and know your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
 
 ## Checks, in order
 
-1. **Persistence.** PRODUCT.md exists. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too, the surface brief naming the approved comp or an `approved` flag in its sidecar; comp-round comps with no recorded pick mean the approval point was skipped, and that is a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and they imply no approval whatever the build path; a code-led build has no comp round at all.
-2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element, and its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Two rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement, because medium is part of the promise. When no approved comp was supplied, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality, CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never actually renders, as contradicted on its face; imitation material is the single most reliable mark of machine-made design. A critique-reference comp, when one arrived on such a build, is provocation rather than spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is the question of what the image dared that the build did not, and the dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. In every material_fixes list, a fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
+0. **Evidence.** Before any other check, verify the required captures exist and every capture is valid. Required: the platform's full viewport set (web: `desktop.png` and `mobile.png`; native: one capture per shipped device class), plus every capture the calling brief names as required, a reported user viewport (`user-<width>.png`) included. Valid: no black or blank regions, content matching what the filename claims (a visit capture showing the About section is invalid), the document top visible where the file claims a full page, dimensions that make sense for the named viewport. A required capture that is absent fails exactly like one that is malformed: a viewport nobody captured is a viewport nobody inspected, and it cannot ship. When any capture fails, the whole review changes shape: return `disposition: recapture` as the first line, then one section, `recapture`, listing each missing or invalid file and what a valid capture of it shows, and stop. Never build a matrix on malformed evidence; a verdict derived from a broken capture launders the breakage into an approval, and the parent owes you a full re-review on valid captures, not a scoring round.
+1. **Persistence.** PRODUCT.md exists. On a comp-led build, `.impeccable/review/hero-repro.png` exists: the hero reproduction checkpoint's capture at the comp's own dimensions; its absence means the reproduction phase ran unproven, a material finding. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too: the surface brief naming the approved comp, or an `approved` flag in its sidecar. Comp-round comps with no recorded pick mean the approval point was skipped, a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and imply no approval whatever the build path; a code-led build has no comp round at all.
+2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element; its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Three rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement; medium is part of the promise. GROUND: the page field's value and temperature against the comp's, sampled from pixels on both sides when tooling allows rather than judged from memory, and read as the net on-screen result where a texture or tile paints over the base color; a ground warmer or cooler than the comp's is contradicted however faithfully the layout matches, and drift toward the rendition prior (warm cream on light grounds, blue-black slate on dark) is the direction to hunt. With no approved comp, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality (CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never renders) as contradicted on its face; imitation material is the single most reliable mark of machine-made design. GROUND narrows rather than lapses: with no comp to sample, a color OWN-WORLD names is the target and the same warmer-or-cooler judgment applies; when OWN-WORLD names none, there is no GROUND authority, and the review says so in place of a verdict, because a target the reviewer invents turns the check into taste. A critique-reference comp on such a build is provocation, not spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is what the image dared that the build did not, and dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. A fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
 3. **Ceiling.** Against the QUALITY BAR card: name the world's native devices the build left unused, frame, depth, lettering treatment, ornament density, motion. The card governs commitment and finish, never composition.
-4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped and that is a material fix ahead of any craft point. Then, for each of the five blocks, does the render keep the promise? Apply the memory test to the first viewport.
+4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped, a material fix ahead of any craft point. Then, for each of the five blocks: does the render keep the promise? Apply the memory test to the first viewport.
 5. **Truth.** Demonstration data authored and labeled synthetic; no invented commercial claims; unanswered claims present as marked placeholders, not omissions. Every image-native region of the approved comp shipped as a real asset, not a gradient standing in for one, and every produced asset visibly present in the screenshots; an asset applied at near-zero opacity or buried behind other paint is a compliance token, not a shipped material.
-6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp, because the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
+6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp: the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
 
 Do not run a second detector pass; mechanical findings belong to the parent's hooks.
 
 ## Disposition
 
-The first line of your return is `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. It is derived, never felt: rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
+The first line of your return is `disposition: recapture`, `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. These four words are the whole vocabulary; never invent another. The word is derived, never felt: recapture when the evidence check failed, rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
 
 ## Output Contract
 
-Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). Missing inputs are named in one line above the sections. No praise, no summary prose.
+Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). A recapture return replaces the five sections with the single `recapture` section from check 0. Missing inputs are named in one line above the sections. No praise, no summary prose.
 
 ## Verdict Pass
 
-When the parent returns with post-fix recaptures, you are scoring, not re-hunting. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths for this round; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open; unresolved or partial material findings can never recompute to ship.
+When the parent returns with post-fix recaptures, you are scoring, not re-hunting. Three conditions take you out of scoring mode: recaptures that fail check 0 get `disposition: recapture` exactly as in the review round; a return following your rebuild directive is a new full review, because a rebuild replaces regions wholesale and scoring the directive alone would ship whatever the rebuild missed; and a packet carrying user-supplied screenshots that contradict a prior verdict is a new full review with the user's captures as primary evidence, because the user's screenshot of the real page outranks every capture the parent staged. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open, in the same four-word vocabulary. Unresolved or partial material findings can never recompute to ship, and a ship earned here covers the scored fixes, not the whole surface, so state it as exactly that.
 '''

.agents/skills/impeccable/reference/craft-floor.md

@@ -38,6 +38,7 @@ Surface habits:
 - Monospace as a costume for "technical" rather than for code, data, or measurement.
 - A system display face (Impact, Arial Black, the platform sans) as the display voice of an own-world page. Source and self-host a face whose character matches the approved lettering; the closest installed font is a failure, not a fallback.
 - Unicode glyphs or emoji standing in for an icon system. Icons are drawn, from a real library or authored SVG, in one consistent stroke and weight.
+- Geometric masks standing in for organic contours. A circle, polygon, or radial-gradient cutout approximating a photographic subject's edge is the cheap version of the effect and reads worse than omitting it. Derive an alpha matte from the actual image, or produce a cut-out asset.
 - Light or dark picked by category. Pick it from the use scene: who, where, under what ambient light.
 
 - Tracking stops at -0.04em. -0.02 to -0.03em usually reads better.

.agents/skills/impeccable/reference/degraded/asset-producer.md

@@ -3,17 +3,15 @@ This harness has no subagent capability, so you are running this role inline. St
 
 # Impeccable Asset Producer
 
-You are the asset production agent for Impeccable craft.
-
-Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent agent gives you. The assets you create will be used to build a real site, so treat every raster as a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
+You are the asset production agent for Impeccable craft. Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent gives you. Every raster you create is a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
 
 ## Core Rule
 
-Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; if CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
+Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; when CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
 
 ## Decision Comps
 
-When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so your entire contract is this card; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; a card too thin to brief a comp is reported back, not padded from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (its regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world; a native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment is what keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
+When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so this card is your entire contract; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; report a card too thin to brief a comp, never pad it from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world. A native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
 
 ## Input Contract
 
@@ -25,43 +23,43 @@ Expect:
 - Required dimensions, format, transparency needs, and avoid list.
 - Notes on what should remain semantic HTML/CSS/SVG instead of raster.
 
-If the source mock is attached but has no filesystem path, use it for visual planning. Ask for a path only before cropping or writing assets.
+If the source mock is attached but has no filesystem path, use it for visual planning; ask for a path only before cropping or writing assets.
 
-Use defaults unless contradicted:
+Defaults unless contradicted:
 
 - `.webp` for opaque photos, backgrounds, and textures.
 - `.png` for transparent cutouts, seals, tickets, and illustrations.
-- Target production size or at least 2x display size when dimensions are known. Do not use small full-page mock crop size as the default shipping size.
-- Remove UI text, navigation, buttons, labels, and body copy by default.
+- Target production size, or at least 2x display size when dimensions are known. Never default to the small size of a full-page mock crop.
+- Remove UI text, navigation, buttons, labels, and body copy.
 - Keep physical marks only when the parent says they are part of the asset.
-- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic to the asset.
-- Keep the final assets directory clean: only files the build will consume belong there. Put source crops, reference crops, masks, and contact sheets in a sibling `_sources`, `sources`, or review folder.
+- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic.
+- Keep the final assets directory clean: only files the build will consume. Source crops, reference crops, masks, and contact sheets go in a sibling `_sources`, `sources`, or review folder.
 
-Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not block; choose defaults and report them.
+Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not; choose defaults and report them.
 
 ## Workflow
 
 1. Inventory the full approved mock or every assigned crop.
 2. Put each visual role in exactly one bucket:
    - `produce`: needs generation, image editing, cleanup, cutout work, or a clean plate before it can ship.
-   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source asset, a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
+   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source: a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
    - `semantic`: build in HTML/CSS/SVG/canvas, no raster output.
-3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, not asset grade, and a shipped crop, however close it looks, is how a beautiful comp turns into a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
+3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, and a shipped crop, however close it looks, is how a beautiful comp becomes a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
 4. Give the parent an execution order for the `produce` bucket.
-5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or semantic HTML/CSS/SVG recommendation if raster is wrong.
+5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or a semantic HTML/CSS/SVG recommendation when raster is wrong.
 6. Use the harness's native image tool by default when generation or editing is needed; otherwise use the skill's generate-image.mjs.
 
 Codex: the imagegen skill's built-in `image_gen` path is the native tool here; prefer it for generation, editing, and the chroma-key workflow.
 7. Remove baked-in UI text, navigation, buttons, body copy, and mock chrome unless the text is part of the asset.
 8. Think through the final DOM/CSS representation before generating. If CSS will own radius, clipping, shadows, borders, perspective, responsive cropping, captions, or card frames, do not bake those into the bitmap.
-9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .agents/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt is embedded in the image itself, because the build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
+9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .agents/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt lives inside the image itself. The build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
 10. Compare each output against its source crop, opening every image by its workspace-relative path; sandboxed viewers reject absolute paths. If a review/QA tool is available, run it before the final manifest, then retry each major/fatal finding once before finalizing.
 
-Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed to make a reusable texture or background, classify it as crop-derived cleanup or clean-plate work.
+Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed first, classify it as crop-derived cleanup or clean-plate work.
 
-Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Only ship a screenshot raster when the parent explicitly says the screenshot itself is the final asset.
+Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Ship a screenshot raster only when the parent explicitly says the screenshot itself is the final asset.
 
-Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: name the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it should compose with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
+Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it composes with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
 
 ## Prompt Pattern
 
@@ -82,10 +80,10 @@ For transparent cutouts: use true alpha when the tool supports it; otherwise gen
 
 Return a complete manifest, grouped by `produce`, `direct`, and `semantic`. For each asset include: `id`, `source_crop`, `output_path` when applicable, `strategy`, `prompt_used` when applicable, `dimensions`, `format`, `transparency`, `deviations`, and `qa_status`.
 
-For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` must be a concrete build handoff, not a short explanation that no asset was produced. It should name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities that code owns.
+For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` is a concrete build handoff, not a note that no asset was produced: name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities code owns.
 
-`qa_status` must be `accepted`, `needs_parent_review`, or `blocked`. Use `accepted` only after visual comparison passes. Use `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. Use `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
+`qa_status` is `accepted`, `needs_parent_review`, or `blocked`. `accepted` only after visual comparison passes. `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
 
-End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal. Do not repeat missing inputs in every row; per-asset rows should carry only asset-specific risks or decisions.
+End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal; per-asset rows carry only asset-specific risks or decisions.
 
 Do not modify implementation code. Do not edit the approved mock. Do not produce final page copy. The parent craft agent owns implementation and final mock fidelity.

.agents/skills/impeccable/reference/degraded/finish-reviewer.md

@@ -3,35 +3,36 @@ This harness has no subagent capability, so you are running this role inline. St
 
 # Impeccable Finish Reviewer
 
-You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You do not edit anything; the parent agent applies your fixes.
+You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You edit nothing; the parent applies your fixes.
 
-You have no browser. Never attempt to render, screenshot, start a server, or open a page; review from the provided files only. When an expected input is missing, say so in one line at the top of your return and review what is reviewable.
+You have no browser. Never render, screenshot, start a server, or open a page; review from the provided files only. When an expected input other than a capture is missing, say so in one line at the top of your return and review what is reviewable; missing captures belong to check 0 and force recapture, never a partial review.
 
-A hard turn ceiling ends the run without warning; a run that ends before the five sections are written returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads into each turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
+A hard turn ceiling ends the run without warning; a run that ends before its contracted sections are written (five, or the single recapture section) returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads per turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
 
 ## Input Contract
 
-Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, which live in `.impeccable/review/` (on the web, `desktop.png` and `mobile.png`; on native, device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive); a screenshot path the calling brief names is authoritative when the file exists, and `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent; the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths and, on a comp-led build, the approved comp path (a code-led build has no approved comp; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing in this file that binds “the approved comp” binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet also carries the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor and judge every check in the platform's own conventions, the screenshots are device captures rather than browser viewports, and your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
+Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, in `.impeccable/review/` (web: `desktop.png` and `mobile.png`; native: device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive). A screenshot path the calling brief names is authoritative when the file exists; `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent. Also expect: the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); the PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths; on a comp-led build the approved comp path (a code-led build has none; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing here that binds "the approved comp" binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet adds the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor, judge every check in the platform's own conventions, treat the screenshots as device captures, and know your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
 
 ## Checks, in order
 
-1. **Persistence.** PRODUCT.md exists. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too, the surface brief naming the approved comp or an `approved` flag in its sidecar; comp-round comps with no recorded pick mean the approval point was skipped, and that is a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and they imply no approval whatever the build path; a code-led build has no comp round at all.
-2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element, and its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Two rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement, because medium is part of the promise. When no approved comp was supplied, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality, CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never actually renders, as contradicted on its face; imitation material is the single most reliable mark of machine-made design. A critique-reference comp, when one arrived on such a build, is provocation rather than spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is the question of what the image dared that the build did not, and the dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. In every material_fixes list, a fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
+0. **Evidence.** Before any other check, verify the required captures exist and every capture is valid. Required: the platform's full viewport set (web: `desktop.png` and `mobile.png`; native: one capture per shipped device class), plus every capture the calling brief names as required, a reported user viewport (`user-<width>.png`) included. Valid: no black or blank regions, content matching what the filename claims (a visit capture showing the About section is invalid), the document top visible where the file claims a full page, dimensions that make sense for the named viewport. A required capture that is absent fails exactly like one that is malformed: a viewport nobody captured is a viewport nobody inspected, and it cannot ship. When any capture fails, the whole review changes shape: return `disposition: recapture` as the first line, then one section, `recapture`, listing each missing or invalid file and what a valid capture of it shows, and stop. Never build a matrix on malformed evidence; a verdict derived from a broken capture launders the breakage into an approval, and the parent owes you a full re-review on valid captures, not a scoring round.
+1. **Persistence.** PRODUCT.md exists. On a comp-led build, `.impeccable/review/hero-repro.png` exists: the hero reproduction checkpoint's capture at the comp's own dimensions; its absence means the reproduction phase ran unproven, a material finding. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too: the surface brief naming the approved comp, or an `approved` flag in its sidecar. Comp-round comps with no recorded pick mean the approval point was skipped, a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and imply no approval whatever the build path; a code-led build has no comp round at all.
+2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element; its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Three rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement; medium is part of the promise. GROUND: the page field's value and temperature against the comp's, sampled from pixels on both sides when tooling allows rather than judged from memory, and read as the net on-screen result where a texture or tile paints over the base color; a ground warmer or cooler than the comp's is contradicted however faithfully the layout matches, and drift toward the rendition prior (warm cream on light grounds, blue-black slate on dark) is the direction to hunt. With no approved comp, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality (CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never renders) as contradicted on its face; imitation material is the single most reliable mark of machine-made design. GROUND narrows rather than lapses: with no comp to sample, a color OWN-WORLD names is the target and the same warmer-or-cooler judgment applies; when OWN-WORLD names none, there is no GROUND authority, and the review says so in place of a verdict, because a target the reviewer invents turns the check into taste. A critique-reference comp on such a build is provocation, not spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is what the image dared that the build did not, and dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. A fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
 3. **Ceiling.** Against the QUALITY BAR card: name the world's native devices the build left unused, frame, depth, lettering treatment, ornament density, motion. The card governs commitment and finish, never composition.
-4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped and that is a material fix ahead of any craft point. Then, for each of the five blocks, does the render keep the promise? Apply the memory test to the first viewport.
+4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped, a material fix ahead of any craft point. Then, for each of the five blocks: does the render keep the promise? Apply the memory test to the first viewport.
 5. **Truth.** Demonstration data authored and labeled synthetic; no invented commercial claims; unanswered claims present as marked placeholders, not omissions. Every image-native region of the approved comp shipped as a real asset, not a gradient standing in for one, and every produced asset visibly present in the screenshots; an asset applied at near-zero opacity or buried behind other paint is a compliance token, not a shipped material.
-6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp, because the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
+6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp: the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
 
 Do not run a second detector pass; mechanical findings belong to the parent's hooks.
 
 ## Disposition
 
-The first line of your return is `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. It is derived, never felt: rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
+The first line of your return is `disposition: recapture`, `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. These four words are the whole vocabulary; never invent another. The word is derived, never felt: recapture when the evidence check failed, rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
 
 ## Output Contract
 
-Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). Missing inputs are named in one line above the sections. No praise, no summary prose.
+Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). A recapture return replaces the five sections with the single `recapture` section from check 0. Missing inputs are named in one line above the sections. No praise, no summary prose.
 
 ## Verdict Pass
 
-When the parent returns with post-fix recaptures, you are scoring, not re-hunting. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths for this round; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open; unresolved or partial material findings can never recompute to ship.
+When the parent returns with post-fix recaptures, you are scoring, not re-hunting. Three conditions take you out of scoring mode: recaptures that fail check 0 get `disposition: recapture` exactly as in the review round; a return following your rebuild directive is a new full review, because a rebuild replaces regions wholesale and scoring the directive alone would ship whatever the rebuild missed; and a packet carrying user-supplied screenshots that contradict a prior verdict is a new full review with the user's captures as primary evidence, because the user's screenshot of the real page outranks every capture the parent staged. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open, in the same four-word vocabulary. Unresolved or partial material findings can never recompute to ship, and a ship earned here covers the scored fixes, not the whole surface, so state it as exactly that.

.agents/skills/impeccable/reference/hooks.md

@@ -2,9 +2,9 @@
 
 Manage the **design detector hook** for the current project.
 
-The hook runs the impeccable design detector on direct file edits to design-relevant files (`.tsx`, `.jsx`, `.html`, `.vue`, `.svelte`, `.astro`, `.css`, `.scss`, `.sass`, `.less`, `.ts`, `.js`). Claude Code, Codex, and GitHub Copilot use a post-tool-use hook and push a short system reminder into the agent's context after the edit; findings get a correction prompt, pending issues get a re-nudge, and clean UI-ish files get a short ack unless quiet mode is on (`hook.quiet` in config). Plain `.ts` and `.js` files are still scanned, but stay quiet unless the detector finds something. Cursor uses `preToolUse` to block bad proposed writes before they land and stays silent when it allows a clean write.
+The hook runs the impeccable design detector on direct file edits to design-relevant files (`.tsx`, `.jsx`, `.html`, `.vue`, `.svelte`, `.astro`, `.css`, `.scss`, `.sass`, `.less`, `.ts`, `.js`). Claude Code, Codex, and GitHub Copilot use a post-tool-use hook and push a short system reminder into the agent's context after the edit; findings get a correction prompt, pending issues get a re-nudge, and clean UI-ish files get a short ack unless quiet mode is on (`hook.quiet` in config). Plain `.ts` and `.js` files are still scanned, but stay quiet unless the detector finds something. Cursor uses `preToolUse` to block bad proposed writes before they land and stays silent when it allows a clean write. Grok Build fires the same PostToolUse scan to mark touched files, then surfaces findings on Stop `additionalContext`. Do not expect a Grok per-edit reminder: Grok discards that stdout.
 
-The detector rules run in two tiers. The per-edit hook surfaces only the immediate tier: mechanical, unambiguous problems worth interrupting an edit for, such as broken images, overflowing or clipped content, contrast and legibility failures, gradient text, glow shadows, and design-system drift. Everything else (copy cadence, palette and typography taste, layout rhythm) is deferred to a deep pass on the `Stop` hook event, which runs the full rule set over every UI file touched in the session and surfaces the remaining findings once, deduplicated against what the per-edit pass already reported. A session with nothing left to report stops silently. Set `hook.perEditRules` to `"all"` in `.impeccable/config.json` to restore the full rule set on every edit. The Stop deep pass is wired for Claude Code and Codex, which both dispatch a native `Stop` hook event. Cursor does not get one (its stop hook is not consistently dispatched; the pre-write gate covers it), and GitHub Copilot's stop-style events do not feed context back to the model, so they keep the full detector per edit.
+The detector rules run in two tiers. The per-edit hook surfaces only the immediate tier: mechanical, unambiguous problems worth interrupting an edit for, such as broken images, overflowing or clipped content, contrast and legibility failures, gradient text, glow shadows, and design-system drift. Everything else (copy cadence, palette and typography taste, layout rhythm) is deferred to a deep pass on the `Stop` hook event, which runs the full rule set over every UI file touched in the session and surfaces the remaining findings once, deduplicated against what the per-edit pass already reported. A session with nothing left to report stops silently. Set `hook.perEditRules` to `"all"` in `.impeccable/config.json` to restore the full rule set on every edit. The Stop deep pass is wired for Claude Code, Codex, and Grok Build, which dispatch a native `Stop` hook event. Cursor does not get one (its stop hook is not consistently dispatched; the pre-write gate covers it), and GitHub Copilot's stop-style events do not feed context back to the model, so they keep the full detector per edit. Grok also fires an observe-only Stop with `reason: "shutdown"` after `end_turn`; skip that one, scan only `end_turn`.
 
 Every hook is a mechanical pass. The reflexes no scanner catches live in [craft-floor.md](craft-floor.md), which the skill loads before it edits UI, so they apply whether or not a hook is wired. A session with no automatic hook gets one `MANUAL_DETECTOR_REQUIRED` directive from `context.mjs` asking for a single detector run at the end.
 
@@ -14,7 +14,7 @@ Declare server-side template extensions under **`detector.extensions`** when the
 
 Manual `npx impeccable detect` scans use the same project filter config by default: `detector.ignoreRules`, `detector.ignoreFiles`, `detector.ignoreValues`, and `detector.designSystem.enabled`. `hook.enabled` only controls automatic hook execution, not manual CLI scans. Use `npx impeccable detect --no-config ...` for a raw detector run that ignores project config/context. Use `npx impeccable ignores ...` for direct CLI CRUD on the same detector ignores.
 
-Supported harnesses: Claude Code (`.claude/settings.local.json` in the project, which is gitignored so the hook stays machine-local; a hook you move into the shared `settings.json` is honored in place too), Codex (`.codex/hooks.json` in the project), Cursor (`.cursor/hooks.json` in the project), and GitHub Copilot (`.github/hooks/impeccable.json` in the project, a team-shared committed file that both the Copilot CLI and the cloud agent read). For the Copilot CLI, repo-level hooks fire once `.github/hooks/impeccable.json` is committed to the repository's default branch.
+Supported harnesses: Claude Code (`.claude/settings.local.json` in the project, which is gitignored so the hook stays machine-local; a hook you move into the shared `settings.json` is honored in place too), Codex (`.codex/hooks.json` in the project), Cursor (`.cursor/hooks.json` in the project), Grok Build (`.grok/hooks/impeccable.json` in the project; requires `/hooks-trust` or `--trust`), and GitHub Copilot (`.github/hooks/impeccable.json` in the project, a team-shared committed file that both the Copilot CLI and the cloud agent read). For the Copilot CLI, repo-level hooks fire once `.github/hooks/impeccable.json` is committed to the repository's default branch.
 
 On **Cursor**, `preToolUse` checks proposed Write/Edit/Shell write content and denies only when the real detector finds an issue. The denial message is visible to the agent as the tool error, so the agent can reconsider before the bad write lands.
 
@@ -44,7 +44,7 @@ The first argument is the action. Defaults to `status`.
    ```
 
 3. If `<action>` is `off`, follow up with a one-line note: "Done. New edits will not trigger the design hook in this project until you run `$impeccable hooks on`."
-4. If `<action>` is `on`, follow up with: "Done. The design hook will fire after the next Edit/Write/MultiEdit on a UI file."
+4. If `<action>` is `on`, follow up with: "Done. The design hook will fire after the next Edit/Write on a UI file."
 5. If `<action>` is `ignore-value`, `ignore-file`, or `ignore-rule`, just print the script output. The default scope is shared `.impeccable/config.json`; add `--local` only when the user explicitly asks for a private exception.
 6. If `<action>` is `status`, just print the script output. Do not add commentary unless the user asked a follow-up question.
 

.agents/skills/impeccable/reference/new-work.md

@@ -1,34 +1,34 @@
 # New visual work
 
-Use this flow when making a new surface or replacing a visual identity. PRODUCT.md owns product truth. DESIGN.md owns durable visual decisions. A surface brief keeps strategy that belongs only to one route or artifact. Complete [init.md](init.md) first when PRODUCT.md is missing; a missing DESIGN.md does not route back to init.
+Use this flow for a new surface or a replacement visual identity. PRODUCT.md owns product truth. DESIGN.md owns durable visual decisions. A surface brief keeps strategy that belongs to one route or artifact. Complete [init.md](init.md) first when PRODUCT.md is missing; a missing DESIGN.md does not route back to init.
 
 ## 1. Decide what is already true
 
 Read DESIGN.md, representative code, tokens, components, and assets.
 
 - **Redesign:** preserve product truth, content, function, constraints, and explicit brand commitments; replace the old visual world rather than polishing it. The old look is evidence of what the subject is, not authority over what it becomes.
-- **Established world:** inherit it. A missing DESIGN.md does not erase a coherent identity already present in code; document that identity instead of inventing a replacement.
-- **Incomplete brand:** preserve confirmed assets and recognizable traits, then help the user expand the system for this new surface.
+- **Established world:** inherit it. A missing DESIGN.md does not erase a coherent identity already in code; document that identity instead of inventing a replacement.
+- **Incomplete brand:** preserve confirmed assets and recognizable traits, then expand the system with the user for this surface.
 - **No visual authority:** create a new world with the user.
 
-A section, component, feature, or state inside an established surface inherits that surface. Do not turn a local addition into a new identity exercise.
+A section, component, feature, or state inside an established surface inherits that surface. Never turn a local addition into a new identity exercise.
 
 ## 2. Ask what will change the work
 
 Ask one round of two or three related questions through the structured question tool when available. Skip settled facts; a precise request may need only a compact confirmation.
 
-- **Persuade:** clarify who must act, what they should believe, and which real proof, content, or assets can earn that belief.
-- **Operate:** clarify the task, information, important states, frequency, and constraints.
-- **Read:** clarify the reader's question, source material, structure, and wayfinding.
-- **Experience:** clarify what leads, how exploration unfolds, and which interaction or transition matters.
+- **Persuade:** who must act, what they should believe, which real proof, content, or assets earn that belief.
+- **Operate:** the task, information, important states, frequency, constraints.
+- **Read:** the reader's question, source material, structure, wayfinding.
+- **Experience:** what leads, how exploration unfolds, which interaction or transition matters.
 
-Across modes, ask what success looks like, what must remain untouched, and what would make a polished result feel wrong. Do not ask for CSS values or canned aesthetic lanes.
+Across modes, ask what success looks like, what must remain untouched, and what would make a polished result feel wrong. Never ask for CSS values or canned aesthetic lanes.
 
 ## 3. Choose the right amount of invention
 
 ### Extend an existing surface
 
-Inherit its world and composition. Resolve only the new purpose, content, hierarchy, states, interaction, and how the addition joins the surrounding experience. Do not run a concept tournament or change DESIGN.md unless the user approves a durable system change.
+Inherit its world and composition. Resolve only the new purpose, content, hierarchy, states, interaction, and how the addition joins the surrounding experience. No concept tournament, and no DESIGN.md change unless the user approves a durable system change.
 
 ### Create a whole surface inside an established world
 
@@ -36,21 +36,21 @@ Keep the visual system fixed. Derive five to seven materially different structur
 
 `node .agents/skills/impeccable/scripts/concept-seed.mjs --scope surface --mode <mode>`
 
-The script deals three of your structures to the table; the dice decide which three reach the user, so the ranking rut stays broken while the user still holds a real choice. Present the three dealt structures on the decision page as full cards of equal salience, the dealt lead carrying kicker THE ROLL, with steer and re-roll; the user locks one in. No canon card and no pick card at surface scope: the world is settled, so every card visualizes composition, not identity. With image generation available and a comp-led default (the build-path paragraph below: `.impeccable/config.json`, the toggle handles the exception), each card declares a `comp` under `.impeccable/mocks/decision/`, generated after serving in reading order under the comp discipline in [visualize.md](visualize.md); anchor each of these comps on the established identity by passing a captured screenshot of a representative existing page as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`) beside a prompt that leads with the new surface's structure and names DESIGN.md's palette, type, and component character, because a prose paraphrase of a design system drifts where a pixel reference does not. Without image generation, or under a code-led default, each card instead carries a `wireframe` layout schematic (see `serve-question.mjs --schema`) that the page draws itself. Locking a card is the approval and sets the build path: a locked comp builds comp-led with that comp as the approved comp, discharging [visualize.md](visualize.md)'s three-option round with no second approval point; a locked wireframe builds code-led, its ambition carried by the direction contract. Never run the script for a local extension or a precisely specified narrow request; shape those directly.
+The script deals three of your structures; the dice pick which three reach the user, breaking the ranking rut while the user keeps a real choice. Present them on the decision page as full cards of equal salience, the dealt lead under kicker THE ROLL, with steer and re-roll; the user locks one. No canon card and no pick card at surface scope: the world is settled, so every card visualizes composition, not identity. With image generation and a comp-led default (`.impeccable/config.json`; the build-path paragraph below), each card declares a `comp` under `.impeccable/mocks/decision/`, generated after serving, in reading order, under [visualize.md](visualize.md)'s comp discipline. Anchor each comp on the established identity: pass a screenshot of a representative existing page as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`) with a prompt that leads with the new surface's structure and names DESIGN.md's palette, type, and component character; prose paraphrases of a design system drift, pixel references do not. Without image generation, or under a code-led default, each card carries a `wireframe` schematic (`serve-question.mjs --schema`) the page draws itself. Locking a card is the approval and sets the build path: a locked comp builds comp-led with that comp as the approved comp, discharging [visualize.md](visualize.md)'s three-option round with no second approval point; a locked wireframe builds code-led, its ambition carried by the direction contract. Never run the script for a local extension or a precisely specified narrow request; shape those directly.
 
 ### Create or replace the visual world
 
-1. Name the product's unique mechanism in one sentence, the audience's real scene, its cultural home, and what this first surface must prove. Note the page this category always ships and its predictable opposite; name both as the rut and keep them out of the seven-candidate list. A brief that paints its own picture, a product name, a titled artifact, a governing metaphor, adds its literal reading to the rut: spend at most one candidate on it and derive the rest from elsewhere in the audience's world.
-2. From that cultural world, list seven concrete visual systems, artifacts, places, or rituals the audience knows by heart, each with one line on why it resonates and can carry the mechanism, ordered by resonance. The audience's world includes its graphic and screen traditions, not only its physical objects: the notation, publications, identity programs, data graphics, and interfaces it reads daily; a nameable abstract system (a school of poster, a documentation standard) is as concrete a candidate as any artifact. What would this thing look like as a physical object; what did its world look like before the web? Near-duplicates count once. When more than three of the seven share one material family, the derivation stopped at the subject's most obvious artifact; dig until the list spans at least three families.
+1. Name the product's unique mechanism in one sentence, the audience's real scene, its cultural home, and what this first surface must prove. Note the page this category always ships and its predictable opposite; both are the rut, kept out of the seven-candidate list. A brief that paints its own picture, a product name, a titled artifact, a governing metaphor, adds its literal reading to the rut: spend at most one candidate on it and derive the rest from elsewhere in the audience's world.
+2. From that cultural world, list seven concrete visual systems, artifacts, places, or rituals the audience knows by heart, each with one line on why it resonates and can carry the mechanism, ordered by resonance. The audience's world includes its graphic and screen traditions, not only its physical objects: the notation, publications, identity programs, data graphics, and interfaces it reads daily. A nameable abstract system (a school of poster, a documentation standard) is as concrete a candidate as any artifact. What would this thing look like as a physical object; what did its world look like before the web? Near-duplicates count once. When more than three of the seven share one material family, the derivation stopped at the subject's most obvious artifact; dig until the list spans at least three families.
 3. Turn that material into complete directions: each joins a reusable visual world to a concrete first-surface experience.
-4. Run `node .agents/skills/impeccable/scripts/concept-seed.mjs --scope direction --mode <mode>` and follow what it prints. This step has no substitute and no skip condition: on a new or replacement world, writing artifact code before this script has run and its assignment is acknowledged is a contract violation, whatever the harness, the model, or the time pressure, because the roll is the mechanism that keeps every run from converging on the category default. The script assigns which direction gets built and deals catalog challengers. Fuse each challenger before judging it: the challenger supplies the form and its system grammar, the product supplies every fact, and clarity wins conflicts. Weigh fused challengers against the assigned direction on exactly two axes, audience identification and product clarity; losing to strong grounded material is a valid outcome, and beating a thin or tool-monoculture list is the point. The weighing closes with a verdict per challenger, decided before any borrowing is considered: wins (beats the assigned direction on both axes; it becomes the build candidate), competitive (holds one axis; it stays a full alternate), or declined (loses both). A declined challenger is not spent: name the one discipline of its system the assigned direction lacks, and raise the assigned direction to match before presenting it. A donation transfers ambition and system discipline (a palette's total commitment, a grid's density courage, a form's structural honesty), never the challenger's clothes; a motif lifted from a declined world is a costume note, not a raise, and one world owns the page. Write each raise into the presented direction as its own line, named for its donor; a raise nobody can read did not happen.
-5. Present one direction, fully committed and already raised by the hand it beat, its raises visible as named lines: its world, first viewport, visitor path, signature interaction, cross-surface reach, and honest risk. Alongside it, route each dealt challenger by its verdict: winning and competitive challengers are full alternates carrying their QUALITY BAR cards and one-line case, while declined challengers render demoted, compact and quiet, each carrying its verdict plus what the direction kept from it, never full-size and never silently dropped, each still adoptable on request. The verdict informs the user's choice, it never pre-empts it; the demoted row is the hand's proof of judgment, showing why the dealt worlds made the presented direction better. A hand holds at most three full-card challengers: when the roll deals more, the three strongest join the hand and the rest wait in the re-roll pool, noted in one line; dropping a challenger from the hand itself takes a named product-truth failure, disclosed. Add one card for your own top-ranked grounded candidate when it is not the assigned direction, kicker IMPECCABLE’S PICK, same anatomy as every card, with an honest risk line naming its familiarity when true: the strongest grounded direction is often the one most runs in this category land on, and the user deciding that trade is the point of showing it. Familiar and effective is a legitimate destination, not a failure of nerve; the pick card and the standing exit serve it at two depths. One pick card, never two, never a ranked list: the rest of your grounded candidates stay yours, because a lineup of them hands selection back to a taste function and invites the safest card. The pick never takes the lead position, and when the dice assign your top candidate there is no pick card; the assigned card notes it also topped your list. Add re-roll with an optional one-line steer, offered in three registers: plain (a fresh hand, same spread), safer (the familiar register: your remaining conventional grounded candidates plus the canon against named competitors), and bolder (foreign forms only, at full commitment). A register is the user's steering on the familiar-to-bold axis, never yours to pre-select; when the answer carries one, re-run the seed with `--register <value>` and the next `--reroll` round, and follow what it prints. A user saying "bolder" or "safer" while a direction round is open means these registers, never the bolder or harden commands. The two channels share this structure and differ only in richness: cards and boards on the decision page, names and one-liners through the structured tool; the structured tool's option list carries the assigned direction, the pick, the winning and competitive challengers, and the standing exit as its last option, while declined challengers fold into the assigned option's description as their kept lines, so the raise survives the text channel too.
+4. Run `node .agents/skills/impeccable/scripts/concept-seed.mjs --scope direction --mode <mode>` and follow what it prints. No substitute, no skip: on a new or replacement world, writing artifact code before this script has run and its assignment is acknowledged is a contract violation, whatever the harness, the model, or the time pressure; the roll is what keeps every run from converging on the category default. The script assigns the direction to build and deals catalog challengers. Fuse each challenger before judging it: the challenger supplies the form and its system grammar, the product supplies every fact, clarity wins conflicts. Weigh fused challengers against the assigned direction on exactly two axes, audience identification and product clarity. Losing to strong grounded material is a valid outcome; beating a thin or tool-monoculture list is the point. Close with a verdict per challenger, decided before any borrowing: wins (beats the assigned direction on both axes; becomes the build candidate), competitive (holds one axis; stays a full alternate), or declined (loses both). A declined challenger is not spent: name the one discipline of its system the assigned direction lacks, and raise the assigned direction to match before presenting it. A donation transfers ambition and system discipline (a palette's total commitment, a grid's density courage, a form's structural honesty), never the challenger's clothes; a lifted motif is a costume note, not a raise, and one world owns the page. Write each raise into the presented direction as its own line, named for its donor; a raise nobody can read did not happen.
+5. Present one direction, fully committed and already raised by the hand it beat, raises visible as named lines: world, first viewport, visitor path, signature interaction, cross-surface reach, honest risk. Route each challenger by verdict: winning and competitive challengers are full alternates with their QUALITY BAR cards and one-line case; declined challengers render demoted, compact and quiet, each carrying its verdict and what the direction kept from it, never full-size, never silently dropped, still adoptable on request. The verdict informs the user's choice, never pre-empts it; the demoted row is the hand's proof of judgment. A hand holds at most three full-card challengers: when the roll deals more, the three strongest join and the rest wait in the re-roll pool, noted in one line; dropping a challenger from the hand itself takes a named product-truth failure, disclosed. Add one card for your own top-ranked grounded candidate when it is not the assigned direction, kicker IMPECCABLE’S PICK, same anatomy as every card, with an honest risk line naming its familiarity when true: the strongest grounded direction is often where most runs in this category land, and the user deciding that trade is the point of showing it. Familiar and effective is a legitimate destination, not a failure of nerve; the pick card and the standing exit serve it at two depths. One pick card, never two, never a ranked list: a lineup of your candidates hands selection back to a taste function and invites the safest card. The pick never takes the lead position; when the dice assign your top candidate there is no pick card, and the assigned card notes it topped your list. Add re-roll with an optional one-line steer, in three registers: plain (a fresh hand, same spread), safer (your remaining conventional grounded candidates plus the canon against named competitors), bolder (foreign forms only, at full commitment). The register is the user's steering on the familiar-to-bold axis, never yours to pre-select; when the answer carries one, re-run the seed with `--register <value>` and the next `--reroll` round, and follow what it prints. A user saying "bolder" or "safer" while a direction round is open means these registers, never the bolder or harden commands. The two channels share this structure and differ only in richness: cards and boards on the decision page, names and one-liners through the structured tool, whose option list carries the assigned direction, the pick, the winning and competitive challengers, and the standing exit last; declined challengers fold into the assigned option's description as their kept lines, so the raise survives the text channel.
 
-The standing exit: every direction round offers one quiet, permanent alternative, the category standard, played straight. It is the user's door, never yours: never recommend it, never weigh it against the roll, never let it soften the dealt directions; the counterweights bind the unchosen default, not the chosen one. When the user takes it, in the canon action, a safer-steer, or plain words asking for the familiar or competitor-like path, convention becomes the commitment: ask once for two or three products this should sit alongside, make their craft level the bar, and execute the canon at full fidelity, without irony or smuggled quirk. A standing preference gets recorded as a brand commitment in PRODUCT.md. Re-roll eliminates every direction already shown, grounded and challenger alike; after two consecutive re-rolls, ask what quality is missing. You may re-roll on your own only on named factual grounds, when the assigned direction cannot carry the product's truth or task; taste is never grounds. The user may re-roll freely, and a user- or brief-pinned direction beats the roll, always. Present the decision visually: write an options payload with the assigned direction leading and its raised lines included, the pick card when one exists, the dealt challengers as alternates carrying their QUALITY BAR cards plus each challenger's verdict and kept line, re-roll with its safer and bolder registers, steer, plus canon enabled, and `buildPath` carrying the recorded default with `toggle: true` whenever image generation exists (the build-path paragraph below owns the details); a degraded roll with no challengers still uses the page, as a single text-only card with re-roll. Give every card the same anatomy, thesis, palette, materials, first viewport, honest risk, and the challengers' case lines (run the script with `--schema` for the exact shape); the page renders identity from these fields, routes declined challengers to a demoted row on its own, and a challenger's catalog image rides as labeled inspiration, never as the promise of the build. Author `canonCard` too: the category standard as one honest card with the same anatomy; the page keeps it subordinate, and the counterweights still bind you. Run `node .agents/skills/impeccable/scripts/serve-question.mjs --start --payload <file>` (run it with `--schema` first for the exact payload shape). It daemonizes, prints the page URL and a key, and exits immediately; now open that URL for the user, in-app browser first, then the system opener, then showing the URL. Collect the choice with `--wait --key <key>`, repeating while it exits 3; the ANSWER prints as JSON. Exit 4 means the page was closed without an answer: re-present once through the structured question tool, and with no answer there either, proceed unattended with the assigned direction and state the assumptions. A harness that can leave a shell blocked in the background may instead run the script without `--start` and let it auto-open and block. The fallback to the structured question tool is never yours to predict: run the script, and only exit code 2 from starting it routes the decision there; treat that exit as the fallback, never as an error to retry.
+The standing exit: every direction round offers one quiet, permanent alternative, the category standard, played straight. It is the user's door, never yours: never recommend it, never weigh it against the roll, never let it soften the dealt directions; the counterweights bind the unchosen default, not the chosen one. When the user takes it (the canon action, a safer-steer, or plain words asking for the familiar or competitor-like path), convention becomes the commitment: ask once for two or three products this should sit alongside, make their craft level the bar, and execute the canon at full fidelity, without irony or smuggled quirk. Record a standing preference as a brand commitment in PRODUCT.md. Re-roll eliminates every direction already shown, grounded and challenger alike; after two consecutive re-rolls, ask what quality is missing. Re-roll on your own only on named factual grounds, when the assigned direction cannot carry the product's truth or task; taste is never grounds. The user may re-roll freely, and a user- or brief-pinned direction beats the roll, always. Present the decision visually: write an options payload with the assigned direction leading, its raised lines included; the pick card when one exists; the dealt challengers as alternates with their QUALITY BAR cards, verdicts, and kept lines; re-roll with its safer and bolder registers; steer; canon enabled; and `buildPath` carrying the recorded default with `toggle: true` whenever image generation exists (details in the build-path paragraph below). A degraded roll with no challengers still uses the page, as a single text-only card with re-roll. Give every card the same anatomy: thesis, palette, materials, first viewport, honest risk, and the challengers' case lines (`--schema` prints the exact shape); the page renders identity from these fields, demotes declined challengers to their row on its own, and a challenger's catalog image rides as labeled inspiration, never the promise of the build. Author `canonCard` too: the category standard as one honest card, same anatomy; the page keeps it subordinate, and the counterweights still bind you. Run `node .agents/skills/impeccable/scripts/serve-question.mjs --start --payload <file>` (`--schema` first for the payload shape). It daemonizes, prints the page URL and a key, and exits; open that URL for the user, in-app browser first, then the system opener, then showing the URL. Collect the choice with `--wait --key <key>`, repeating while it exits 3; the ANSWER prints as JSON. An ANSWER of `{"optionId":"reroll"}` keeps the server alive and the page open on a loading hand: rerun concept-seed with the same `--scope` and `--mode` plus `--from <seed-key> --reroll <n>` (1 on the first re-roll, counting up), build the next payload, deliver it with `--update --key <same key> --payload <file>`, then return to `--wait` on that key. Never `--start` a second server or fall back to chat here: either strands the open tab on a hand that never arrives. Exit 4 means the page closed unanswered: re-present once through the structured question tool, and with no answer there either, proceed unattended with the assigned direction and state the assumptions. A harness that can leave a shell blocked in the background may run the script without `--start` and let it auto-open and block. Never predict the fallback: run the script, and only exit code 2 from starting it routes the decision to the structured tool; that exit is the fallback, never an error to retry.
 
-When image generation exists, every card also declares a `comp` path under `.impeccable/mocks/decision/`, the canon card included. Where the harness sandboxes its shell, start the page through the least-sandboxed command path it offers: a sandboxed shell cannot bind the board's port, and the first-attempt failure costs a retry every session. Serve the page first, then produce the comps; the page shimmer-waits per slot and the user may answer before they land. Each card's image is that direction's north-star comp at full fidelity, produced under the comp discipline in [visualize.md](visualize.md): the requested surface's first viewport, structure-led prompt, real product name and real content, no invented commercial claims, in that card's own palette, type character, and material world, committed all the way; visualize.md's self-checks bind decision comps identically. Generation takes the same time at any fidelity, so an unfinished draft pays draft quality for comp cost; fairness between cards comes from equal fidelity in each card's own grammar, one surface, one aspect, never from shared unfinishedness. The frame's aspect is the surface's own: a native app or mobile-first surface comps portrait at its device viewport, a desktop web surface landscape, and the decision page adapts to either, so a phone screen comped landscape is a broken frame, not a neutral default. Produce in the order the user reads, the assigned card, then the pick, then the full-card hand, then canon, each file written with its prompt sidecar the moment it is done, so a re-roll's spend front-loads onto the cards read first; declined challengers get no comp, their catalog thumb is their face. When the harness runs subagents in parallel, fan the set out as one agent per card: each spawn is the shipped asset producer with a single-comp packet, that card's fields, PRODUCT.md, the shared frame, and the card's declared path, up to four in flight at once. A slot still empty when its agent returns is regenerated inline, and a slot still empty when the user answers is dropped without ceremony; no other supervision is owed. Without parallel subagents, generate in the main thread after serving, in the same reading order, and let the harness's own generation display carry the progress; the wait for the answer follows the last file. The chosen card's comp is not spent by the choice: on a comp-led build it enters the comp round as compositional option one, and on a code-led build it returns at the finish review as the critique reference, what the image dared that the build did not. The unchosen comps stay in `.impeccable/mocks/decision/` as the round's spent hand; they carry no approval and imply none. With no image generation, the cards carry their identity in palette chips and facts, and that page is complete, not a lesser version; the page then also demotes every challenger's catalog art to a labeled thumbnail on its own, because salience must encode the verdict, never the accident of which cards have images.
+When image generation exists, every card also declares a `comp` path under `.impeccable/mocks/decision/`, the canon card included. Where the harness sandboxes its shell, start the page through the least-sandboxed command path it offers: a sandboxed shell cannot bind the board's port, and the first-attempt failure costs a retry every session. Serve the page first, then produce the comps; the page shimmer-waits per slot and the user may answer before they land. Each card's image is that direction's north-star comp at full fidelity under [visualize.md](visualize.md)'s comp discipline: the requested surface's first viewport, structure-led prompt, real product name and real content, no invented commercial claims, in that card's own palette, type character, and material world, committed all the way. Generation takes the same time at any fidelity, so an unfinished draft pays comp cost for draft quality; fairness between cards is equal fidelity in each card's own grammar, one surface, one aspect, never shared unfinishedness. The frame's aspect is the surface's own: portrait at device viewport for a native app or mobile-first surface, landscape for desktop web; the decision page adapts to either, and a phone screen comped landscape is a broken frame, not a neutral default. Produce in reading order, the assigned card, then the pick, then the full-card hand, then canon, each file written with its prompt sidecar the moment it is done, so a re-roll's spend front-loads onto the cards read first; declined challengers get no comp, their catalog thumb is their face. With parallel subagents, fan out one agent per card: each spawn is the shipped asset producer with a single-comp packet, that card's fields, PRODUCT.md, the shared frame, and the card's declared path, up to four in flight. Regenerate inline any slot still empty when its agent returns; drop without ceremony any slot still empty when the user answers. No other supervision is owed. Without parallel subagents, generate in the main thread after serving, same order, and let the harness's own generation display carry the progress; the wait for the answer follows the last file. The chosen card's comp is not spent by the choice: comp-led, it enters the comp round as compositional option one; code-led, it returns at the finish review as the critique reference, what the image dared that the build did not. Unchosen comps stay in `.impeccable/mocks/decision/` as the round's spent hand; they carry no approval and imply none. With no image generation, cards carry their identity in palette chips and facts, and that page is complete, not a lesser version; the page then also demotes every challenger's catalog art to a labeled thumbnail on its own, because salience must encode the verdict, never the accident of which cards have images.
 
-The execution contract, comp-led or code-led, is a workflow preference, not a per-surface decision, so no round asks it: the recorded default rides every round and the page's toggle handles the exception. Read the default from `.impeccable/config.json` (`buildPath`), with the gitignored `.impeccable/config.local.json` winning where one machine differs from the team's committed value; with neither, comp-led is the default whenever image generation exists. Author every direction and surface payload with `buildPath: { "value": <default>, "toggle": true }`; the page renders a footer toggle with the trade stated beside it, and the ANSWER returns `buildPath` plus `buildPathFlipped`. A flipped value binds that session only and is never written back, with one exception, and it is the only thing inside a round that earns a question about this preference (init records it up front on projects that get the chance): when `buildPathFlipped` comes back true on a project that records no `buildPath` at all, ask once after the round closes whether to keep it as the standing default. Either answer ends in a write to `.impeccable/config.json`; the answer picks the value, never whether to record one. Yes writes the flipped value, and "no, just this once" writes the value they flipped away from, which is the standing default they just confirmed by declining. Ask on the flip and never on the untouched default, because a user who left the toggle alone has told you nothing. A declined offer nothing writes down is an offer the next session makes again. When the user asks in words to change the standing default, update the file without asking. **Comp-led**: the chosen card's comp is law, generated before building when it does not exist yet, and the finish review audits the build against it; boldest composition on the table, fix rounds expected; comp-led makes the comp non-optional, no silent skipping. **Code-led**: no comp of this page and no apology for it; the QUALITY BAR boards still calibrate finish, and the ambition moves into the written contract, the FIRST VIEWPORT block plus a named signature interaction and motion grammar, which the finish reviewer audits in behavior; code-led is not a discount on commitment, the direction still lands fully committed in code. A code-led round still declares each card's comp path as a flip reserve: when the user flips the toggle to comp mid-round, `--wait` returns once with BUILD PATH FLIPPED while the page shimmers the slots; generate each open card's comp into its declared path then, lead first, and wait again. The flip back is free, and a comp that already rendered rides at the finish review as the critique reference. Without image generation there is no toggle and no choice: code-led is the only path, stated in one line rather than asked. The old two-card execution-contract round is retired; `followup: true` remains the general mechanism for delivering any later round over the same table via `--update`.
+The execution contract, comp-led or code-led, is a workflow preference, not a per-surface decision; no round asks it. The recorded default rides every round and the page's toggle handles the exception. Read the default from `.impeccable/config.json` (`buildPath`), the gitignored `.impeccable/config.local.json` winning where one machine differs from the team's committed value; with neither, comp-led is the default whenever image generation exists. Author every direction and surface payload with `buildPath: { "value": <default>, "toggle": true }`; the page renders a footer toggle with the trade stated beside it, and the ANSWER returns `buildPath` plus `buildPathFlipped`. A flipped value binds that session only and is never written back, with one exception, the only question this preference ever earns inside a round (init records it up front on projects that get the chance): when `buildPathFlipped` comes back true on a project that records no `buildPath` at all, ask once after the round closes whether to keep it as the standing default. Either answer writes `.impeccable/config.json`; the answer picks the value, never whether to record one. Yes writes the flipped value; "no, just this once" writes the value they flipped away from, the standing default they just confirmed by declining. Ask on the flip, never on the untouched default: a user who left the toggle alone told you nothing. A declined offer nothing writes down is an offer the next session makes again. When the user asks in words to change the standing default, update the file without asking. **Comp-led**: the chosen card's comp is law, generated before building when it does not yet exist, and the finish review audits the build against it; boldest composition on the table, fix rounds expected, and the comp is non-optional, no silent skipping. **Code-led**: no comp of this page and no apology for it; the QUALITY BAR boards still calibrate finish, and the ambition moves into the written contract, the FIRST VIEWPORT block plus a named signature interaction and motion grammar, which the finish reviewer audits in behavior; code-led is not a discount on commitment. A code-led round still declares each card's comp path as a flip reserve: when the user flips the toggle to comp mid-round, `--wait` returns once with BUILD PATH FLIPPED while the page shimmers the slots; generate each open card's comp into its declared path then, lead first, and wait again. The flip back is free, and a comp that already rendered rides at the finish review as the critique reference. Without image generation there is no toggle and no choice: code-led is the only path, stated in one line rather than asked. The old two-card execution-contract round is retired; `followup: true` remains the general mechanism for delivering any later round over the same table via `--update`.
 
 Catalog worlds are working systems, not mood references. When one survives, carry its palette and material, type and composition, topology, controls and state, and responsive rules into the product. When the source is itself an interface language, commit to its native grammar across navigation, content, controls, and states. Open the QUALITY BAR board and hero for the world you build the moment the choice lands, even if you viewed another card earlier; the ANSWER line names the chosen card's images (when the harness only reads files or runs sandboxed, download them into the workspace and open the relative path; sandboxed viewers reject absolute paths outside it). They set the craft level the build must reach, a rendered reference's finish, commitment, and art direction, never the composition; your surface serves this product.
 
@@ -64,11 +64,11 @@ Pick a color strategy before picking colors: Restrained (neutrals plus one accen
 
 Choose faces like objects from the subject's world, in the mode's register. Operate and Read surfaces are well served by system stacks and workhorse UI faces; Persuade and Experience surfaces want faces with a point of view, and these training-data defaults mean you stopped looking: Fraunces, Playfair Display, Cormorant, Lora, Crimson, Newsreader, Syne, Space Grotesk, Space Mono, IBM Plex, Inter-as-display, DM Sans, DM Serif, Outfit, Plus Jakarta Sans, Instrument Sans. Naming one of these faces anyway requires a reason no other face could satisfy, and a subject association is never that reason: books wanting a serif, bookshops wanting hand-lettering, and tech wanting a mono are the associations the list exists to break.
 
-Calibration: AI-generated interfaces cluster around a few looks regardless of subject: warm cream ground, high-contrast serif display, and a terracotta or signal-red accent; near-black with one neon accent and glowing edges; broadsheet-editorial hairlines, italic display serif, and small tracked mono labels. All are legitimate when the brief calls for them. Where the brief leaves the aesthetic free, landing in one means the self-check failed: if someone could guess your aesthetic from the category alone, or from category-plus-avoidance, rework until neither answer is obvious. Energy is not the enemy of trust: a brief's negative constraints (no gamification, no hype) rule out those devices, not exuberance, and adjectives describing the product's behavior (quiet support, calm coaching) do not dictate the surface's energy. A bookish, warm, or child-facing subject does not soften the calibration: book cloth, thread, jackets, endpapers, and shelf ephemera span the whole saturated spectrum, and cream paper is the smallest corner of that world; landing on cream plus serif for a book subject is the default wearing the subject's clothes. A brief-pinned world pins the world, not its softest rendition: the pinned world's full material range stays in play, and a rendition that matches what any model ships for that world failed the self-check at execution rather than selection.
+Calibration: AI-generated interfaces cluster around a few looks regardless of subject: warm cream ground, high-contrast serif display, and a terracotta or signal-red accent; near-black with one neon accent and glowing edges; broadsheet-editorial hairlines, italic display serif, and small tracked mono labels. All are legitimate when the brief calls for them. Where the brief leaves the aesthetic free, landing in one means the self-check failed: if someone could guess your aesthetic from the category alone, or from category-plus-avoidance, rework until neither answer is obvious. Energy is not the enemy of trust: a brief's negative constraints (no gamification, no hype) rule out those devices, not exuberance, and adjectives describing the product's behavior (quiet support, calm coaching) do not dictate the surface's energy. A bookish, warm, or child-facing subject does not soften the calibration: book cloth, thread, jackets, endpapers, and shelf ephemera span the whole saturated spectrum, and cream paper is the smallest corner of that world; landing on cream plus serif for a book subject is the default wearing the subject's clothes. A brief-pinned world pins the world, not its softest rendition: the pinned world's full material range stays in play, and a rendition matching what any model ships for that world failed the self-check at execution rather than selection.
 
 ## 5. Record the decision
 
-Before code, state the chosen direction as a contract in the artifact's opening comment, five short blocks, 150 words at most, in a form that survives the production build: an HTML comment in the emitted markup, never only a templating-frontmatter comment, placed as the first child of the document's body in the root layout, never inside a slotted or child component (some compilers, Astro among them, strip a slot's leading comment while keeping deeper ones). After the first production build, grep the built output for the seed key; a contract the build erased is a contract nobody can audit. THESIS: the one idea this surface owns and the category-default arrangement it refuses. OWN-WORLD: the palette and component language, specific enough to be recognizable with all content removed. STORY: what the visitor understands, believes, and does. FIRST VIEWPORT: the exact composition, what is where and at what scale, and where the primary action sits. FORM: the chosen form, its position on your ordered list, and the seed key the script printed. Close the comment with one more line, FINISH: the run's exit condition, verbatim "unreviewed and undocumented is unfinished; this build ends with the finish review, the verdict, and DESIGN.md". The comment tops the artifact you re-open on every edit, the one reminder that survives a long build: a page that looks complete with the FINISH line undischarged is not done, it is abandoned at the finish line. If a block reads like a mood, the direction is not decided yet; the finishing review audits the render against this contract.
+Before code, state the chosen direction as a contract in the artifact's opening comment, five short blocks, 150 words at most, in a form that survives the production build: an HTML comment in the emitted markup, never only a templating-frontmatter comment, placed as the first child of the document's body in the root layout, never inside a slotted or child component (some compilers, Astro among them, strip a slot's leading comment while keeping deeper ones). After the first production build, grep the built output for the seed key; a contract the build erased is a contract nobody can audit. THESIS: the one idea this surface owns and the category-default arrangement it refuses. OWN-WORLD: the palette and component language, specific enough to be recognizable with all content removed. STORY: what the visitor understands, believes, and does. FIRST VIEWPORT: the exact composition, what is where and at what scale, and where the primary action sits. FORM: the chosen form, its position on your ordered list, and the seed key the script printed. Close with one more line, FINISH: the run's exit condition, verbatim "unreviewed and undocumented is unfinished; this build ends with the finish review, the verdict, DESIGN.md, and every shipping raster carrying its provenance". The comment tops the artifact you re-open on every edit, the one reminder that survives a long build: a page that looks complete with the FINISH line undischarged is not done, it is abandoned at the finish line. If a block reads like a mood, the direction is not decided yet; the finishing review audits the render against this contract.
 
 On a new or replacement world, DESIGN.md is written at finish, from the built world, by the shipped documenter (section 7); a rulebook written before the build gets defended against reality instead of describing it, and hands the design-system detector an unstable target. A new world shipped with no DESIGN.md is still an incomplete run. An ordinary extension does not rewrite DESIGN.md.
 
@@ -80,18 +80,18 @@ If the work establishes durable strategy for a route or artifact, read its exist
 
 Keep the brief small: scope and visitor mode; audience, job, action/task, proof/content, and constraints; chosen direction and memorable moment; unresolved decisions. Do not copy global product truth or DESIGN.md tokens into it.
 
-On a comp-led build, whenever any image generation is available, a harness-native tool or the API fallback context.mjs reports, the locked direction is visualized before it is built, never skipped: load [visualize.md](visualize.md) and follow it, three compositional options put before the user for approval, the chosen card's decision comp plus two variations. This step is proven to produce the most compositional and ambitious work. On a code-led build the comp round is skipped by contract, never by drift: the ambition it would have carried lives in the direction contract's FIRST VIEWPORT block and named signature interaction, and the finish reviewer audits those promises in behavior.
+On a comp-led build, whenever any image generation is available (a harness-native tool or the API fallback context.mjs reports), the locked direction is visualized before it is built, never skipped: load [visualize.md](visualize.md) and follow it, three compositional options put before the user for approval, the chosen card's decision comp plus two variations. This step is proven to produce the most compositional and ambitious work. On a code-led build the comp round is skipped by contract, never by drift: the ambition it would have carried lives in the direction contract's FIRST VIEWPORT block and named signature interaction, and the finish reviewer audits those promises in behavior.
 
 For `shape`, return the selected direction to [shape.md](shape.md) and stop before persistence or implementation.
 
 ## 6. Build with full commitment
 
-When an approved comp exists, the comp is king, and the build happens in phases. Phase one is reproduction: rebuild the comp at its own breakpoint until a screenshot at the comp's width and height overlaps it near pixel-perfectly, materials, components, elevation, assets, and implied design language included. Exactly three concessions exist: fonts (the closest obtainable face), icons (exact match unless the user already chose an icon library), and genuine defects in the generated comp such as spelling errors. Everything else must match, and models systematically believe their HTML, CSS, and SVG recreation succeeded when it did not, so the overlap comparison is the authority, never your conviction: set the screenshot beside the freshly reopened comp image at identical dimensions after every region, never beside your memory of it, and when a region keeps losing that comparison, stop recreating it in code and produce it as a rendered asset composited into the page. The comp also outranks every written record of it: when the recorded brief or inventory commits to less than the comp shows, a softer texture, a sparser field, a sculpted plate reduced to flat CSS, correct the record upward to the comp; qualifiers like subtle, restrained, and low-contrast, and counts rounded down to a comfortable fraction, are how approved materials die between approval and build. A produced material must then survive to the screen: a texture buried under a nearly opaque color wash ships the wash, not the material, so judge every material by the screenshot beside the comp, never by the stylesheet. Only when reproduction holds does phase two begin: static regions that should live become animated or interactive, reveals and motion are added, then responsiveness across the surface's devices. Where the comp does not cover the whole surface, continue building the remainder inside the comp's recorded world and design language; a component the comp never shows inherits the recorded system's corner language, line weights, and materials, and may not introduce container styles, border weights, or chrome the comp never uses.
+When an approved comp exists, the comp is king, and the build happens in phases. The comp is a spatial contract, not a mood board: only the user can downgrade its authority, in explicit words, and difficulty never infers a downgrade. Phase one is reproduction: rebuild the comp at its own breakpoint until a screenshot at the comp's width and height overlaps it near pixel-perfectly, materials, components, elevation, assets, and implied design language included. Exactly three concessions exist: fonts (the closest obtainable face), icons (exact match unless the user already chose an icon library), and genuine defects in the generated comp such as spelling errors. Everything else must match, and models systematically believe their HTML, CSS, and SVG recreation succeeded when it did not, so the overlap comparison is the authority, never your conviction: set the screenshot beside the freshly reopened comp image at identical dimensions after every region, never beside your memory of it, and when a region keeps losing that comparison, stop recreating it in code and produce it as a rendered asset composited into the page. The comp also outranks every written record of it: when the recorded brief or inventory commits to less than the comp shows, a softer texture, a sparser field, a sculpted plate reduced to flat CSS, correct the record upward to the comp; qualifiers like subtle, restrained, and low-contrast, and counts rounded down to a comfortable fraction, are how approved materials die between approval and build. A produced material must then survive to the screen: a texture buried under a nearly opaque color wash ships the wash, not the material, so judge every material by the screenshot beside the comp, never by the stylesheet. Every color the brief records gets that comparison by number, not by eye: sample the build screenshot's ground, dominant fields, and accents the same way each record was taken (an interior patch average where the record is an average, both end colors where the record is a gradient) and set each value against its recorded counterpart (sampled from the comp itself when the brief lacks one), and when a texture or tile paints over a base token, measure the net on-screen value, because the eye files a drifted color under the same color word and the number is what catches it. Judge the gap like a colorist, not a diff tool: a difference with a color name (warmer, grayer, darker than the record) is drift to fix, while a few digits of render and compression noise are the same color. Only when reproduction holds does phase two begin: static regions that should live become animated or interactive, reveals and motion are added, then responsiveness across the surface's devices. Where the comp does not cover the whole surface, continue building the remainder inside the comp's recorded world and design language; a component the comp never shows inherits the recorded system's corner language, line weights, and materials, and may not introduce container styles, border weights, or chrome the comp never uses.
 
 Build the assigned direction, not a safer interpretation of it. The form supplies structure, reading order, component conventions, and native motion; the product supplies every fact. Commit every atom: nav, buttons, inputs, and links are rebuilt in the form's vocabulary, and a stock component inside a committed form is a lapse. Land the first build fully committed; committing is the hard part, and the passes that follow exist to make the committed thing clear and effective, never to dilute it. In unattended work, the safe rendition is the known risk.
 
 - **The first viewport is a thesis, not a header.** Demonstrate the mechanism immediately, at the scale the form has in life; do not trap the concept inside a standard hero or card shell. The memory test: if someone left after one viewport, what would they describe an hour later? If the honest answer is a mood, the concept has not committed yet.
-- **Prove the hero before building past it.** When an approved comp exists, render the first viewport, capture it, and set it beside the comp's first viewport before any later section: the hero carries the run's ambition, and every following section inherits its shortfall. Judge scale and density as quantities, a field at a tenth of the comp's coverage or type at half its weight is a different design, and a five-minute retry here is what a rebuild verdict at the finish costs when this check is skipped.
+- **Prove the hero before building past it.** When an approved comp exists, render the first viewport, capture it at the comp's own pixel dimensions, and set it beside the comp's first viewport before any later section: the hero carries the run's ambition, and every following section inherits its shortfall. Save that capture as `.impeccable/review/hero-repro.png` (create the directory); the finish reviewer verifies it exists, so a skipped checkpoint is a visible checkpoint. Judge scale and density as quantities, a field at a tenth of the comp's coverage or type at half its weight is a different design, and a five-minute retry here is what a rebuild verdict at the finish costs when this check is skipped.
 - **Prove, don't claim.** Show the subject doing its job: the interface at work, the mechanism dramatized, specifics a competitor could not copy-paste. Sections that restate a claim in different words add length, not substance. Demonstration data is design material: author it at full fidelity and label it synthetic; claims stay uninventable.
 - **Author the assets; never substitute chrome.** Great surfaces live on carefully made content: names, entries, copy, covers, thumbnails, textures. In greenfield work every blank the ask round left open is yours to author at production fidelity; content is authorable, claims are labelable, no section is omittable. An unanswered commercial claim ships as a clearly marked placeholder on the user's replacement list. When image generation exists, producing the design's imagery is part of building, at the scale the composition needs: a viewport that wants atmosphere gets a full-bleed layered scene, and a library of small centered subjects standardized for tidiness forecloses it. Gradients, glass, and generic icon tiles where an authored asset belongs are the gap wearing chrome; icons drawn in the world's own grammar are the remedy, not the target.
 - **Build the form's web leverage.** When the chosen world names a technique (canvas, WebGL, view transitions, generative motion), build the technique itself, not a static imitation of it; the graceful fallback serves constrained clients, it is not the default experience.
@@ -103,8 +103,18 @@ Preserve semantics, accessibility, performance, responsiveness, project conventi
 
 ## 7. Inspect and finish
 
-Inspect the surface's target sizes in one batched screenshot round: desktop and mobile on the web; on a native platform (`ios` / `android` / `adaptive`), the shipped device classes per OS, captured from the simulator or emulator the way the platform reference's Verifying the build section describes. Critique the render against the user's request and the direction contract, fix material gaps, and confirm with one final round; two rounds is the ceiling, and fixes batch between them rather than earning per-tweak screenshots. When an approved comp exists, the critique is a side-by-side: view the comp region and the build region together, the hero and each section as its own crop at legible scale, never one full-page thumbnail, which hides exactly the failures that matter, crude controls, wrong lettering character, flattened material, behind a superficially similar section order. On a Persuade surface, verify the mode did its job: a first-time visitor should know what this is, why it matters, and what to do within seconds, in the form's own vocabulary.
+Inspect the surface's target sizes in one batched screenshot round: desktop and mobile on the web; on a native platform (`ios` / `android` / `adaptive`), the shipped device classes per OS, captured from the simulator or emulator the way the platform reference's Verifying the build section describes. When the harness reports the user's actual viewport (an in-app browser's size, a named resolution), add that width to the set: the width that breaks is the one the user sees first. Critique the render against the user's request and the direction contract, fix material gaps, and confirm with one final round; two rounds is the ceiling, and fixes batch between them rather than earning per-tweak screenshots. When an approved comp exists, the critique is a side-by-side: view the comp region and the build region together, the hero and each section as its own crop at legible scale, never one full-page thumbnail, which hides exactly the failures that matter, crude controls, wrong lettering character, flattened material, behind a superficially similar section order. On a Persuade surface, verify the mode did its job: a first-time visitor should know what this is, why it matters, and what to do within seconds, in the form's own vocabulary.
 
-After the second inspection round the build thread's polishing is over: no further defect hunts, micro-edit scripts, or rebuilds here; whatever remains ships through the handoffs, where a fresh context does the finding better and cheaper. On the web, where this harness runs no design hook, run `node .agents/skills/impeccable/scripts/detect.mjs --json` on the changed targets once here, fix what is mechanical, and pass the remaining findings to the reviewer; a hookless web build that skips this ships every tell the hook exists to catch. A native platform skips the detector entirely: it reads HTML and CSS and has no verdict on native code, so the reviewer's floor check is the only slop gate and the input packet says so. Capture the screenshots into `.impeccable/review/`, one file per captured viewport (on the web, `desktop.png` and `mobile.png`; on native, one per device class, such as `phone.png` and `tablet.png`, suffixed per OS on adaptive), creating that directory when the harness does not; the paths you pass the reviewer are its spec, and that directory is where it looks when a passed path is missing. Then spawn the shipped finish reviewer, `impeccable-finish-reviewer` (`impeccable_finish_reviewer` in codex; `/impeccable-finish-reviewer` in Cursor; on GitHub Copilot say "Use the impeccable-finish-reviewer agent"), with the original request, confirmed answers, the artifact path, the screenshot paths, its direction contract, existing hook findings, the QUALITY BAR card and approved comp paths (on a code-led build there is no approved comp; the chosen decision comp rides in that slot as the critique reference, named as such), the craft-floor reference path, and on a native platform the platform reference path(s), [ios.md](ios.md) / [android.md](android.md), both on adaptive, plus one line saying no detector ran, so the reviewer judges in the platform's conventions rather than the web's. The reviewer has no browser; screenshots you fail to pass are checks it cannot run. Never read the shipped agents' definition files before spawning; the harness loads them at spawn, and you owe only the input packet. Wait on any agent with one long timeout rather than a loop of short polls, and spend the wait on the next independent step. Verify its return carries the five contract sections; on an empty or thrashed return, respawn once with the same inputs before doing anything else. This review never runs inside the build thread and never inherits it: spawn the reviewer fresh, with no forked conversation history (`fork_turns: 0` in codex); a reviewer that inherits your transcript inherits your framing, your optimism, and your abstractions, and everything it needs travels in the inputs above. Only a harness whose tool surface has no subagent capability at all substitutes a fresh in-thread pass after stepping fully out of the build context, run from [degraded/finish-reviewer.md](degraded/finish-reviewer.md), and a substituted or failed-and-replaced review is disclosed in one line at finish, never silently. When the reviewer's first material fix is a rebuild directive, fidelity failed wholesale rather than in patches, so skip the fix batch and execute the rebuild immediately: re-derive the named regions, produce the named assets, and send the result back for a verdict, telling the user what is happening rather than asking permission to fix a failure. The user is consulted only when a second rebuild directive arrives, both verdicts on the table, or when rebuilding would discard content the user approved. Otherwise apply the material fixes in one batch, rebuild once, and recapture the same viewports over the same files. A recapture measures positions, loading, and overflow; it cannot measure whether a fix reached the quality the finding named, so send the recaptured screenshots back to the same reviewer for a verdict scoring every material fix resolved, partial, or unresolved (through the harness's agent continuation; without one, run the scoring fresh from [degraded/finish-reviewer.md](degraded/finish-reviewer.md)'s Verdict Pass). Fixes scored partial or unresolved get another batch, recapture, and verdict. Two rounds is the budget an unattended run ends at; an attended session's ceiling belongs to the user, so when the second verdict still lists open items, put the table in front of them and let them choose between shipping as it stands and funding another round. Whoever is deciding, stop the moment a round resolves nothing, and the reviewer's findings are the only list you work from, never your own re-opened hunt. Report the final verdict table to the user as it stands, open items included, under the reviewer's own disposition word: a table with open material findings is never announced as a pass, and never under a softer label than the reviewer wrote. Do not run a second detector.
+A capture is evidence only when it is valid, and you validate before you send. Settle or disable entrance motion first: an element hidden by animation timing reads as a missing element and gets fixed into a regression. Capture full-page shots from the document top. Capture the comp comparison at the comp's own pixel dimensions. Then open every file once and confirm it shows what its name claims: no black or blank regions, no wrong section behind a right filename, no half-loaded state. A malformed capture sent onward costs the whole round; the reviewer answers it with `disposition: recapture` and nothing it reviewed binds.
 
-Then spawn the shipped documenter, `impeccable-documenter` (`impeccable_documenter` in codex), with the project root, the artifact path, the direction contract, PRODUCT.md, the [document.md](document.md) reference path, and the boundary to write at; it records DESIGN.md and the sidecar from the built world, ground truth over intention; without subagents the pass runs from [degraded/documenter.md](degraded/documenter.md). A clean detector pass is not finished; finished is the contract kept, the comp honored, the review closed, and the system recorded.
+After the second inspection round the build thread's polishing is over: no further defect hunts, micro-edit scripts, or rebuilds here; whatever remains ships through the handoffs, where a fresh context does the finding better and cheaper. On the web, where this harness runs no design hook, run `node .agents/skills/impeccable/scripts/detect.mjs --json` on the changed targets once here, fix what is mechanical, and pass the remaining findings to the reviewer; a hookless web build that skips this ships every tell the hook exists to catch. A native platform skips the detector entirely: it reads HTML and CSS and has no verdict on native code, so the reviewer's floor check is the only slop gate and the input packet says so. Capture the screenshots into `.impeccable/review/`, one file per captured viewport (on the web, `desktop.png` and `mobile.png`, plus `user-<width>.png` whenever the user's viewport joined the inspected set; on native, one per device class, such as `phone.png` and `tablet.png`, suffixed per OS on adaptive), creating that directory when the harness does not; the paths you pass the reviewer are its spec, every viewport you inspected is named required in the packet, and that directory is where it looks when a passed path is missing.
+
+Then spawn the shipped finish reviewer, `impeccable-finish-reviewer` (`impeccable_finish_reviewer` in codex; `/impeccable-finish-reviewer` in Cursor; on GitHub Copilot say "Use the impeccable-finish-reviewer agent"), with the original request, confirmed answers, the artifact path, the screenshot paths, the direction contract, existing hook findings, the QUALITY BAR card and approved comp paths (a code-led build has no approved comp; the chosen decision comp rides in that slot as the critique reference, named as such), the craft-floor reference path, and on a native platform the platform reference path(s), [ios.md](ios.md) / [android.md](android.md), both on adaptive, plus one line saying no detector ran, so the reviewer judges in the platform's conventions rather than the web's. The reviewer has no browser; screenshots you fail to pass are checks it cannot run. Never read the shipped agents' definition files before spawning; the harness loads them at spawn, and you owe only the input packet. Wait on any agent with one long timeout rather than a loop of short polls, and spend the wait on the next independent step. Verify the return carries the five contract sections (a recapture return carries one, its recapture list); on an empty or thrashed return, respawn once with the same inputs. This review never runs inside the build thread and never inherits it: spawn the reviewer fresh, with no forked conversation history (`fork_turns: 0` in codex); a reviewer that inherits your transcript inherits your framing, your optimism, and your abstractions, and everything it needs travels in the inputs above. Only a harness with no subagent capability at all substitutes a fresh in-thread pass after stepping fully out of the build context, run from [degraded/finish-reviewer.md](degraded/finish-reviewer.md), and a substituted or failed-and-replaced review is disclosed in one line at finish, never silently.
+
+Act on the disposition word; there are exactly four. **recapture**: the evidence failed, not the build. Recapture what the return names under the capture-validity rules, then run a full review over the new evidence. A review conducted on invalid evidence binds nothing, and a verdict pass may never follow it. **rebuild**: fidelity failed wholesale, not in patches. Skip the fix batch and execute the rebuild immediately: re-derive the named regions, produce the named assets, and send the result back for a fresh full review, never a verdict pass; a rebuild replaces regions wholesale, so the whole matrix runs again over the recaptures. Tell the user what is happening rather than asking permission to fix a failure. Consult the user only on a second rebuild directive, both verdicts on the table, or when rebuilding would discard content the user approved. **ship**: nothing is owed; report the verdict at its scope and continue to the documenter. **fix**: apply the material fixes in one batch, rebuild once, and recapture the same viewports over the same files. A recapture measures positions, loading, and overflow; it cannot measure whether a fix reached the quality the finding named, so send the recaptured screenshots back to the same reviewer for a verdict scoring every material fix resolved, partial, or unresolved (through the harness's agent continuation; without one, run the scoring fresh from [degraded/finish-reviewer.md](degraded/finish-reviewer.md)'s Verdict Pass). Fixes scored partial or unresolved get another batch, recapture, and verdict. Two rounds is the budget an unattended run ends at; an attended session's ceiling belongs to the user, so when the second verdict still lists open items, put the table in front of them and let them choose between shipping as it stands and funding another round. Whoever decides, stop the moment a round resolves nothing, and the reviewer's findings are the only list you work from, never your own re-opened hunt. Do not run a second detector.
+
+A rebuild and a fix round share one asset rule: a raster either round creates or replaces is still asset work under [visualize.md](visualize.md)'s Produce section and keeps its **provenance** like every build raster, and a raster the round abandons is deleted in the same batch. Before either round's result goes back for review or verdict, run `node .agents/skills/impeccable/scripts/embed-prompt.mjs --scan <asset-dir...>` over the directories the artifact's rasters ship from and clear every file it reports by embedding what it is missing: the exact generation prompt for a produced raster, the origin for a sourced, stock, or pre-existing one. The scan only reads; deletion is reserved for rasters the round abandoned, never for a file the scan flagged.
+
+Report the final verdict under the reviewer's own disposition word and at its actual scope. A verdict pass scores the listed fixes and nothing else: "the reviewer scored all three fixes resolved" is a claim it supports, "no material issues remain" is not. A table with open material findings is never announced as a pass, never softened, and never dressed as whole-surface approval when only a fix list was scored. When the user answers a ship with evidence against it, their own screenshot, a named mismatch with the comp, that evidence outranks every capture you made: put their material in the packet and spawn a fresh reviewer for a new full review. Patching inline and self-certifying is how a rejected page ships twice.
+
+Then spawn the shipped documenter, `impeccable-documenter` (`impeccable_documenter` in codex), with the project root, the artifact path, the direction contract, PRODUCT.md, the [document.md](document.md) reference path, and the boundary to write at; it records DESIGN.md and the sidecar from the built world, ground truth over intention; without subagents the pass runs from [degraded/documenter.md](degraded/documenter.md). The documenter runs after the last correction lands: when any fix round follows the documentation, re-run the documenter over the changed surface, because a DESIGN.md describing a layout that no longer exists turns defects into system guidance. A clean detector pass is not finished; finished is the contract kept, the comp honored, the review closed, and the system recorded.

.agents/skills/impeccable/reference/visualize.md

@@ -1,51 +1,55 @@
 # Visualize: Direction Comps & Asset Production
 
-Load this from [new-work.md](new-work.md) on a comp-led build, when image generation is available (a harness-native tool or the API fallback context.mjs reports). A code-led execution contract skips this file by design, not by drift: its ambition lives in the written direction contract and is audited in behavior, so do not load it for a code-led round. PRODUCT.md and DESIGN.md are preconditions. New-work has already resolved the visual world; this file must not reopen it. A surface-scope structure round that already put three visualized cards before the user (new-work.md, established world) has discharged this round: the locked card’s comp is the approved comp, so record the approval and continue at After approval; generate nothing new.
+Load this from [new-work.md](new-work.md) on a comp-led build, when image generation is available (a harness-native tool or the API fallback context.mjs reports). A code-led contract skips this file by design, not by drift; do not load it then. PRODUCT.md and DESIGN.md are preconditions. New-work has already resolved the visual world; this file must not reopen it. A surface-scope structure round that already put three visualized cards before the user (new-work.md, established world) has discharged this round: the locked card's comp is the approved comp, so record the approval and continue at After approval; generate nothing new.
 
-The purpose of a probe is to test composition, narrative, hierarchy, density, focal moment, signature use, and image requirements. It is not a second identity workshop. Keep DESIGN.md's palette, typography direction, material language, component character, imagery stance, and motion grammar fixed.
+A probe tests composition, narrative, hierarchy, density, focal moment, signature use, and image requirements. It is not a second identity workshop. Keep DESIGN.md's palette, typography direction, material language, component character, imagery stance, and motion grammar fixed.
 
 ## Generate three compositional options
 
-Render three distinct high-fidelity north-star comps of the requested surface, with whatever generation capability exists, saved under `.impeccable/mocks/` so they survive the session. Comp at the surface's own viewport: portrait at device size for a native app or mobile-first surface, desktop landscape otherwise; a phone screen comped landscape misstates the composition before anything gets built against it. Comps are the build thread's own work, never delegated: the thread that writes the comp prompts holds the direction's full context, and it has already seen every comp when the build starts. Open every image you produce or reference by its workspace-relative path, never an absolute one: sandboxed viewers reject absolute paths, and everything under the project root has a relative path. Base them on the real content and the surface concepts already developed with the user. On an established world, anchor every comp on the real identity: capture a screenshot of a representative existing page and pass it as a reference image (the harness image tool’s input image, or `generate-image.mjs --ref`); the prompt then leads with the new surface’s structure while the reference carries palette, type, and component character, because DESIGN.md words alone drift where a pixel reference does not. Name what the reference contributes and what it must not: chrome, palette, type, and component character carry over; the reference page’s own content does not, so a banner, hero, or card lifted verbatim from the reference is the reference leaking, not fidelity. Three is the number: one comp invites rubber-stamping, and the spread between three is what surfaces the composition worth building. The chosen card's decision comp is the first of the three: it already renders this direction at full fidelity under this file's discipline, so this round generates two more that vary what the first held fixed, and all three go to the approval point together. Only a round that arrives with no decision comp, a degraded roll, an identity-mode page, a direction pinned without the decision round, renders all three here.
+Render three distinct high-fidelity north-star comps of the requested surface, saved under `.impeccable/mocks/` so they survive the session. Comp at the surface's own viewport: portrait at device size for a native app or mobile-first surface, desktop landscape otherwise; a phone screen comped landscape misstates the composition before anything is built against it. Comps are the build thread's own work, never delegated: the thread that writes the prompts holds the direction's full context and has seen every comp when the build starts. Open every image by its workspace-relative path; sandboxed viewers reject absolute paths, and everything under the project root has a relative one. Base the comps on real content and the surface concepts already developed with the user. On an established world, anchor every comp on the real identity: capture a screenshot of a representative existing page and pass it as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`); the prompt leads with the new surface's structure while the reference carries palette, type, and component character, because DESIGN.md words alone drift where a pixel reference does not. Name what the reference contributes and what it must not: chrome, palette, type, and component character carry over; the reference page's own content does not, and a banner, hero, or card lifted verbatim is the reference leaking, not fidelity. Three is the number: one comp invites rubber-stamping; the spread between three surfaces the composition worth building. The chosen card's decision comp is the first of the three: it already renders this direction at full fidelity under this discipline, so generate two more that vary what the first held fixed, and send all three to the approval point together. Only a round arriving with no decision comp (a degraded roll, an identity-mode page, a direction pinned without the decision round) renders all three here.
 
-- A comp is a designed surface, not a picture of the subject. Lead the generation prompt with the surface's own structure, whatever regions this design actually has, named in order with their scale relationships; a page with no navigation states that instead of inventing one, and an unconventional surface states its unconventional skeleton. A prompt that leads with the world's atmosphere gets a vignette back: the model paints the fish market instead of the fish market's website. Self-check every render: if it could hang as a poster, or reads as a photograph or scene with some text on it, it is not a comp; regenerate with the layout scaffold stated more literally.
-- The inverse is also a failure: a surface with none of its subject in it. The subject appears as the content the regions exist to hold; the world dresses the frame and never displaces what the frame exists to show. The deletion usually rides in on the prompt's exclusion list, so exclusions bind invented claims, and a medium ban belongs to the committed imagery stance, never to caution. Before accepting a render, point at the subject: a render that depicts everything about the world and nothing of the subject fails however faithful its atmosphere, so regenerate with the subject's content named region by region.
-- A comp is judged as the shipped screen: the visitor's job must be readable from the image alone. Name the surface's mode from the render with no caption; a render whose mode cannot be read back is art direction without a surface, so regenerate with the visitor's job as the prompt's spine.
-- Commitment is depth, not coverage. The world enters through one dominant move plus the material, type, and spacing that support it, and the remaining regions hold still so that move can be read; a region that simply does its job in the world's own grammar carries the direction further than a region performing the concept. The check cuts competition, never content: a quieted region keeps its information and stops performing. Where the direction names a focal moment, a second element competing with it at the same scale means the comp is shouting; where it names none, several regions performing the concept at once is the same shout. Regenerate keeping the strongest move and quieting the rest. Busy is louder, not bolder.
+- A comp is a designed surface, not a picture of the subject. Lead the prompt with the surface's own structure: the regions this design has, named in order with their scale relationships; a page with no navigation says so instead of inventing one, and an unconventional surface states its unconventional skeleton. A prompt that leads with atmosphere gets a vignette back: the model paints the fish market instead of the fish market's website. Self-check every render: if it could hang as a poster, or reads as a photograph with some text on it, it is not a comp; regenerate with the layout scaffold stated more literally.
+- The inverse is also a failure: a surface with none of its subject in it. The subject appears as the content the regions hold; the world dresses the frame and never displaces what the frame shows. The deletion usually rides in on the prompt's exclusion list, so exclusions bind invented claims, and a medium ban belongs to the committed imagery stance, never to caution. Before accepting a render, point at the subject; a render that depicts everything about the world and nothing of the subject fails however faithful its atmosphere. Regenerate with the subject's content named region by region.
+- Judge a comp as the shipped screen: the visitor's job must be readable from the image alone. Name the surface's mode from the render with no caption; a render whose mode cannot be read back is art direction without a surface. Regenerate with the visitor's job as the prompt's spine.
+- Commitment is depth, not coverage. The world enters through one dominant move plus the material, type, and spacing that support it; the remaining regions hold still so that move can be read. A region that simply does its job in the world's grammar carries the direction further than a region performing the concept. The check cuts competition, never content: a quieted region keeps its information and stops performing. A second element competing with the named focal moment at the same scale means the comp is shouting; with no named focal moment, several regions performing the concept at once is the same shout. Regenerate keeping the strongest move and quieting the rest. Busy is louder, not bolder.
 - When the user shortlisted multiple concepts, spread the three across them.
 - When one direction is committed, vary the structural uncertainty an image can resolve: topology, sequence, density, hierarchy, focal composition, or interaction framing.
-- Show enough beyond the opening moment to prove the concept can govern the whole requested surface.
+- Show enough beyond the opening moment to prove the concept can govern the whole surface.
 - Do not generate a palette artifact, ask new atmosphere questions, introduce a different type voice, or invent a new motif. If the committed world cannot support the concept, return to the concept shortlist rather than changing the world.
 
-Treat each comp as a direction test, not a screenshot specification. Core UI text, responsive behavior, accessibility, semantics, and interaction states remain implementation responsibilities.
+Each comp is a direction test, not a screenshot specification. Core UI text, responsive behavior, accessibility, semantics, and interaction states remain implementation responsibilities.
 
 ## One approval point
 
-Show the three together on the decision page (`serve-question.mjs`, one option per comp with the comp as its hero), or in the harness only when it renders images inline; a text-only surface does not count as display. Ask what should carry forward, what feels false to the world, and whether the selected surface concept should be approved, combined, revised, or rejected. Then stop and wait. A structured simulated user counts as attended and receives the same question.
+Show the three together on the decision page (`serve-question.mjs`, one option per comp with the comp as its hero), or in the harness only when it renders images inline; a text-only surface does not count as display. Ask what should carry forward, what feels false to the world, and whether the selected concept should be approved, combined, revised, or rejected. Then stop and wait. A structured simulated user counts as attended and receives the same question.
 
 Do not begin code until the user approves a direction or explicitly delegates the choice. If they delegate, choose using the task brief, PRODUCT.md, and DESIGN.md, and state the evidence. Approval refines the task concept; it does not modify DESIGN.md.
 
-This approval point has no substitute and no skip condition. When the structured question tool errors, fall back to the decision page; only after both fail may you treat the choice as delegated, and a delegated pick is still recorded exactly as an approval is and disclosed in your first reply, not your last. The finish reviewer treats a build whose comp round produced comps with no recorded approval as carrying a material finding; decision comps under `.impeccable/mocks/decision/` are the direction round's hand, not comp-round output, and imply no approval on their own.
+This approval point has no substitute and no skip condition. When the structured question tool errors, fall back to the decision page; only after both fail may you treat the choice as delegated, and a delegated pick is recorded exactly as an approval is and disclosed in your first reply, not your last. The finish reviewer treats comp-round comps with no recorded approval as a material finding; decision comps under `.impeccable/mocks/decision/` are the direction round's hand, not comp-round output, and imply no approval on their own.
 
-After approval, record the choice where tools can find it: the approved comp's path goes in the surface brief, and the approved comp's `.json` prompt sidecar gains `"approved": true` (every comp generated through `generate-image.mjs` has one; create it if a native tool didn't). The sidecar travels with the mocks folder, so the approval survives sessions and machines that never see the brief. Then summarize the composition and the parts of the comp that must not be literalized, return to new-work.md, record the direction contract from the approved surface concept, and build.
+After approval, record the choice where tools can find it: the approved comp's path goes in the surface brief, and its `.json` prompt sidecar gains `"approved": true` (every comp generated through `generate-image.mjs` has one; create it if a native tool didn't). The sidecar travels with the mocks folder, so the approval survives sessions and machines that never see the brief. Summarize the composition and the parts of the comp that must not be literalized, return to new-work.md, record the direction contract from the approved concept, and build.
 
 ## Inventory implementation fidelity
 
-Before building, read the approved comp as a design system and record it in the brief: component grammar, corner language, line weights, elevation treatment, and the type ramp, because everything the comp does not show gets built from this record, and without it the fallback is the model's stock kit of square boxes, 1px grids, bento cells, and hard shadows. Then inventory the comp's major visible ingredients in writing (a short table in the surface brief or working notes; the finish reviewer audits shipped assets against it) and choose an implementation medium for each: semantic HTML/CSS/SVG, existing project asset, generated raster, sourced raster, icon library, canvas/WebGL, or accepted omission. The same written inventory names the comp's compositional commitments: navigation items and icons, headline levels and their scale relationship, signature geometry such as seams, masks, and overlaps, and each section's arrangement and density. The primary action gets its own row with its own medium: when the comp dissolves, stamps, erodes, or otherwise physically works the main CTA, that treatment is signature material on the page's most important element, and shrinking it to a border trick or a few decorative pixels is the compliance-token version of commitment. An element never written down is the element the build silently drops, and the direction contract's 150 words cannot carry this list, so this inventory is where it lives.
+Before building, read the approved comp as a design system and record it in the brief: component grammar, corner language, line weights, elevation treatment, and the type ramp. Everything the comp does not show gets built from this record; without it the fallback is the model's stock kit of square boxes, 1px grids, bento cells, and hard shadows. Then inventory the comp's major visible ingredients in writing (a short table in the surface brief or working notes; the finish reviewer audits shipped assets against it) and choose an implementation medium for each: semantic HTML/CSS/SVG, existing project asset, generated raster, sourced raster, icon library, canvas/WebGL, or accepted omission. The same inventory names the comp's compositional commitments: navigation items and icons, headline levels and their scale relationship, signature geometry such as seams, masks, and overlaps, and each section's arrangement and density. The primary action gets its own row with its own medium: when the comp dissolves, stamps, erodes, or otherwise physically works the main CTA, that treatment is signature material on the page's most important element, and shrinking it to a border trick is the compliance-token version of commitment. An element never written down is the element the build silently drops; the direction contract's 150 words cannot carry this list, so it lives here.
 
-The medium column is where an approved design most often dies, so it obeys a gate: the medium is decided by what the comp region shows, never by what feels buildable in the current stack. A human figure, a product object, machinery, or any material with lighting and depth is raster whatever the stack, and so is any texture by that name alone: woven cloth, paper grain, fabric, leather, brushed metal need no depth argument, because a CSS gradient or layered background is not a texture medium and "layered CSS textures" is not a medium at all. Writing "silhouette" for a photographic figure, or "CSS" for a sculpted panel's finish or a cotton field's weave, is not a medium choice, it is the quiet deletion of the approved design, and it is how a comp full of physical material becomes a flat page with the same section order. Style does not move this boundary: a comp region with perspective, shading, figure drawing, or dense mechanical detail is illustration however line-drawn it looks, and no build session can author illustration as vectors, so it regenerates as raster like any photograph. Authored SVG covers what a session can specify exactly, diagrams with countable elements, controls, flat shape systems, and it ends where drawing skill begins; an instruction-manual world does not convert its illustrations into diagrams, it makes them line-art illustrations. Produce such regions by regenerating them cleanly, with the approved comp and its embedded prompt as the reference for a fresh render at asset resolution; never crop pixels out of the comp itself, whose effective resolution sits far below asset grade. Dropping an image-native region instead of producing it is a scope decision the user makes at the approval point, never a silent flattening after it. Generated imagery is a material, not a claim: evidence rules bind assertions, specs, testimonials, and photographs presented as real, never render fidelity, so "no photography on hand" forbids fake proof, not an illustrated hero.
+The record is sampled, never estimated: read the comp's page **ground**, each dominant field, and each accent's actual hex from its pixels (ImageMagick, Python with PIL, any pixel-reading tool on the machine) and write the values into the same record. Take a flat field from any interior pixel, a textured or grainy one as the average of an interior patch (crop a swatch, scale it to one pixel), and a gradient as its two end colors; never sample an edge, where antialiasing blends neighbors into colors the design never chose. An adjective is a direction, not a record: cream covers everything from near-white to beige, charcoal a third of the value scale, and wherever no number pins a color, the rendition prior picks the spot. Sampled values supersede the palette chips on the decision and composition cards: those were authored before this comp existed, and a chip that disagrees with the comp's pixels is a draft the approval retired.
 
-The gate runs both ways: precise geometry, hard-edged shape systems, diagrams, expressive motion, shaders, and anything interactive are vector and GPU territory (SVG, canvas, WebGL), where a raster flattens what should move, scale, and respond, and code executed safely and professionally remains first-class there. A field or texture built from many small elements carries a quantity commitment either way: write down its approximate density and coverage ("thousands of glyphs over two-thirds of the fold, dense at the top fading into the path"), because a field rebuilt at a tenth of its density passes every checklist and still is not the design. TYPE rows carry the same discipline: name the face's compression class, and render one headline word against the comp before building on it; a visibly wider or lighter silhouette means the face is wrong, and every section built on it inherits the miss. Raster is for what the world paints; code is for what the world draws, animates, or reacts with, and choosing code there is ambition, not economy. Every `produce` entry is produced before the build ships, through the asset producer or in the current thread; an inventory with unproduced entries is an unfinished build, and this gate is where imagery-free pages come from when it is skipped.
+The medium column is where an approved design most often dies, so it obeys a gate: the medium is decided by what the comp region shows, never by what feels buildable in the current stack. A human figure, a product object, machinery, or any material with lighting and depth is raster whatever the stack; so is any texture by name alone: woven cloth, paper grain, fabric, leather, brushed metal need no depth argument, because a CSS gradient is not a texture medium and "layered CSS textures" is not a medium at all. Writing "silhouette" for a photographic figure, or "CSS" for a sculpted panel's finish, is not a medium choice; it is the quiet deletion of the approved design, and it is how a comp full of physical material becomes a flat page with the same section order. Style does not move this boundary: a comp region with perspective, shading, figure drawing, or dense mechanical detail is illustration however line-drawn it looks, and no build session can author illustration as vectors, so it regenerates as raster like any photograph. Authored SVG covers what a session can specify exactly (diagrams with countable elements, controls, flat shape systems) and ends where drawing skill begins; an instruction-manual world keeps its illustrations as line-art illustrations, not diagrams. Produce such regions by regenerating them cleanly, with the approved comp and its embedded prompt as the reference for a fresh render at asset resolution; never crop pixels out of the comp itself, whose effective resolution sits far below asset grade. Dropping an image-native region is a scope decision the user makes at the approval point, never a silent flattening after it. Generated imagery is a material, not a claim: evidence rules bind assertions, specs, testimonials, and photographs presented as real, never render fidelity; "no photography on hand" forbids fake proof, not an illustrated hero.
+
+The gate runs both ways: precise geometry, hard-edged shape systems, diagrams, expressive motion, shaders, and anything interactive are vector and GPU territory (SVG, canvas, WebGL), where a raster flattens what should move, scale, and respond. A field or texture built from many small elements carries a quantity commitment either way: write down its approximate density and coverage ("thousands of glyphs over two-thirds of the fold, dense at the top fading into the path"), because a field rebuilt at a tenth of its density passes every checklist and still is not the design. TYPE rows carry the same discipline: name the face's compression class, and render one headline word against the comp before building on it; a visibly wider or lighter silhouette means the face is wrong, and every section built on it inherits the miss. Raster is for what the world paints; code is for what the world draws, animates, or reacts with, and choosing code there is ambition, not economy. Every `produce` entry is produced before the build ships, through the asset producer or in the current thread; an inventory with unproduced entries is an unfinished build, and this gate is where imagery-free pages come from when it is skipped.
 
 Pay special attention to the dominant composition, signature use, image-native content, second-fold system, and any interaction the still image only implies.
 
-Treat the comp as a north star, not something to trace, and know what that allows: translation into semantic, responsive, accessible code, never recomposition. Keeping the palette and mood while redrawing the topology is a second art direction, not an adaptation. Do not rasterize core UI text or controls. Do not substitute a different visual driver after approval without asking.
+The comp is a north star, not something to trace, and know what that allows: translation into semantic, responsive, accessible code, never recomposition. Keeping the palette and mood while redrawing the topology is a second art direction, not an adaptation. Do not rasterize core UI text or controls. Do not substitute a different visual driver after approval without asking.
 
 ## Produce only the assets the build needs
 
-Generation context is part of the asset: a build composed by a thread that never saw the prompts places assets it does not understand. So prefer generating build-critical imagery in the build thread when the budget allows, and when a subagent produces assets instead, every asset must carry its prompt, and the builder reads those prompts before composing a single one of them. The carrier is uniform across harnesses: after generating any image with any tool, native or `generate-image.mjs` (which does it automatically), run `node .agents/skills/impeccable/scripts/embed-prompt.mjs <image> --prompt "<the prompt used>"` so the intent lives inside the file itself and survives copies between machines and harnesses; `--read` recovers it from any impeccable-generated image.
+Generation context is part of the asset: a build composed by a thread that never saw the prompts places assets it does not understand. Prefer generating build-critical imagery in the build thread when the budget allows; when a subagent produces assets instead, every asset carries its prompt, and the builder reads those prompts before composing. The carrier is uniform across harnesses: after generating any image with any tool, native or `generate-image.mjs` (which does it automatically), run `node .agents/skills/impeccable/scripts/embed-prompt.mjs <image> --prompt "<prompt>"` with the exact string the generation tool received, pasted whole, so the intent lives inside the file and survives copies between machines and harnesses; a summary reconstructed from memory records an asset that was never made. `--read` recovers the prompt from any impeccable-generated image, and `--scan <dir>` lists every raster in a directory still missing one. The embedded prompt plus the asset's row in the written inventory is the raster's **provenance**, and every raster the artifact references carries it; a sourced, stock, or pre-existing raster with no generation prompt embeds its origin instead.
 
-When the harness runs subagents, spawn the shipped asset producer every time, even when the inventory's produce bucket looks empty: its manifest is the independent second opinion on your media, and runs that skipped the spawn are the runs whose cotton became CSS. An honestly empty manifest costs one cheap spawn; a wrongly empty produce bucket costs the build its materials. Use the producer, `impeccable-asset-producer` (`impeccable_asset_producer` in codex; `/impeccable-asset-producer` in Cursor; on GitHub Copilot say "Use the impeccable-asset-producer agent"): give it the approved comp, output paths, required dimensions and formats, transparency needs, crop notes, and what must remain semantic code. Otherwise produce the minimum required assets in the current thread by the book: load [degraded/asset-producer.md](degraded/asset-producer.md) and follow it inline, with whatever generation exists, the native tool or generate-image.mjs.
+Provenance is owed for the run, not the build phase: a raster created or replaced later, in a fix batch or a reviewer's rebuild, is produced under this same section, prompt embedded and inventory row added, because the inventory is how the next thread knows what ships. A raster a fix abandons or supersedes is deleted from the assets directory in the same batch; an unreferenced raster with no record is a provenance leak, not a spare.
+
+When the harness runs subagents, spawn the shipped asset producer every time, even when the inventory's produce bucket looks empty: its manifest is the independent second opinion on your media, and the runs that skipped the spawn are the runs whose cotton became CSS. An honestly empty manifest costs one cheap spawn; a wrongly empty produce bucket costs the build its materials. Use `impeccable-asset-producer` (`impeccable_asset_producer` in codex; `/impeccable-asset-producer` in Cursor; on GitHub Copilot say "Use the impeccable-asset-producer agent"): give it the approved comp, output paths, required dimensions and formats, transparency needs, crop notes, and what must remain semantic code. Without subagents, produce the minimum required assets in the current thread by the book: load [degraded/asset-producer.md](degraded/asset-producer.md) and follow it inline, with whatever generation exists.
 
 Convert images with a converter context.mjs reported at boot (the IMAGE_TOOLS line); probe only when it reported none, at most once per session, never per image.
 

.agents/skills/impeccable/scripts/context.mjs

@@ -1013,6 +1013,27 @@ async function fetchLatestSkillVersion() {
   }
 }
 
+// Destroy fetch's global undici dispatcher before process.exit(): a live
+// keep-alive socket trips a libuv assertion on Windows/Node 24 after a
+// successful boot (nodejs/node#56645, issue #573).
+async function destroyFetchDispatcher() {
+  const dispatcher = globalThis[Symbol.for('undici.globalDispatcher.1')];
+  if (dispatcher && typeof dispatcher.destroy === 'function') {
+    try { await dispatcher.destroy(); } catch { /* exit regardless */ }
+  }
+}
+
+// Drain the boot payload before process.exit(): a live pipe that has not
+// flushed yet is truncated when Node tears down (issue #573 review). Then
+// close fetch so Windows teardown does not abort on the keep-alive socket.
+async function finishCli(output) {
+  await new Promise((resolve) => {
+    process.stdout.write(output, () => resolve());
+  });
+  await destroyFetchDispatcher();
+  process.exit(0);
+}
+
 // Two instructions used to sit in one directive: ask, and "if they agree, run
 // it". Nothing gated the second on an answer, and the same sentence said to
 // continue without waiting, so a run that could never establish agreement was
@@ -1159,8 +1180,7 @@ async function cli() {
     appendImageToolsDirective(parts);
     appendStalenessDirective(parts, ctx, cliOptions);
     if (updateDirective) parts.push(updateDirective);
-    process.stdout.write(parts.join('\n\n---\n\n') + '\n');
-    process.exit(0);
+    await finishCli(parts.join('\n\n---\n\n') + '\n');
   }
   const parts = [`# PRODUCT.md\n\n${ctx.product.trim()}`];
   if (ctx.hasDesign) {
@@ -1206,7 +1226,7 @@ async function cli() {
     }
   }
   if (updateDirective) parts.push(updateDirective);
-  process.stdout.write(parts.join('\n\n---\n\n') + '\n');
+  await finishCli(parts.join('\n\n---\n\n') + '\n');
 }
 
 function parseCliOptions(args) {

.agents/skills/impeccable/scripts/detector/browser/injected/index.mjs

@@ -626,7 +626,7 @@ if (IS_BROWSER) {
       if (currentStyle.filter && currentStyle.filter !== 'none') reasons.add('filter');
       if (currentStyle.backdropFilter && currentStyle.backdropFilter !== 'none') reasons.add('backdrop filter');
 
-      const solidBg = parseRgb(currentStyle.backgroundColor);
+      const solidBg = parseRgb(currentStyle.backgroundColor) || parseAnyColor(currentStyle.backgroundColor);
       if (solidBg && solidBg.a >= 0.95 && (!bgImage || bgImage === 'none')) break;
       current = current.parentElement;
     }
@@ -688,7 +688,7 @@ if (IS_BROWSER) {
       // starve the url()-backed texts this mode exists to sample.
       if (options.imageOnly && !reasons.includes('image background')) continue;
 
-      const textColor = parseRgb(style.color);
+      const textColor = parseRgb(style.color) || parseAnyColor(style.color);
       const fontSize = parseFloat(style.fontSize) || 16;
       const fontWeight = parseInt(style.fontWeight) || 400;
       const isLargeText = fontSize >= WCAG_LARGE_TEXT_PX || (fontSize >= WCAG_LARGE_BOLD_TEXT_PX && fontWeight >= 700);
@@ -985,7 +985,7 @@ if (IS_BROWSER) {
         return sample;
       }
     }
-    const bg = parseRgb(style.backgroundColor);
+    const bg = parseRgb(style.backgroundColor) || parseAnyColor(style.backgroundColor);
     if (bg && bg.a > 0.05) return { status: 'sampled', color: bg, method: 'solid-background' };
     return { status: 'unresolved', reason: 'no readable background' };
   }
@@ -1115,7 +1115,7 @@ if (IS_BROWSER) {
     }
 
     const style = getComputedStyle(el);
-    const textColor = parseRgb(style.color) || candidate.textColor;
+    const textColor = parseRgb(style.color) || parseAnyColor(style.color) || candidate.textColor;
     if (!textColor) return { ...candidate, status: 'unresolved', confidence: 'none', reason: 'unreadable text color' };
 
     const rect = getDirectTextRect(el) || el.getBoundingClientRect();

.agents/skills/impeccable/scripts/detector/cli/main.mjs

@@ -105,6 +105,13 @@ function formatFindings(findings, jsonMode) {
 // `optionsFor` maps a local path to scan options carrying that path's own
 // project design system (or base options when null). Falls back to a plain
 // object so direct/legacy callers still work.
+async function detectLocalFile(filePath, options) {
+  if (HTML_EXTENSIONS.has(path.extname(filePath).toLowerCase())) {
+    return detectHtml(filePath, options);
+  }
+  return detectText(fs.readFileSync(filePath, 'utf-8'), filePath, options);
+}
+
 async function handleStdin(optionsFor = () => ({})) {
   const resolve = typeof optionsFor === 'function' ? optionsFor : () => optionsFor;
   const chunks = [];
@@ -114,9 +121,7 @@ async function handleStdin(optionsFor = () => ({})) {
     const parsed = JSON.parse(input);
     const fp = parsed?.tool_input?.file_path;
     if (fp && fs.existsSync(fp)) {
-      const options = resolve(fp);
-      return HTML_EXTENSIONS.has(path.extname(fp).toLowerCase())
-        ? detectHtml(fp, options) : detectText(fs.readFileSync(fp, 'utf-8'), fp, options);
+      return detectLocalFile(fp, resolve(fp));
     }
   } catch { /* not JSON */ }
   return detectText(input, '<stdin>', resolve(null));
@@ -374,16 +379,10 @@ async function detectCli() {
           }
 
           for (const file of files) {
-            const ext = path.extname(file).toLowerCase();
             // Each file resolves its own project design system (cached by root),
             // so a scan spanning sibling projects applies the right rules per file.
             const fileOptions = scanOptionsFor(file);
-            let fileFindings;
-            if (HTML_EXTENSIONS.has(ext)) {
-              fileFindings = await detectHtml(file, fileOptions);
-            } else {
-              fileFindings = detectText(fs.readFileSync(file, 'utf-8'), file, fileOptions);
-            }
+            const fileFindings = await detectLocalFile(file, fileOptions);
             // Annotate findings with import context
             const importers = importedByMap.get(file);
             if (importers && importers.size > 0) {
@@ -396,13 +395,8 @@ async function detectCli() {
           }
         } else if (stat.isFile()) {
           if (shouldIgnoreDetectionFile(resolved, process.cwd(), detectionConfig)) continue;
-          const ext = path.extname(resolved).toLowerCase();
           const fileOptions = scanOptionsFor(resolved);
-          if (HTML_EXTENSIONS.has(ext)) {
-            allFindings.push(...await detectHtml(resolved, fileOptions));
-          } else {
-            allFindings.push(...detectText(fs.readFileSync(resolved, 'utf-8'), resolved, fileOptions));
-          }
+          allFindings.push(...await detectLocalFile(resolved, fileOptions));
         }
       }
     } finally {

.agents/skills/impeccable/scripts/detector/design-system.mjs

@@ -13,6 +13,11 @@ const FALLBACK_DIRS = ['.agents/context', 'docs'];
 // CLI can't import (separate tree). `.git` and `package.json` are the common
 // boundaries; `.impeccable` is our own project marker.
 const PROJECT_ROOT_MARKERS = ['.git', 'package.json', '.impeccable'];
+// Monorepo-root recognition, mirroring context.mjs's isMonorepoRoot: declared
+// workspace globs (package.json `workspaces`, pnpm-workspace.yaml `packages:`)
+// or a marker file beside apps/ or packages/ children.
+const MONOREPO_MARKER_FILES = ['pnpm-workspace.yaml', 'turbo.json', 'nx.json', 'lerna.json'];
+const MONOREPO_FALLBACK_PROJECT_DIRS = ['apps', 'packages'];
 const COLOR_CHANNEL_TOLERANCE = 6;
 // Shadow blacks at different alphas are different tokens (0.28 vs 0.55 is the
 // difference between a documented shadow and drift), so shadow matching cannot
@@ -575,14 +580,179 @@ function designSystemStartDir(targetPath, cwd = process.cwd()) {
   }
 }
 
+// Same two groups as context.mjs's readProjectPatternGroups: Impeccable
+// projectRoots govern any path they match (positive or negated); package-manager
+// globs only apply to paths the Impeccable group does not match.
+function readWorkspacePatternGroups(dir) {
+  const impeccable = [];
+  for (const name of ['config.json', 'config.local.json']) {
+    const roots = safeReadJson(path.join(dir, '.impeccable', name))?.projectRoots;
+    if (Array.isArray(roots)) {
+      impeccable.push(...roots.filter(entry => typeof entry === 'string' && entry.trim()).map(entry => entry.trim()));
+    }
+  }
+  const pkg = [];
+  const workspaces = safeReadJson(path.join(dir, 'package.json'))?.workspaces;
+  if (Array.isArray(workspaces)) pkg.push(...workspaces);
+  else if (Array.isArray(workspaces?.packages)) pkg.push(...workspaces.packages);
+  const lernaPackages = safeReadJson(path.join(dir, 'lerna.json'))?.packages;
+  if (Array.isArray(lernaPackages)) pkg.push(...lernaPackages);
+  try {
+    let inPackages = false;
+    for (const line of fs.readFileSync(path.join(dir, 'pnpm-workspace.yaml'), 'utf-8').split(/\r?\n/)) {
+      const trimmed = stripInlineYamlComment(line).trim();
+      if (!trimmed || trimmed.startsWith('#')) continue;
+      const flow = trimmed.match(/^packages:\s*\[(.*)\]\s*$/);
+      if (flow) {
+        pkg.push(...flow[1].split(',').map(entry => entry.trim().replace(/^['"]|['"]$/g, '')).filter(Boolean));
+        break;
+      }
+      if (/^packages:\s*$/.test(trimmed)) { inPackages = true; continue; }
+      if (!inPackages) continue;
+      const item = trimmed.match(/^-\s*(.+)$/);
+      if (item) pkg.push(item[1].trim().replace(/^['"]|['"]$/g, ''));
+      else if (/^[A-Za-z0-9_-]+:\s*/.test(trimmed)) break;
+    }
+  } catch { /* no pnpm-workspace.yaml */ }
+  return [impeccable, pkg];
+}
+
+function readWorkspacePatterns(dir) {
+  return readWorkspacePatternGroups(dir).flat();
+}
+
+function isMonorepoRoot(dir) {
+  if (readWorkspacePatterns(dir).some(pattern => !String(pattern).trim().startsWith('!'))) return true;
+  if (!MONOREPO_MARKER_FILES.some(file => fs.existsSync(path.join(dir, file)))) return false;
+  return MONOREPO_FALLBACK_PROJECT_DIRS.some(name => {
+    try {
+      return fs.readdirSync(path.join(dir, name), { withFileTypes: true }).some(entry => entry.isDirectory());
+    } catch {
+      return false;
+    }
+  });
+}
+
+function monorepoOwnsPath(root, boundaryDir) {
+  const rel = path.relative(root, boundaryDir);
+  if (!rel || rel.startsWith('..') || path.isAbsolute(rel)) return false;
+  const relSegments = rel.split(path.sep).filter(Boolean);
+
+  function normalizeWorkspacePattern(pattern) {
+    return String(pattern || '')
+      .trim()
+      .replace(/^['"]|['"]$/g, '')
+      .replace(/^\.\//, '')
+      .replace(/\/+$/, '');
+  }
+
+  function escapeRegExp(s) {
+    return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+  }
+
+  function segmentMatches(patternSegment, relSegment) {
+    if (patternSegment === '*') return true;
+    if (!patternSegment.includes('*')) return patternSegment === relSegment;
+    const re = new RegExp(`^${escapeRegExp(patternSegment).replace(/\\\*/g, '[^/]*')}$`);
+    return re.test(relSegment);
+  }
+
+  function matchGlobSegments(patternSegments, relSegments) {
+    function rec(pi, ri) {
+      if (pi === patternSegments.length) return ri === relSegments.length;
+      if (patternSegments[pi] === '**') {
+        if (pi === patternSegments.length - 1) return true;
+        for (let k = ri; k <= relSegments.length; k++) {
+          if (rec(pi + 1, k)) return true;
+        }
+        return false;
+      }
+      if (ri >= relSegments.length) return false;
+      if (!segmentMatches(patternSegments[pi], relSegments[ri])) return false;
+      return rec(pi + 1, ri + 1);
+    }
+    return rec(0, 0);
+  }
+
+  // Negations like !packages/excluded must also cover nested dirs under that path.
+  function matchesNegation(pattern) {
+    const patternSegments = normalizeWorkspacePattern(pattern).split('/').filter(Boolean);
+    if (!patternSegments.length) return false;
+    if (patternSegments.includes('**')) return matchGlobSegments(patternSegments, relSegments);
+    if (relSegments.length < patternSegments.length) return false;
+    for (let i = 0; i < patternSegments.length; i++) {
+      if (!segmentMatches(patternSegments[i], relSegments[i])) return false;
+    }
+    return true;
+  }
+
+  // Positive globs identify workspace packages at exact depth (`*` is a direct
+  // child). A nested package.json under that package is still owned: the
+  // ancestor directory of glob length must itself be a package.
+  function positiveOwns(pattern) {
+    const patternSegments = normalizeWorkspacePattern(pattern).split('/').filter(Boolean);
+    if (!patternSegments.length) return false;
+    if (patternSegments.includes('**')) return matchGlobSegments(patternSegments, relSegments);
+    if (relSegments.length < patternSegments.length) return false;
+    for (let i = 0; i < patternSegments.length; i++) {
+      if (!segmentMatches(patternSegments[i], relSegments[i])) return false;
+    }
+    if (relSegments.length === patternSegments.length) return true;
+    const ancestorDir = path.join(root, ...relSegments.slice(0, patternSegments.length));
+    return fs.existsSync(path.join(ancestorDir, 'package.json'));
+  }
+
+  function groupOwns(rawPatterns) {
+    const patterns = rawPatterns.map(normalizeWorkspacePattern).filter(Boolean);
+    if (!patterns.length) return null;
+    const excluded = patterns.some((pattern) => (
+      pattern.startsWith('!') && matchesNegation(pattern.slice(1))
+    ));
+    const included = patterns.filter((pattern) => !pattern.startsWith('!')).some(positiveOwns);
+    if (!excluded && !included) return null;
+    if (excluded) return false;
+    return true;
+  }
+
+  const [impeccable, pkg] = readWorkspacePatternGroups(root);
+  const fromImpeccable = groupOwns(impeccable);
+  if (fromImpeccable !== null) return fromImpeccable;
+  const fromPkg = groupOwns(pkg);
+  if (fromPkg !== null) return fromPkg;
+  if ([...impeccable, ...pkg].some((pattern) => !normalizeWorkspacePattern(pattern).startsWith('!'))) {
+    return false;
+  }
+  return relSegments.length >= 2 && MONOREPO_FALLBACK_PROJECT_DIRS.includes(relSegments[0]);
+}
+
+// Both forms of the home directory. The walk compares path strings, and a
+// symlinked home (e.g. /home -> /var/home) never string-matches the physical
+// paths a cwd-resolved target produces, which would let the post-boundary walk
+// sail through $HOME and inherit from it.
+function homeDirForms() {
+  const homeDir = path.resolve(os.homedir());
+  const forms = new Set([homeDir]);
+  try {
+    forms.add(fs.realpathSync(homeDir));
+  } catch { /* keep the logical form only */ }
+  return forms;
+}
+
 // Walk up from `startDir` to the directory that governs the target's design
 // system, mirroring skill/scripts/context.mjs's project-boundary semantics:
 //
 //   - A directory carrying a DESIGN.md (directly or in a fallback dir) IS the
 //     design root — that's where the rules live.
 //   - A directory carrying a project marker (.git / package.json / .impeccable)
-//     but no DESIGN.md is a project BOUNDARY: the walk stops with no design
-//     system, so a sibling project never inherits a parent's or cwd's rules.
+//     but no DESIGN.md is a project BOUNDARY. A nested package.json inherits
+//     the ancestor DESIGN.md only when that ancestor's workspace declarations
+//     include the path (negations win; a nested package under a matched
+//     workspace still inherits). Marker-only roots (turbo/nx/lerna/pnpm
+//     with no globs) still own apps/<name> and packages/<name>. A stray nested
+//     package that matches no glob does not inherit. This is detect's
+//     contamination contract, not skill-context's repoRoot fallback for
+//     excluded paths. A nested separate repository (.git with no workspace
+//     declaration) still inherits nothing (issue #570).
 //   - Reaching the home directory / filesystem root with neither means no
 //     design system at all — never process.cwd()'s.
 //
@@ -590,15 +760,33 @@ function designSystemStartDir(targetPath, cwd = process.cwd()) {
 // runs out. This is the fix for cross-project contamination.
 export function findDesignRoot(startDir) {
   let dir = path.resolve(startDir);
-  const homeDir = path.resolve(os.homedir());
+  const homeDirs = homeDirForms();
+  let boundary = null;
   while (true) {
-    if (resolveDesignMdPath(dir)) return { dir, hasDesign: true };
-    if (PROJECT_ROOT_MARKERS.some((marker) => fs.existsSync(path.join(dir, marker)))) {
-      return { dir, hasDesign: false };
+    if (!boundary && resolveDesignMdPath(dir)) return { dir, hasDesign: true };
+    if (boundary) {
+      // Past the boundary the walk only looks for the monorepo root that owns
+      // the workspace path (workspace globs including negations, or marker-only
+      // apps/packages fallback). Monorepo-root before .git, same order as
+      // context.mjs: a workspace root carrying its own .git is still recognized,
+      // while a .git that declares no workspaces is a separate repository and
+      // stops the walk with nothing inherited. The home directory is never an
+      // owning root, same as context.mjs's findMonorepoRoot, which stops at
+      // homeDir before its monorepo check.
+      if (!homeDirs.has(dir) && isMonorepoRoot(dir)) {
+        if (monorepoOwnsPath(dir, boundary.dir)) return { dir, hasDesign: !!resolveDesignMdPath(dir) };
+        return boundary;
+      }
+      if (fs.existsSync(path.join(dir, '.git'))) return boundary;
+    } else if (PROJECT_ROOT_MARKERS.some((marker) => fs.existsSync(path.join(dir, marker)))) {
+      boundary = { dir, hasDesign: false };
+      // A boundary that is itself a monorepo root, or a separate repository
+      // with its own .git, inherits nothing from above.
+      if (isMonorepoRoot(dir) || fs.existsSync(path.join(dir, '.git'))) return boundary;
     }
-    if (dir === homeDir) return null;
+    if (homeDirs.has(dir)) return boundary;
     const parent = path.dirname(dir);
-    if (parent === dir) return null;
+    if (parent === dir) return boundary;
     dir = parent;
   }
 }

.agents/skills/impeccable/scripts/detector/detect-antipatterns-browser.js

@@ -773,14 +773,22 @@ function extractColorFunctionTokens(value) {
 function parseGradientColors(bgImage) {
   if (!bgImage || !bgImage.includes('gradient')) return [];
   const colors = [];
+  const tokenSpans = [];
+  let from = 0;
   // Stops arrive in whatever syntax the author wrote and the browser kept.
   // A dark ground painted as `linear-gradient(oklch(...), oklch(...))` used
   // to read as a gradient with no stops at all.
   for (const token of extractColorFunctionTokens(bgImage)) {
+    const start = bgImage.indexOf(token, from);
+    if (start < 0) break;
+    tokenSpans.push({ start, end: start + token.length });
+    from = start + token.length;
     const c = parseAnyColor(token);
     if (c) colors.push(c);
   }
   for (const m of bgImage.matchAll(/#([0-9a-f]{6}|[0-9a-f]{3})\b/gi)) {
+    // Nested hex inside color-mix is an ingredient, not a stop (issue #578).
+    if (tokenSpans.some(s => m.index >= s.start && m.index < s.end)) continue;
     const h = m[1];
     if (h.length === 6) {
       colors.push({ r: parseInt(h.slice(0,2),16), g: parseInt(h.slice(2,4),16), b: parseInt(h.slice(4,6),16), a: 1 });
@@ -1955,20 +1963,19 @@ function scanCssTextForGlow(content) {
   return results;
 }
 
-// Decorative grid or line-field backgrounds drawn with hairline
+// Decorative two-axis grid backgrounds drawn with hairline
 // linear-gradient layers tiled by a fixed pixel cell. Shared by the HTML
 // pattern pass and the regex source engine so standalone CSS, component
 // styles, and inline styles receive the same coverage. Both signals must
 // co-occur in one declaration block; unrelated rules must not add up across
-// the file. Returns [{ index, snippet }], capped at one finding per source to
-// match the page-level HTML check's existing behavior.
+// the file. A single hairline is a line, divider, or rail, not a grid, even
+// when tiled by a 2D px cell. Returns [{ index, snippet }], capped at one
+// finding per source to match the page-level HTML check's existing behavior.
 function scanCssTextForGridBackground(content) {
   const hairlineRe = /\b\d{1,3}px\s*,\s*transparent\s+\d{1,3}px/gi;
   const invertedHairlineRe = /transparent\s+calc\(100%\s*-\s*\d{1,3}px\)/gi;
   const sizeDeclPxRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\b/i;
-  const sizeDeclPxPairRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\s+\d{1,3}px/i;
   const shorthandPxAnyRe = /\/\s*\d{1,3}px\b/;
-  const shorthandPxPairRe = /\/\s*\d{1,3}px\s+\d{1,3}px/;
   const bgDeclRe = /\bbackground(?:-image)?\s*:\s*([^;{}"']*)/gi;
   const blockRe = /\{([^{}]*)\}|style\s*=\s*"([^"]*)"|style\s*=\s*'([^']*)'/gi;
   let blk;
@@ -1985,13 +1992,10 @@ function scanCssTextForGridBackground(content) {
     }
     if (hairlineCount === 0) continue;
     const hasPxCell = sizeDeclPxRe.test(block) || shorthandPxAnyRe.test(bgJoined);
-    const hasPxPairCell = sizeDeclPxPairRe.test(block) || shorthandPxPairRe.test(bgJoined);
-    if ((hairlineCount >= 2 && hasPxCell) || hasPxPairCell) {
+    if (hairlineCount >= 2 && hasPxCell) {
       return [{
         index: blk.index,
-        snippet: hairlineCount >= 2
-          ? 'two-axis grid-line gradient background'
-          : 'px-tiled hairline line-field background',
+        snippet: 'two-axis grid-line gradient background',
       }];
     }
   }
@@ -3986,7 +3990,7 @@ function checkElementAIPaletteDOM(el) {
   }
 
   // Check for neon text (vivid cyan/purple color on dark background)
-  const textColor = parseRgb(style.color);
+  const textColor = parseRgb(style.color) || parseAnyColor(style.color);
   if (textColor && hasChroma(textColor, 80)) {
     const hue = getHue(textColor);
     const isAIPalette = (hue >= 160 && hue <= 200) || (hue >= 260 && hue <= 310);
@@ -7281,7 +7285,7 @@ if (IS_BROWSER) {
       if (currentStyle.filter && currentStyle.filter !== 'none') reasons.add('filter');
       if (currentStyle.backdropFilter && currentStyle.backdropFilter !== 'none') reasons.add('backdrop filter');
 
-      const solidBg = parseRgb(currentStyle.backgroundColor);
+      const solidBg = parseRgb(currentStyle.backgroundColor) || parseAnyColor(currentStyle.backgroundColor);
       if (solidBg && solidBg.a >= 0.95 && (!bgImage || bgImage === 'none')) break;
       current = current.parentElement;
     }
@@ -7343,7 +7347,7 @@ if (IS_BROWSER) {
       // starve the url()-backed texts this mode exists to sample.
       if (options.imageOnly && !reasons.includes('image background')) continue;
 
-      const textColor = parseRgb(style.color);
+      const textColor = parseRgb(style.color) || parseAnyColor(style.color);
       const fontSize = parseFloat(style.fontSize) || 16;
       const fontWeight = parseInt(style.fontWeight) || 400;
       const isLargeText = fontSize >= WCAG_LARGE_TEXT_PX || (fontSize >= WCAG_LARGE_BOLD_TEXT_PX && fontWeight >= 700);
@@ -7640,7 +7644,7 @@ if (IS_BROWSER) {
         return sample;
       }
     }
-    const bg = parseRgb(style.backgroundColor);
+    const bg = parseRgb(style.backgroundColor) || parseAnyColor(style.backgroundColor);
     if (bg && bg.a > 0.05) return { status: 'sampled', color: bg, method: 'solid-background' };
     return { status: 'unresolved', reason: 'no readable background' };
   }
@@ -7770,7 +7774,7 @@ if (IS_BROWSER) {
     }
 
     const style = getComputedStyle(el);
-    const textColor = parseRgb(style.color) || candidate.textColor;
+    const textColor = parseRgb(style.color) || parseAnyColor(style.color) || candidate.textColor;
     if (!textColor) return { ...candidate, status: 'unresolved', confidence: 'none', reason: 'unreadable text color' };
 
     const rect = getDirectTextRect(el) || el.getBoundingClientRect();

.agents/skills/impeccable/scripts/detector/engines/regex/detect-text.mjs

@@ -42,6 +42,7 @@ function shouldRunPageAnalyzers(content, filePath) {
 }
 
 const JS_SOURCE_EXTS = new Set(['.js', '.jsx', '.ts', '.tsx', '.mjs', '.cjs']);
+const STYLESHEET_EXTS = new Set(['.css', '.scss', '.sass', '.less']);
 const REGEX_PREFIX_KEYWORDS = new Set(['await', 'case', 'default', 'delete', 'do', 'else', 'in', 'instanceof', 'new', 'of', 'return', 'throw', 'typeof', 'void', 'yield']);
 const BLOCK_BRACE_PREFIX_KEYWORDS = new Set(['do', 'else', 'finally', 'try']);
 
@@ -256,6 +257,153 @@ function stripCssComments(content) {
   return content.replace(/\/\*[\s\S]*?\*\//g, comment => comment.replace(/[^\n]/g, ' '));
 }
 
+function blankHtmlComments(text) {
+  return text.replace(/<!--[\s\S]*?-->/g, comment => comment.replace(/[^\n]/g, ' '));
+}
+
+function blankCssLineCommentsInStyleBlocks(text) {
+  const re = /<style\b[^>]*>([\s\S]*?)<\/style>/gi;
+  let output = '';
+  let lastIndex = 0;
+  let match;
+  while ((match = re.exec(text)) !== null) {
+    const inner = match[1];
+    const openLength = match[0].length - inner.length - '</style>'.length;
+    output += text.slice(lastIndex, match.index);
+    output += match[0].slice(0, openLength);
+    output += blankCssLineComments(inner);
+    output += match[0].slice(openLength + inner.length);
+    lastIndex = re.lastIndex;
+  }
+  return output + text.slice(lastIndex);
+}
+
+function blankHtmlAndCssCommentsOutsideScripts(text) {
+  const re = /<script\b[^>]*>[\s\S]*?<\/script>/gi;
+  let output = '';
+  let lastIndex = 0;
+  let match;
+  while ((match = re.exec(text)) !== null) {
+    output += blankCssLineCommentsInStyleBlocks(stripCssComments(blankHtmlComments(text.slice(lastIndex, match.index))));
+    output += match[0];
+    lastIndex = re.lastIndex;
+  }
+  return output + blankCssLineCommentsInStyleBlocks(stripCssComments(blankHtmlComments(text.slice(lastIndex))));
+}
+
+function blankCssLineComments(text) {
+  let output = '';
+  let state = 'code';
+  let urlDepth = 0;
+  for (let i = 0; i < text.length; i++) {
+    const char = text[i];
+    const next = text[i + 1];
+    if (state === 'line') {
+      if (char === '\n') {
+        output += '\n';
+        state = 'code';
+      } else {
+        output += ' ';
+      }
+      continue;
+    }
+    if (state === 'single' || state === 'double') {
+      output += char;
+      if (char === '\\' && next) {
+        output += next;
+        i++;
+      } else if ((state === 'single' && char === "'") || (state === 'double' && char === '"')) {
+        state = 'code';
+      }
+      continue;
+    }
+    const prev = output.length ? output[output.length - 1] : '';
+    if (char === '/' && next === '/' && urlDepth === 0 && prev !== ':' && prev !== '(' && prev !== '\\') {
+      output += '  ';
+      i++;
+      state = 'line';
+      continue;
+    }
+    if (char === "'") state = 'single';
+    else if (char === '"') state = 'double';
+    if (char === '(') {
+      const behind = output.replace(/\s+$/, '');
+      if (urlDepth > 0 || /url$/i.test(behind)) urlDepth++;
+    } else if (char === ')' && urlDepth) {
+      urlDepth--;
+    }
+    output += char;
+  }
+  return output;
+}
+
+function findAstroFrontmatterClose(text) {
+  if (!text.startsWith('---')) return -1;
+  let cursor = text.indexOf('\n');
+  if (cursor === -1) return -1;
+  cursor += 1;
+  while (cursor < text.length) {
+    if (text[cursor - 1] === '\n' && text.startsWith('---', cursor)) {
+      let end = cursor + 3;
+      while (text[end] === ' ' || text[end] === '\t') end++;
+      if (end >= text.length || text[end] === '\n' || text[end] === '\r') return cursor - 1;
+    }
+    const char = text[cursor];
+    const next = text[cursor + 1];
+    if (char === "'" || char === '"') {
+      const close = findQuotedStringEnd(text, cursor, char);
+      if (close === -1) return -1;
+      cursor = close + 1;
+      continue;
+    }
+    if (char === '`') {
+      const close = findTemplateLiteralEnd(text, cursor);
+      if (close === -1) return -1;
+      cursor = close + 1;
+      continue;
+    }
+    if (char === '/' && next === '/') {
+      const lineEnd = text.indexOf('\n', cursor);
+      if (lineEnd === -1) return -1;
+      cursor = lineEnd;
+      continue;
+    }
+    if (char === '/' && next === '*') {
+      const commentEnd = text.indexOf('*/', cursor + 2);
+      if (commentEnd === -1) return -1;
+      cursor = commentEnd + 2;
+      continue;
+    }
+    if (char === '/' && next !== '/' && next !== '*') {
+      const close = findRegexLiteralEnd(text, cursor);
+      if (close !== -1) {
+        cursor = close + 1;
+        continue;
+      }
+    }
+    cursor++;
+  }
+  return -1;
+}
+
+function blankAstroFrontmatterComments(text) {
+  const close = findAstroFrontmatterClose(text);
+  if (close === -1) return text;
+  return stripJsComments(text.slice(0, close)) + text.slice(close);
+}
+
+function blankCommentsForMatchers(text, ext) {
+  if (PAGE_ANALYZER_EXTS.has(ext)) {
+    const withFrontmatter = ext === '.astro' ? blankAstroFrontmatterComments(text) : text;
+    return blankHtmlAndCssCommentsOutsideScripts(withFrontmatter);
+  }
+  if (STYLESHEET_EXTS.has(ext)) {
+    const withoutBlocks = stripCssComments(text);
+    return ext === '.css' ? withoutBlocks : blankCssLineComments(withoutBlocks);
+  }
+  return text;
+}
+
 function firstOverusedGoogleFont(text) {
   return extractGoogleFontFamilies(text).find(f => OVERUSED_FONTS.has(f)) || '';
 }
@@ -1028,14 +1176,13 @@ function detectText(content, filePath, options = {}) {
   const ext = extFromFilePath(filePath);
   const commentStrippedSource = JS_SOURCE_EXTS.has(ext) ? stripJsComments(content, {
     jsx: ext === '.js' || ext === '.jsx' || ext === '.tsx',
-  }) : content;
+  }) : blankCommentsForMatchers(content, ext);
   const source = stripCssInJsComments(commentStrippedSource, ext);
   const lines = source.split('\n');
 
   // Run regex matchers on the full file content (catches Tailwind classes, inline styles)
   // Enable block context for CSS files where related properties span multiple lines
-  const cssLike = new Set(['.css', '.scss', '.sass', '.less']);
-  findings.push(...runRegexMatchers(lines, filePath, 0, cssLike.has(ext) || null, {
+  findings.push(...runRegexMatchers(lines, filePath, 0, STYLESHEET_EXTS.has(ext) || null, {
     profile,
     phase: 'source',
   }));
@@ -1050,7 +1197,7 @@ function detectText(content, filePath, options = {}) {
     scanCssTextForPseudoStripe(text).map(hit =>
       finding(hit.id, filePath, hit.snippet, lineOffset + text.slice(0, hit.index).split('\n').length));
 
-  if (cssLike.has(ext)) {
+  if (STYLESHEET_EXTS.has(ext)) {
     findings.push(...scanInsetStripeCss(content, filePath));
     findings.push(...pseudoStripeFindings(content, 0));
   }
@@ -1078,7 +1225,8 @@ function detectText(content, filePath, options = {}) {
     }, () => extractStyleBlocks(content, ext))
     : extractStyleBlocks(content, ext);
   for (const block of styleBlocks) {
-    const blockLines = block.content.split('\n');
+    const blockContent = blankCssLineComments(stripCssComments(block.content));
+    const blockLines = blockContent.split('\n');
     findings.push(...runRegexMatchers(blockLines, filePath, block.startLine - 1, true, {
       profile,
       phase: 'style-block',
@@ -1089,8 +1237,8 @@ function detectText(content, filePath, options = {}) {
     // 1-based, so the offset is startLine - 2; startLine - 1 double-counted and
     // reported every selector one line low. runRegexMatchers keeps startLine - 1
     // because it indexes its split lines from zero.
-    findings.push(...scanInsetStripeCss(block.content, filePath, block.startLine - 2));
-    findings.push(...pseudoStripeFindings(block.content, block.startLine - 2));
+    findings.push(...scanInsetStripeCss(blockContent, filePath, block.startLine - 2));
+    findings.push(...pseudoStripeFindings(blockContent, block.startLine - 2));
   }
 
   // Extract and scan CSS-in-JS template literals

.agents/skills/impeccable/scripts/detector/engines/static-html/css-cascade.mjs

@@ -835,10 +835,11 @@ class StaticElement {
     }
   }
   closest(selector) {
+    const matcher = this._doc.matcherFor(selector);
     let cur = this.node;
     while (cur && cur.type === 'tag') {
       try {
-        if (this._doc.is(cur, selector)) return this._doc.wrap(cur);
+        if (matcher(cur)) return this._doc.wrap(cur);
       } catch {
         return null;
       }
@@ -862,9 +863,10 @@ class StaticDocument {
     this.root = root;
     this.selectAll = modules.selectAll;
     this.selectOne = modules.selectOne;
-    this.is = modules.is;
+    this.compile = modules.compile;
     this.domutils = modules.domutils;
     this._wrappers = new WeakMap();
+    this._compiledSelectors = new Map();
     this._styleMap = new WeakMap();
     this._hoverStyleMap = new WeakMap();
     this._accentDashPseudo = new WeakSet();
@@ -882,6 +884,20 @@ class StaticDocument {
     }
     return wrapped;
   }
+  matcherFor(selector) {
+    let matcher = this._compiledSelectors.get(selector);
+    if (!matcher) {
+      try {
+        matcher = this.compile(selector);
+      } catch (err) {
+        // Cache the failure as a rethrower so a bad selector still reaches
+        // closest()'s catch on every call, first and repeat alike.
+        matcher = () => { throw err; };
+      }
+      this._compiledSelectors.set(selector, matcher);
+    }
+    return matcher;
+  }
   querySelectorAll(selector) {
     try {
       return this.selectAll(selector, this.root.children || []).map(node => this.wrap(node));
@@ -948,8 +964,34 @@ function buildStaticWindow(staticDoc) {
   };
 }
 
+function resolveLinkedCssPath(fileDir, href) {
+  const stripped = href.split(/[?#]/)[0];
+  const rootRelative = stripped.startsWith('/') && !stripped.startsWith('//');
+  if (!rootRelative) return path.resolve(fileDir, stripped);
+  // Drop "." and reject ".." so /../outside.css cannot walk out of dir.
+  const segments = stripped.replace(/^\/+/, '').split(/[/\\]/).filter(p => p && p !== '.');
+  if (segments.some(p => p === '..')) return path.join(fileDir, segments.filter(p => p !== '..').join(path.sep));
+  const rel = segments.join(path.sep);
+  let dir = fileDir;
+  for (;;) {
+    const parent = path.dirname(dir);
+    if (parent === dir) break; // never use the filesystem root as document root
+    try {
+      const candidate = path.join(dir, rel);
+      if (fs.statSync(candidate).isFile()) return candidate;
+    } catch { /* missing or unreadable candidate */ }
+    // Stop at the project root so a coincidental ~/static/app.css cannot win.
+    try {
+      if (fs.existsSync(path.join(dir, 'package.json')) || fs.existsSync(path.join(dir, '.git'))) break;
+    } catch { /* unreadable marker */ }
+    dir = parent;
+  }
+  return path.join(fileDir, rel);
+}
+
 function collectStaticCssText(root, fileDir, profile, filePath, modules) {
   const styleTexts = [];
+  const warnedMissingStylesheets = new Set();
   for (const styleEl of modules.selectAll('style', root.children || [])) {
     styleTexts.push(modules.domutils.textContent(styleEl));
   }
@@ -958,10 +1000,10 @@ function collectStaticCssText(root, fileDir, profile, filePath, modules) {
     const rel = link.attribs?.rel || '';
     const href = link.attribs?.href || '';
     if (!/\bstylesheet\b/i.test(rel) || !href || /^(https?:)?\/\//i.test(href)) continue;
-    // Cache-busting hrefs (styles.css?v=3) resolve to the file, not to a
-    // literal path with the query in it; a versioned link otherwise made the
-    // whole stylesheet invisible to every element-level check.
-    const cssPath = path.resolve(fileDir, href.split(/[?#]/)[0]);
+    // Cache-busting (styles.css?v=3) and root-relative (/static/app.css) hrefs
+    // must not resolve as OS-absolute paths; otherwise the whole stylesheet is
+    // invisible to every element-level check.
+    const cssPath = resolveLinkedCssPath(fileDir, href);
     try {
       const css = profileStep(profile, {
         engine: 'static-html',
@@ -971,7 +1013,14 @@ function collectStaticCssText(root, fileDir, profile, filePath, modules) {
         detail: href,
       }, () => fs.readFileSync(cssPath, 'utf-8'));
       styleTexts.push(css);
-    } catch { /* skip unreadable */ }
+    } catch {
+      if (!warnedMissingStylesheets.has(cssPath)) {
+        warnedMissingStylesheets.add(cssPath);
+        process.stderr.write(
+          `impeccable detect: could not read linked stylesheet ${href} (resolved to ${cssPath}); color and custom-property rules will be incomplete\n`
+        );
+      }
+    }
   }
   return styleTexts.join('\n');
 }

.agents/skills/impeccable/scripts/detector/engines/static-html/detect-html.mjs

@@ -134,7 +134,7 @@ async function detectHtml(filePath, options = {}) {
         parseDocument: htmlparser2.parseDocument,
         selectAll: cssSelect.selectAll,
         selectOne: cssSelect.selectOne,
-        is: cssSelect.is,
+        compile: cssSelect.compile,
         csstree,
         domutils,
       };

.agents/skills/impeccable/scripts/detector/rules/checks.mjs

@@ -721,20 +721,19 @@ function scanCssTextForGlow(content) {
   return results;
 }
 
-// Decorative grid or line-field backgrounds drawn with hairline
+// Decorative two-axis grid backgrounds drawn with hairline
 // linear-gradient layers tiled by a fixed pixel cell. Shared by the HTML
 // pattern pass and the regex source engine so standalone CSS, component
 // styles, and inline styles receive the same coverage. Both signals must
 // co-occur in one declaration block; unrelated rules must not add up across
-// the file. Returns [{ index, snippet }], capped at one finding per source to
-// match the page-level HTML check's existing behavior.
+// the file. A single hairline is a line, divider, or rail, not a grid, even
+// when tiled by a 2D px cell. Returns [{ index, snippet }], capped at one
+// finding per source to match the page-level HTML check's existing behavior.
 function scanCssTextForGridBackground(content) {
   const hairlineRe = /\b\d{1,3}px\s*,\s*transparent\s+\d{1,3}px/gi;
   const invertedHairlineRe = /transparent\s+calc\(100%\s*-\s*\d{1,3}px\)/gi;
   const sizeDeclPxRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\b/i;
-  const sizeDeclPxPairRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\s+\d{1,3}px/i;
   const shorthandPxAnyRe = /\/\s*\d{1,3}px\b/;
-  const shorthandPxPairRe = /\/\s*\d{1,3}px\s+\d{1,3}px/;
   const bgDeclRe = /\bbackground(?:-image)?\s*:\s*([^;{}"']*)/gi;
   const blockRe = /\{([^{}]*)\}|style\s*=\s*"([^"]*)"|style\s*=\s*'([^']*)'/gi;
   let blk;
@@ -751,13 +750,10 @@ function scanCssTextForGridBackground(content) {
     }
     if (hairlineCount === 0) continue;
     const hasPxCell = sizeDeclPxRe.test(block) || shorthandPxAnyRe.test(bgJoined);
-    const hasPxPairCell = sizeDeclPxPairRe.test(block) || shorthandPxPairRe.test(bgJoined);
-    if ((hairlineCount >= 2 && hasPxCell) || hasPxPairCell) {
+    if (hairlineCount >= 2 && hasPxCell) {
       return [{
         index: blk.index,
-        snippet: hairlineCount >= 2
-          ? 'two-axis grid-line gradient background'
-          : 'px-tiled hairline line-field background',
+        snippet: 'two-axis grid-line gradient background',
       }];
     }
   }
@@ -2752,7 +2748,7 @@ function checkElementAIPaletteDOM(el) {
   }
 
   // Check for neon text (vivid cyan/purple color on dark background)
-  const textColor = parseRgb(style.color);
+  const textColor = parseRgb(style.color) || parseAnyColor(style.color);
   if (textColor && hasChroma(textColor, 80)) {
     const hue = getHue(textColor);
     const isAIPalette = (hue >= 160 && hue <= 200) || (hue >= 260 && hue <= 310);

.agents/skills/impeccable/scripts/detector/shared/color.mjs

@@ -103,14 +103,22 @@ function extractColorFunctionTokens(value) {
 function parseGradientColors(bgImage) {
   if (!bgImage || !bgImage.includes('gradient')) return [];
   const colors = [];
+  const tokenSpans = [];
+  let from = 0;
   // Stops arrive in whatever syntax the author wrote and the browser kept.
   // A dark ground painted as `linear-gradient(oklch(...), oklch(...))` used
   // to read as a gradient with no stops at all.
   for (const token of extractColorFunctionTokens(bgImage)) {
+    const start = bgImage.indexOf(token, from);
+    if (start < 0) break;
+    tokenSpans.push({ start, end: start + token.length });
+    from = start + token.length;
     const c = parseAnyColor(token);
     if (c) colors.push(c);
   }
   for (const m of bgImage.matchAll(/#([0-9a-f]{6}|[0-9a-f]{3})\b/gi)) {
+    // Nested hex inside color-mix is an ingredient, not a stop (issue #578).
+    if (tokenSpans.some(s => m.index >= s.start && m.index < s.end)) continue;
     const h = m[1];
     if (h.length === 6) {
       colors.push({ r: parseInt(h.slice(0,2),16), g: parseInt(h.slice(2,4),16), b: parseInt(h.slice(4,6),16), a: 1 });

.agents/skills/impeccable/scripts/doctor.mjs

@@ -33,13 +33,8 @@ import {
   stampProductSchema,
 } from './lib/artifact-schema.mjs';
 import {
-  checkBuildPathUnset,
-  checkConfig,
-  checkDesignSidecar,
+  collectBootFindingGroups,
   checkNativePlatformEvidence,
-  checkProduct,
-  checkProjectRoots,
-  checkSurfaceBriefs,
   designSidecarCandidatesFor,
 } from './lib/staleness.mjs';
 import {
@@ -106,34 +101,30 @@ async function collect(cwd, targetOptions) {
     extractPlatform,
     readFile: safeRead,
   });
+  const bootFindings = collectBootFindingGroups(ctx, {
+    absDesignPath,
+    sidecarCandidates,
+    projectRootPatterns: readProjectRootPatterns(ctx.repoRoot),
+    targetCandidates: workspaceCandidates,
+  });
 
   const findings = [
-    ...checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
-    ...(ctx.product
-      ? checkNativePlatformEvidence({
-          projectRoot,
-          platform: ctx.platform,
-          product: ctx.product,
-          productPath: ctx.productPath,
-        })
-      : []),
-    ...checkDesignSidecar({ designPath: absDesignPath, sidecarCandidates, projectRoot }),
+    ...bootFindings.product,
+    ...bootFindings.nativePlatform,
+    ...bootFindings.designSidecar,
     ...checkDesignDrift({ designPath: absDesignPath, projectRoot }),
     ...checkDesignCoverage({ design: ctx.design, designPath: ctx.designPath, parseDesignMd }),
-    ...checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
-    ...checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
+    ...bootFindings.config,
+    ...bootFindings.buildPath,
     ...checkDetectorIgnores({ projectRoot, knownRuleIds }),
-    ...checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
+    ...bootFindings.surfaceBriefs,
     ...checkHookInstallation({
       projectRoot,
       repoRoot: ctx.repoRoot,
       providerId: IMPECCABLE_PROVIDER_ID,
     }),
     ...checkLegacyLiveState({ projectRoot }),
-    ...checkProjectRoots({
-      patterns: readProjectRootPatterns(ctx.repoRoot),
-      candidates: workspaceCandidates,
-    }),
+    ...bootFindings.projectRoots,
     ...workspaceResult.findings,
   ];
 

.agents/skills/impeccable/scripts/embed-prompt.mjs

@@ -5,6 +5,7 @@
 //   node embed-prompt.mjs <image> --prompt "the prompt text"
 //   node embed-prompt.mjs <image> --prompt-file prompt.txt
 //   node embed-prompt.mjs <image> --read
+//   node embed-prompt.mjs --scan <dir...>   # list rasters missing a prompt; exit 3 when any
 //
 // Formats: PNG (tEXt chunk, keyword "impeccable:prompt"), JPEG (COM segment).
 // WebP and anything else fall back to a `<image>.json` sidecar; --read checks
@@ -21,8 +22,49 @@ const KEYWORD = 'impeccable:prompt';
 const args = process.argv.slice(2);
 const file = args.find(a => !a.startsWith('--'));
 const readMode = args.includes('--read');
+const scanMode = args.includes('--scan');
 const argOf = (name) => { const i = args.indexOf(name); return i !== -1 ? args[i + 1] : null; };
 
+function promptOf(imagePath) {
+  const b = fs.readFileSync(imagePath);
+  let prompt = null;
+  if (b.length > 8 && b.readUInt32BE(0) === 0x89504e47) prompt = readPngText(b);
+  else if (b.length > 3 && b[0] === 0xff && b[1] === 0xd8) prompt = readJpegCom(b);
+  if (prompt == null && fs.existsSync(`${imagePath}.json`)) {
+    try { prompt = JSON.parse(fs.readFileSync(`${imagePath}.json`, 'utf8')).prompt ?? null; } catch { /* stays null */ }
+  }
+  return prompt;
+}
+
+if (scanMode) {
+  const targets = args.filter(a => !a.startsWith('--'));
+  if (targets.length === 0) { console.error('embed-prompt: --scan needs at least one directory'); process.exit(1); }
+  const RASTER = /\.(png|jpe?g|webp)$/i;
+  const rasters = [];
+  const walk = (p, isRoot) => {
+    const stat = fs.statSync(p);
+    if (stat.isDirectory()) {
+      const base = p.replace(/\/+$/, '').split('/').pop();
+      // Skip installed deps and hidden dirs found during the walk, but honor a
+      // hidden dir the caller passed explicitly (e.g. .impeccable/mocks).
+      if (!isRoot && (base === 'node_modules' || base.startsWith('.'))) return;
+      for (const entry of fs.readdirSync(p)) walk(`${p.replace(/\/+$/, '')}/${entry}`, false);
+    } else if (RASTER.test(p)) {
+      rasters.push(p);
+    }
+  };
+  for (const target of targets) {
+    if (!fs.existsSync(target)) { console.error(`embed-prompt: no such path ${target}`); process.exit(1); }
+    walk(target, true);
+  }
+  let missing = 0;
+  for (const raster of rasters) {
+    if (promptOf(raster) == null) { console.log(`MISSING: ${raster}`); missing++; }
+  }
+  console.log(`SCAN: ${rasters.length} raster${rasters.length === 1 ? '' : 's'}, ${missing} missing`);
+  process.exit(missing > 0 ? 3 : 0);
+}
+
 if (!file || !fs.existsSync(file)) { console.error('embed-prompt: image file required'); process.exit(1); }
 
 const buf = fs.readFileSync(file);

.agents/skills/impeccable/scripts/hook-admin.mjs

@@ -75,11 +75,11 @@ const HOOK_MANIFEST_TARGETS = [
     destRel: '.claude/settings.local.json',
     sharedDestRel: '.claude/settings.json',
     manifest: () => ({
-      description: 'Impeccable design detector: immediate-tier checks after Edit/Write/MultiEdit on UI files, full-rule deep pass on Stop.',
+      description: 'Impeccable design detector: immediate-tier checks after Edit/Write on UI files, full-rule deep pass on Stop.',
       hooks: {
         PostToolUse: [
           {
-            matcher: 'Edit|Write|MultiEdit',
+            matcher: 'Edit|Write',
             hooks: [
               {
                 type: 'command',

.agents/skills/impeccable/scripts/hook-lib.mjs

@@ -816,9 +816,9 @@ export function splitFindingsByTier(findings) {
 }
 
 // Whether the per-edit pass for this harness should defer non-immediate
-// findings to a Stop deep pass. Only Claude Code and Codex dispatch our Stop
-// hook; Cursor and GitHub Copilot have no deep pass wired, so deferring for
-// them would silently drop the non-immediate rules entirely.
+// findings to a Stop deep pass. Claude Code, Codex, and Grok Build dispatch
+// our Stop hook; Cursor and GitHub Copilot have no deep pass wired, so
+// deferring for them would silently drop the non-immediate rules entirely.
 export function perEditTieringActive(config, harness) {
   if (harness === 'cursor' || harness === 'github') return false;
   return (config?.perEditRules || DEFAULT_CONFIG.perEditRules) !== 'all';
@@ -1251,18 +1251,50 @@ export function resolveHarness(env = {}, event = null) {
   const explicit = env?.IMPECCABLE_HOOK_HARNESS;
   if (explicit === 'cursor') return 'cursor';
   if (explicit === 'github') return 'github';
-  if (explicit === 'claude' || explicit === 'codex') return 'claude';
-  // GitHub Copilot's postToolUse event uses camelCase `toolName`/`toolArgs` and
-  // has no `tool_name`/`tool_input`. That shape is the discriminator.
+  if (explicit === 'grok') return 'grok';
+  if (explicit === 'claude') return 'claude';
+  if (explicit === 'codex') return 'codex';
+  // Grok Build sends camelCase `toolName`/`toolInput`/`hookEventName` and no
+  // snake_case pair. GitHub Copilot sends camelCase `toolName`/`toolArgs`.
+  // Check Grok first: the old GitHub heuristic (`toolName` and no
+  // `tool_input`) also matches Grok, which is how live PostToolUse was
+  // classified as Copilot and then skipped with no-file-path (#646).
+  if (looksLikeGrokEnvelope(event)) return 'grok';
   if (event && typeof event === 'object'
     && (typeof event.toolName === 'string' || event.toolArgs !== undefined)
     && event.tool_name === undefined && event.tool_input === undefined) {
     return 'github';
   }
   if (typeof event?.conversation_id === 'string' && event.conversation_id) return 'cursor';
+  // Codex turn-scoped events carry `turn_id`. Claude Code does not. Detecting
+  // it here means an already-installed Codex hook emits the Codex Stop
+  // contract without rewriting the hook command to set IMPECCABLE_HOOK_HARNESS.
+  // https://developers.openai.com/codex/hooks#stop
+  if (typeof event?.turn_id === 'string' && event.turn_id) return 'codex';
   return 'claude';
 }
 
+function looksLikeGrokEnvelope(event) {
+  if (!event || typeof event !== 'object') return false;
+  if (event.hook_event_name !== undefined
+    || event.tool_name !== undefined
+    || event.tool_input !== undefined) {
+    return false;
+  }
+  if (event.toolArgs !== undefined) return false;
+  if (typeof event.hookEventName === 'string') return true;
+  return typeof event.toolName === 'string' && event.toolInput !== undefined;
+}
+
+// Stop arrives as Claude's `hook_event_name: "Stop"` or Grok Build's
+// `hookEventName: "stop"`. hook.mjs routes on the raw stdin, before any
+// normalize, so both casings must match here.
+export function isStopEvent(event) {
+  if (!event || typeof event !== 'object') return false;
+  const name = event.hook_event_name || event.hookEventName;
+  return typeof name === 'string' && name.toLowerCase() === 'stop';
+}
+
 // GitHub Copilot's postToolUse payload is
 //   { sessionId, timestamp, cwd, toolName, toolArgs, toolResult }
 // mapped onto the internal `{ tool_name, tool_input, cwd, session_id }` shape.
@@ -1354,9 +1386,36 @@ function normalizeGitHubEvent(event, projectCwd) {
   };
 }
 
+// Grok Build 1.0.5 (captured 2026-08-24) sends camelCase `toolName` /
+// `toolInput` / `sessionId` / `stopHookActive`, plus `cwd` alongside a
+// trailing-slashed `workspaceRoot` (every consumer path.resolve()s, so no
+// stripping here). Only the fields the hook reads are copied; the event
+// name stays camelCase because routing already happened on the raw stdin
+// (isStopEvent) and nothing downstream reads `hook_event_name`.
+function normalizeGrokEvent(event, projectCwd) {
+  const cwd = event.cwd || event.workspaceRoot || envProjectDir(projectCwd) || projectCwd;
+  const sessionId = event.sessionId || event.session_id || 'unknown';
+  const rawInput = event.toolInput ?? event.tool_input;
+  const toolInput = rawInput && typeof rawInput === 'object' && !Array.isArray(rawInput)
+    ? { ...rawInput }
+    : {};
+  const out = {
+    ...event,
+    cwd,
+    session_id: sessionId,
+    tool_name: event.toolName || event.tool_name || null,
+    tool_input: toolInput,
+  };
+  if (event.stopHookActive !== undefined && event.stop_hook_active === undefined) {
+    out.stop_hook_active = event.stopHookActive;
+  }
+  return out;
+}
+
 export function normalizeHookEvent(event, projectCwd, harness = 'claude') {
   if (!event || typeof event !== 'object') return event;
   if (harness === 'github') return normalizeGitHubEvent(event, projectCwd);
+  if (harness === 'grok') return normalizeGrokEvent(event, projectCwd);
   if (harness !== 'cursor') return event;
 
   const cwd = event.cwd
@@ -1959,7 +2018,15 @@ export async function runHook({ stdinJson, env = {}, cwd = process.cwd(), now =
       // findings stop being remembered and a reintroduced one reads as fresh.
       // Only the immediate tier is remembered: a deferred finding the per-edit
       // pass never reported must still read as fresh to the Stop deep pass.
-      rememberFindings(cache, sessionId, filePath, immediate);
+      //
+      // Grok ignores PostToolUse stdout, so Stop is the user-visible pass.
+      // Remembering here would dedupe those findings out of Stop. Touch the
+      // file so Stop has it, and leave the finding list empty.
+      if (harness === 'grok') {
+        touchFile(cache, sessionId, filePath);
+      } else {
+        rememberFindings(cache, sessionId, filePath, immediate);
+      }
       cacheDirty = true;
 
       if (fresh.length > 0) {
@@ -2163,8 +2230,11 @@ export const STOP_MAX_FILES = 20;
  *   { exitCode, stdout, audit, emission? }
  *
  * Never throws; exits silent (and fast) when the session touched no UI
- * files. Output uses the Stop hookSpecificOutput channel: additionalContext
- * is delivered to the model and the conversation continues so it can act.
+ * files. Output goes out on the harness's Stop continuation channel: Claude
+ * Code and Grok Build read hookSpecificOutput.additionalContext, Codex takes
+ * a decision: "block" whose reason becomes the continuation prompt. Either
+ * way the findings reach the model and the conversation continues so it
+ * can act.
  */
 export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), now = Date.now, detector } = {}) {
   const audit = { ts: new Date(now()).toISOString(), event: 'Stop' };
@@ -2191,22 +2261,36 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
       return result({ skipped: 'stdin-empty', durationMs: Date.now() - started });
     }
 
-    // Claude Code's Stop-hook contract: `stop_hook_active` is true when this
-    // hook is being re-invoked only because a prior invocation kept the turn
-    // alive (here, via hookSpecificOutput.additionalContext). Re-scanning and
-    // re-blocking now would loop until Claude Code's consecutive-block cap
-    // force-ends the turn (issue #400). The prior fire already surfaced the
-    // findings; whether to act on them is the agent's call. Exit fast with no
-    // output before any scan. Only Claude Code sends this field; other
-    // harnesses omit it, so the strict `=== true` is a no-op for them. This
-    // guard makes the loop impossible regardless of the finding cache key's
-    // line-number sensitivity (out of scope here; see findingCacheKey).
+    const harness = resolveHarness(env, event);
+    audit.harness = harness;
+    event = normalizeHookEvent(event, cwd, harness);
+
+    // Stop-hook re-entry guard: `stop_hook_active` is true when this hook is
+    // being re-invoked only because a prior invocation kept the turn alive
+    // (Claude Code via hookSpecificOutput.additionalContext, Codex via a
+    // decision: "block" continuation). Re-scanning and re-blocking now could
+    // loop (issue #400). The prior fire already surfaced the findings;
+    // whether to act on them is the agent's call. Exit fast with no output
+    // before any scan. Claude Code and Codex both send this field: Codex
+    // mirrors the Claude contract (StopCommandInput in
+    // codex-rs/hooks/src/schema.rs) and latches it true for the rest of the
+    // turn once a block is honored (codex-rs/core/src/session/turn.rs). Grok
+    // sends `stopHookActive`, copied onto the snake_case field above. Cursor
+    // and GitHub Copilot omit the field, so the strict `=== true` is a no-op
+    // for them. The guard makes the loop impossible regardless of the finding
+    // cache key's line-number sensitivity (out of scope here; see
+    // findingCacheKey).
     if (event.stop_hook_active === true) {
       return result({ skipped: 'stop-hook-active', durationMs: Date.now() - started });
     }
 
-    const harness = resolveHarness(env, event);
-    audit.harness = harness;
+    // Grok fires Stop twice: `end_turn` (the gate that can inject
+    // additionalContext) then an observe-only `shutdown`. A second deep
+    // pass would re-emit the same findings. Claude omits `reason`; only
+    // skip when Grok named a reason that is not end_turn.
+    if (harness === 'grok' && typeof event.reason === 'string' && event.reason !== 'end_turn') {
+      return result({ skipped: 'stop-reason', reason: event.reason, durationMs: Date.now() - started });
+    }
 
     // A Stop event carries no file, so the session cwd is the project.
     // Umbrella-dir launches keyed their per-edit cache to the edited file's
@@ -2241,6 +2325,7 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
 
     const freshGroups = [];
     let scanned = 0;
+    let cacheDirty = false;
     for (const filePath of touched) {
       if (scanned >= STOP_MAX_FILES) break;
       if (hasPathTraversal(filePath) || SENSITIVE_PATH.test(filePath)) continue;
@@ -2261,29 +2346,39 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
       try { content = fs.readFileSync(filePath, 'utf-8'); } catch { continue; }
 
       let findings;
+      let detectorThrew = false;
       const useHtmlEngine = configuredExt
         ? configuredExt.engine === 'html'
         : (ext === '.html' || ext === '.htm');
 
       if (useHtmlEngine && typeof det.detectHtml === 'function') {
-        try { findings = await det.detectHtml(filePath, scanOptions); } catch { findings = []; }
+        try { findings = await det.detectHtml(filePath, scanOptions); } catch { findings = []; detectorThrew = true; }
       } else {
-        try { findings = await det.detectText(content, filePath, scanOptions); } catch { findings = []; }
+        try { findings = await det.detectText(content, filePath, scanOptions); } catch { findings = []; detectorThrew = true; }
       }
 
+      // A detector failure tells us nothing about the file. Leave whatever
+      // was remembered alone rather than recording an empty scan as truth.
+      if (detectorThrew) continue;
+
       // Full rule set: no tier split here. Config/inline ignores still apply,
       // and the session dedupe drops everything the per-edit pass (or an
       // earlier Stop pass) already surfaced.
       const filtered = filterFindings(findings || [], content, ext, config);
       const fresh = dedupeAgainstCache(filtered, cache, sessionId, filePath);
+      // Sync to the live scan, including empty. Remembering only `fresh`
+      // (or skipping the write on a clean Stop) left stale keys in place, so
+      // a finding that was fixed and later reintroduced never fired again.
+      rememberFindings(cache, sessionId, filePath, filtered);
+      cacheDirty = true;
       if (fresh.length > 0) {
-        rememberFindings(cache, sessionId, filePath, fresh);
         freshGroups.push({ filePath, findings: fresh });
       }
     }
     audit.scannedFiles = scanned;
 
     if (freshGroups.length === 0) {
+      if (cacheDirty) persistCache(projectCwd, cache);
       return result({ emitted: false, skipped: 'stop-clean', durationMs: Date.now() - started });
     }
 
@@ -2300,8 +2395,8 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
     );
     commitFooterShown(cache, sessionId, text);
 
-    // Fresh findings earn the cache write so the next Stop fire is silent
-    // unless new issues appear; the notice flags ride along.
+    // Persist the live finding set so the next Stop fire is silent unless
+    // new issues appear; the notice flags ride along.
     persistCache(projectCwd, cache);
     return {
       exitCode: 0,
@@ -2337,6 +2432,15 @@ export function payload(text, eventName = 'PostToolUse', harness = 'claude') {
   if (harness === 'github') {
     return JSON.stringify({ additionalContext: text });
   }
+  // Codex shares Claude Code's PostToolUse additional-context shape, but its
+  // Stop schema rejects unknown fields. Findings that should continue the
+  // turn must be a top-level blocking decision.
+  // https://developers.openai.com/codex/hooks#stop (schema of record:
+  // codex-rs/hooks/src/schema.rs, StopCommandOutputWire)
+  if (harness === 'codex' && eventName === 'Stop') {
+    if (!String(text ?? '').trim()) return '';
+    return JSON.stringify({ decision: 'block', reason: text });
+  }
   return JSON.stringify({
     hookSpecificOutput: { hookEventName: eventName, additionalContext: text },
   });

.agents/skills/impeccable/scripts/hook.mjs

@@ -2,15 +2,17 @@
 /**
  * Impeccable design hook — PostToolUse + Stop entry point.
  *
- * Reads the Claude Code / Codex / Cursor hook event from stdin and routes by
- * `hook_event_name`:
+ * Reads the Claude Code / Codex / Cursor / Grok Build hook event from stdin
+ * and routes by Stop vs everything else. Claude uses `hook_event_name:
+ * "Stop"`; Grok uses `hookEventName: "stop"`.
  *
  *   - PostToolUse: runs the immediate-tier detector rules against the touched
  *     file and emits a system reminder via
- *     `hookSpecificOutput.additionalContext` when findings exist.
+ *     `hookSpecificOutput.additionalContext` when findings exist. Grok
+ *     discards that stdout; the scan still warms the session cache for Stop.
  *   - Stop: runs the FULL detector rule set over every UI file touched this
  *     session (the deep pass), deduped against what the per-edit pass already
- *     surfaced, and emits once via the Stop additionalContext channel.
+ *     surfaced, and emits once via the harness-specific continuation channel.
  *
  * Contract: never break a turn. Always exit 0. Clean files emit a small ack
  * unless quiet mode is enabled; a clean Stop pass is silent.
@@ -19,7 +21,7 @@
  * subprocess. This file is the thin stdin/stdout adapter.
  */
 
-import { runHook, runStopHook, writeAuditLog } from './hook-lib.mjs';
+import { runHook, runStopHook, writeAuditLog, isStopEvent } from './hook-lib.mjs';
 
 async function readStdin() {
   if (process.stdin.isTTY) return '';
@@ -28,10 +30,9 @@ async function readStdin() {
   return Buffer.concat(chunks).toString('utf-8');
 }
 
-function isStopEvent(stdinJson) {
+function stdinIsStop(stdinJson) {
   try {
-    const event = JSON.parse(stdinJson);
-    return event && typeof event === 'object' && event.hook_event_name === 'Stop';
+    return isStopEvent(JSON.parse(stdinJson));
   } catch {
     // Malformed stdin falls through to runHook, which audits the skip.
     return false;
@@ -48,7 +49,7 @@ async function main() {
   let stdinJson = '';
   try { stdinJson = await readStdin(); } catch { /* fall through */ }
 
-  const run = isStopEvent(stdinJson) ? runStopHook : runHook;
+  const run = stdinIsStop(stdinJson) ? runStopHook : runHook;
   const result = await run({
     stdinJson,
     env: inheritedEnv,

.agents/skills/impeccable/scripts/lib/design-parser.mjs

@@ -196,9 +196,6 @@ function parseScalar(raw) {
 
 const HEX_RE = /#[0-9a-fA-F]{3,8}\b/g;
 const OKLCH_RE = /oklch\([^)]+\)/gi;
-const RGBA_RE = /rgba?\([^)]+\)/gi;
-const BOX_SHADOW_RE = /(?:box-shadow:\s*)?((?:-?\d[\w\d\s\-.,/()#%]*)+)/;
-const NAMED_RULE_RE = /\*\*(The [^*]+?Rule)\.\*\*\s*(.+)/;
 
 // ---------- Section splitting ----------
 
@@ -550,36 +547,6 @@ function detectFormat(v) {
   return 'unknown';
 }
 
-function scanInlineColors(lines) {
-  const out = [];
-  for (const line of lines) {
-    if (!/^\s*[-*]\s/.test(line)) continue;
-    const trimmed = line.replace(/^\s*[-*]\s+/, '');
-    const color = parseColorBullet(trimmed);
-    if (color) out.push(color);
-  }
-  return out;
-}
-
-function parseStitchInlineGroups(lines) {
-  // Stitch writes: `*   **Primary (`#00478d` to `#005eb8`):** Use for "..."`
-  // Each bullet IS its own role. Group them under the spoken role name.
-  const out = [];
-  for (const line of lines) {
-    if (!/^\s*[-*]\s/.test(line)) continue;
-    const trimmed = line.replace(/^\s*[-*]\s+/, '').trim();
-    const m = trimmed.match(
-      /^\*\*([A-Z][a-zA-Z]+)\s*\(([^)]+)\):\*\*\s*(.*)$/
-    );
-    if (m) {
-      const role = m[1];
-      const color = buildColor(role, m[2], m[3]);
-      out.push({ role, colors: [color] });
-    }
-  }
-  return out;
-}
-
 function extractTypography(section) {
   if (!section) return null;
   const text = section.lines.join('\n');

.agents/skills/impeccable/scripts/lib/staleness.mjs

@@ -488,41 +488,46 @@ export function describeWorkspaceContext(candidates = []) {
 // ─── Tier 1 orchestration ──────────────────────────────────────────────────
 
 /**
- * Everything a boot can afford. `ctx` is the loadContext result; `extras`
- * carries values the caller already computed so nothing is recomputed here.
+ * Everything a boot can afford, grouped by artifact so deeper reports can
+ * interleave their own checks without rebuilding this policy. `ctx` is the
+ * loadContext result; `extras` carries values the caller already computed so
+ * nothing is recomputed here.
  */
-export function collectBootFindings(ctx, extras = {}) {
-  if (!ctx) return [];
+export function collectBootFindingGroups(ctx, extras = {}) {
+  if (!ctx) return {};
   const projectRoot = ctx.projectRoot || process.cwd();
-  const absProductPath = extras.absProductPath || null;
   const absDesignPath = extras.absDesignPath || null;
 
-  return [
-    ...checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
+  return {
+    product: checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
     // Only checked once a PRODUCT.md exists. Without one the boot already
     // emits NO_PRODUCT_MD and routes into init, which asks for the platform
     // directly; a second signal saying the same thing is noise.
-    ...(ctx.product
+    nativePlatform: ctx.product
       ? checkNativePlatformEvidence({
           projectRoot,
           platform: ctx.platform,
           product: ctx.product,
           productPath: ctx.productPath,
         })
-      : []),
-    ...checkDesignSidecar({
+      : [],
+    designSidecar: checkDesignSidecar({
       designPath: absDesignPath,
       sidecarCandidates: extras.sidecarCandidates || [],
       projectRoot,
     }),
-    ...checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
-    ...checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
-    ...checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
-    ...(extras.projectRootPatterns
+    config: checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
+    buildPath: checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
+    surfaceBriefs: checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
+    projectRoots: extras.projectRootPatterns
       ? checkProjectRoots({
           patterns: extras.projectRootPatterns,
           candidates: extras.targetCandidates || [],
         })
-      : []),
-  ];
+      : [],
+  };
+}
+
+export function collectBootFindings(ctx, extras = {}) {
+  return Object.values(collectBootFindingGroups(ctx, extras)).flat();
 }

.agents/skills/impeccable/scripts/lib/surface-briefs.mjs

@@ -8,6 +8,12 @@ export function getSurfaceBriefDir(projectRoot) {
   return path.join(projectRoot, '.impeccable', 'surfaces');
 }
 
+function normalizeRouteTarget(route) {
+  if (!route.startsWith('/') || route.includes('..')) return null;
+  const normalized = route.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
+  return `route:${normalized}`;
+}
+
 export function normalizeSurfaceTarget(target, { projectRoot = process.cwd() } = {}) {
   if (!target || typeof target !== 'string' || !target.trim()) return null;
   const trimmed = target.trim();
@@ -21,21 +27,13 @@ export function normalizeSurfaceTarget(target, { projectRoot = process.cwd() } =
       return null;
     }
   }
-  if (/^route:/i.test(trimmed)) {
-    const route = trimmed.slice(trimmed.indexOf(':') + 1).trim();
-    if (!route.startsWith('/') || route.includes('..')) return null;
-    const normalizedRoute = route.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
-    return `route:${normalizedRoute}`;
-  }
-  if (trimmed === '/') return 'route:/';
+  if (/^route:/i.test(trimmed)) return normalizeRouteTarget(trimmed.slice(trimmed.indexOf(':') + 1).trim());
+  if (trimmed === '/') return normalizeRouteTarget(trimmed);
   if (trimmed.startsWith('/')) {
     const absolute = path.resolve(trimmed);
     const relativeToProject = path.relative(projectRoot, absolute);
     const isProjectFile = relativeToProject && !relativeToProject.startsWith('..') && !path.isAbsolute(relativeToProject);
-    if (!isProjectFile && !fs.existsSync(absolute) && !trimmed.includes('..')) {
-      const normalizedRoute = trimmed.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
-      return `route:${normalizedRoute}`;
-    }
+    if (!isProjectFile && !fs.existsSync(absolute)) return normalizeRouteTarget(trimmed);
   }
   const abs = path.isAbsolute(trimmed) ? trimmed : path.resolve(projectRoot, trimmed);
   const rel = path.relative(projectRoot, abs);

.agents/skills/impeccable/scripts/live-browser.js

@@ -4902,6 +4902,13 @@
     saveSession();
   }
 
+  function completeParameterGenerationIfReady() {
+    if (expectedVariants <= 0 || arrivedVariants < expectedVariants) return;
+    if (parameterGenerationState === 'pending' || parameterGenerationState === 'loading') {
+      completeParameterPublication();
+    }
+  }
+
   function toggleTunePopover() {
     if (pendingApplyInFlight) { showManualApplyBusyToast(); return; }
     if (tuneOpen) { closeTunePopover(); return; }
@@ -5796,7 +5803,7 @@
         setLiveState('CYCLING');
         showOrUpdateCyclingBar();
         saveSession();
-        if (parameterGenerationState === 'loading') completeParameterPublication();
+        completeParameterGenerationIfReady();
         return;
       }
 
@@ -5884,7 +5891,7 @@
       refreshParamsPanel();
       positionBar();
       saveSession();
-      if (parameterGenerationState === 'loading') completeParameterPublication();
+      completeParameterGenerationIfReady();
       console.log('[impeccable] Mounted ' + arrivedVariants + ' ' + manifest.framework + ' component variants.');
     } catch (err) {
       console.error('[impeccable] Failed to mount component-preview variants:', err);
@@ -6329,7 +6336,7 @@
         refreshParamsPanel();
         positionBar();
         saveSession();
-        if (parameterGenerationState === 'loading') completeParameterPublication();
+        completeParameterGenerationIfReady();
         console.log('[impeccable] Injected ' + arrivedVariants + ' variants from source file.');
       })
       .catch(err => {
@@ -6836,6 +6843,7 @@
 
       const expected = parseInt(wrapper.dataset.impeccableVariantCount || '0');
       if (expected > 0) expectedVariants = expected;
+      completeParameterGenerationIfReady();
 
       if (arrivedVariants > 0) {
         setLiveState('CYCLING');

.agents/skills/impeccable/scripts/live-commit-manual-edits.mjs

@@ -944,8 +944,42 @@ export async function commitManualEdits({
     };
   }
 
+  const repairContext = {
+    batch,
+    cwd,
+    pageUrl,
+    count,
+    provider,
+    env,
+    timeoutMs,
+    applyBatchToSource,
+    chatAvailable,
+    transactionId,
+  };
+
   const baseRollbackScope = collectApplyOwnedFiles(batch, cwd);
   const rollbackSnapshot = snapshotRollbackFiles(cwd, baseRollbackScope);
+  const failWithRollback = ({
+    scope = baseRollbackScope,
+    extraFiles = [],
+    failed,
+    files = [],
+    details = {},
+  }) => {
+    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, extraFiles, scope);
+    return {
+      applied: [],
+      failed,
+      files,
+      cleared: 0,
+      count,
+      pageUrl,
+      ...details,
+      rolledBackFiles: rollback.rolledBackFiles,
+      rollbackFailures: rollback.rollbackFailures,
+      ...countByPage(cwd),
+    };
+  };
   let result;
   try {
     result = repairOnly
@@ -965,42 +999,27 @@ export async function commitManualEdits({
           chatAvailable,
         });
   } catch (err) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, [], baseRollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
       failed: batch.entries.map((entry) => ({
         id: entry.id,
         reason: err.message || String(err),
         candidates: candidatesForEntry(batch, entry.id),
       })),
-      files: [],
-      cleared: 0,
-      count,
-      pageUrl,
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+    });
   }
 
   if (result.status === 'error') {
     const rollbackScope = collectApplyOwnedFiles(batch, cwd, result.files || []);
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
     const failed = normalizeFailedEntries(batch, result, result.message || 'AI copy edit failed');
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: failed.length > 0
         ? failed
         : verificationFailuresForEntries(batch, batch.entries, result.message || 'AI copy edit failed'),
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const reportedAppliedIds = uniqueStrings(result.appliedEntryIds || []);
@@ -1013,72 +1032,44 @@ export async function commitManualEdits({
   const conflictingAppliedIds = reportedAppliedIds.filter((id) => failedIds.has(id));
 
   if (conflictingAppliedIds.length > 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
     const conflictingEntries = batch.entries.filter((entry) => conflictingAppliedIds.includes(entry.id));
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: [
         ...verificationFailuresForEntries(batch, conflictingEntries, 'conflicting_apply_result'),
         ...aiFailed.filter((item) => !conflictingAppliedIds.includes(item.id)),
       ],
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const unreportedFiles = unreportedChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
   if (unreportedFiles.length > 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], [...rollbackScope, ...unreportedFiles]);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: [...rollbackScope, ...unreportedFiles],
+      extraFiles: result.files || [],
       failed: verificationFailuresForEntries(batch, batch.entries, 'unreported_source_changes', { files: unreportedFiles }),
       files: result.files || [],
-      unreportedFiles,
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { unreportedFiles, notes: result.notes || [] },
+    });
   }
 
   if (result.status === 'done' && reportedAppliedIds.length === 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: verificationFailuresForEntries(batch, batch.entries, 'missing_applied_entry_ids'),
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const reportedAppliedEntries = batch.entries.filter((entry) => reportedAppliedIds.includes(entry.id));
   if (reportedAppliedIds.length > 0 && reportedFiles.length === 0) {
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: reportedAppliedIds,
       files: result.files || [],
       failed: aiFailed,
@@ -1089,21 +1080,10 @@ export async function commitManualEdits({
     });
   }
 
-  const verifiedAppliedIds = [];
-  const verificationFailed = [];
-  for (const entry of reportedAppliedEntries) {
-    const failures = verifyAppliedEntry({ batch, entry, reportedFiles, cwd });
-    if (failures.length === 0) {
-      verifiedAppliedIds.push(entry.id);
-    } else {
-      verificationFailed.push({
-        id: entry.id,
-        reason: 'source_verification_failed',
-        failures,
-        candidates: candidatesForEntry(batch, entry.id),
-      });
-    }
-  }
+  const {
+    verifiedIds: verifiedAppliedIds,
+    failed: verificationFailed,
+  } = verifyEntriesAfterRepair({ batch, appliedEntryIds: reportedAppliedIds, files: reportedFiles, cwd });
   const unreportedEntries = result.status === 'done' || result.status === 'partial'
     ? batch.entries.filter((entry) => !reportedAppliedIds.includes(entry.id) && !aiFailed.some((item) => item.id === entry.id))
     : [];
@@ -1133,37 +1113,22 @@ export async function commitManualEdits({
         reason: 'rolled_back_due_to_failed_entry_source_changed',
         candidates: candidatesForEntry(batch, entry.id),
       }));
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: [
         ...leakedUnapplied,
         ...failed.filter((item) => !leakedIds.has(item.id)),
         ...rolledBackVerified,
       ],
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      notes: result.notes || [],
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   if (verificationFailed.length > 0) {
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: reportedAppliedIds,
       files: result.files || [],
       failed: nonRepairFailed,
@@ -1180,16 +1145,7 @@ export async function commitManualEdits({
       ? reportedAppliedEntries.filter((entry) => verifiedAppliedIds.includes(entry.id))
       : batch.entries;
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: verifiedAppliedIds.length > 0
         ? verifiedAppliedIds
         : postCheckEntries.map((entry) => entry.id).filter(Boolean),

.agents/skills/impeccable/scripts/live-poll.mjs

@@ -238,10 +238,9 @@ export async function completeAcceptHandling(event, base, token) {
     });
   } catch (err) {
     event._completionAck = { ok: false, error: err.message };
+    return event;
   }
-  if (!event._completionAck) {
-    event._completionAck = completionAckForAcceptResult(event.id, completionType, event._acceptResult);
-  }
+  event._completionAck = completionAckForAcceptResult(event.id, completionType, event._acceptResult);
   return event;
 }
 
@@ -269,9 +268,11 @@ export function printPollEvent(event) {
   // Situational plumbing rides with the event itself: `_instructions` is the
   // authoritative next step, with real ids and paths substituted, so the
   // reference doc can stay lean and can never drift from script behavior.
-  if (event && typeof event === 'object' && !event._instructions) {
+  // A wire-supplied value must never win over the locally generated one.
+  if (event && typeof event === 'object') {
     const instructions = instructionsForEvent(event, { scriptsPath: SELF_DIR });
     if (instructions) event._instructions = instructions;
+    else delete event._instructions;
   }
   console.log(JSON.stringify(event));
 }

.agents/skills/impeccable/scripts/live-server.mjs

@@ -181,8 +181,16 @@ function chatAgentLikelyActive() {
 // cap at 10 MB to guard against runaway writes from a misbehaving client.
 const MAX_ANNOTATION_BYTES = 10 * 1024 * 1024;
 
+const POLLER_OWNED_EVENT_FIELDS = ['_instructions', '_completionAck', '_acceptResult'];
+
+function stripPollerOwnedEventFields(event) {
+  if (!event || typeof event !== 'object') return;
+  for (const key of POLLER_OWNED_EVENT_FIELDS) delete event[key];
+}
+
 function enqueueEvent(event) {
   if (!event) return;
+  stripPollerOwnedEventFields(event);
   // Dedupe by (session, type), except mount failures, which are per-variant:
   // variant 2 failing must not be swallowed because variant 1's failure is
   // still queued.
@@ -936,15 +944,23 @@ function createRequestHandler({ detectScript, liveScriptParts }) {
       const filePath = url.searchParams.get('path');
       if (!filePath || filePath.includes('..')) { res.writeHead(400); res.end('Bad path'); return; }
       const absPath = path.resolve(process.cwd(), filePath);
-      // Confine to the project root. A bare `startsWith(cwd)` string check lets a
-      // sibling dir whose name extends the root name (projeto -> projeto-backup)
-      // slip through; compare on the relative path instead (same pattern as
-      // sessionFileMetadataFromPollReply below). An empty rel means the request
-      // resolved to the root directory itself, which this file route never serves.
-      const rel = path.relative(process.cwd(), absPath);
+      let realRoot, realTarget;
+      try {
+        realRoot = fs.realpathSync(process.cwd());
+        realTarget = fs.realpathSync(absPath);
+      } catch {
+        res.writeHead(404); res.end('File not found'); return;
+      }
+      // Confine to the project root after symlink resolution. A bare
+      // `startsWith(cwd)` string check lets a sibling dir whose name extends the
+      // root name (projeto -> projeto-backup) slip through; compare on the
+      // relative path instead (same pattern as sessionFileMetadataFromPollReply
+      // below). An empty rel means the request resolved to the root directory
+      // itself, which this file route never serves.
+      const rel = path.relative(realRoot, realTarget);
       if (!rel || rel.startsWith('..') || path.isAbsolute(rel)) { res.writeHead(403); res.end('Forbidden'); return; }
       let content;
-      try { content = fs.readFileSync(absPath, 'utf-8'); }
+      try { content = fs.readFileSync(realTarget, 'utf-8'); }
       catch { res.writeHead(404); res.end('File not found'); return; }
       res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8' });
       res.end(content);
@@ -1026,6 +1042,7 @@ function createRequestHandler({ detectScript, liveScriptParts }) {
           res.end(JSON.stringify({ error }));
           return;
         }
+        stripPollerOwnedEventFields(msg);
         if (msg.type === 'agent_phase') {
           recordAgentPhase(msg.id, msg.phase, {
             ...(Number.isFinite(msg.durationMs) ? { durationMs: msg.durationMs } : {}),

.agents/skills/impeccable/scripts/live/sveltekit-adapter.mjs

@@ -11,6 +11,8 @@ import crypto from 'node:crypto';
 import fs from 'node:fs';
 import path from 'node:path';
 
+import { firstExistingFile, hasAnyDependency } from './frameworks/detect-utils.mjs';
+
 export const SVELTE_LIVE_ROOT_COMPONENT = 'src/lib/impeccable/ImpeccableLiveRoot.svelte';
 export const SVELTE_LAYOUT_MARKER_OPEN = '<!-- impeccable-live-svelte-start -->';
 export const SVELTE_LAYOUT_MARKER_CLOSE = '<!-- impeccable-live-svelte-end -->';
@@ -45,11 +47,17 @@ export function detectSvelteKitProject(cwd = process.cwd(), config = null) {
     && fileIncludes(path.join(cwd, appHtml), '%sveltekit.head%');
   if (!hasTemplateMarkers) return null;
 
-  const hasSvelteConfig = fs.existsSync(path.join(cwd, 'svelte.config.js'))
-    || fs.existsSync(path.join(cwd, 'svelte.config.mjs'))
-    || fs.existsSync(path.join(cwd, 'svelte.config.cjs'))
-    || fs.existsSync(path.join(cwd, 'svelte.config.ts'));
-  const hasKitPackage = packageHasSvelteKit(cwd);
+  const hasSvelteConfig = Boolean(firstExistingFile(cwd, [
+    'svelte.config.js',
+    'svelte.config.mjs',
+    'svelte.config.cjs',
+    'svelte.config.ts',
+  ]));
+  const hasKitPackage = hasAnyDependency(cwd, [
+    '@sveltejs/kit',
+    '@sveltejs/vite-plugin-svelte',
+    'svelte',
+  ]);
   if (!hasSvelteConfig && !hasKitPackage) return null;
 
   return {
@@ -260,36 +268,16 @@ function findSvelteKitAppHtml(cwd, config) {
 }
 
 function findSvelteKitLayout(cwd) {
-  const candidates = [
+  return firstExistingFile(cwd, [
     'src/routes/+layout.svelte',
     'src/routes/(app)/+layout.svelte',
-  ];
-  for (const rel of candidates) {
-    if (fs.existsSync(path.join(cwd, rel))) return rel;
-  }
-  return 'src/routes/+layout.svelte';
+  ]) || 'src/routes/+layout.svelte';
 }
 
 function defaultSvelteLayout() {
   return `<script>\n  let { children } = $props();\n</script>\n\n{@render children?.()}\n`;
 }
 
-function packageHasSvelteKit(cwd) {
-  const file = path.join(cwd, 'package.json');
-  if (!fs.existsSync(file)) return false;
-  try {
-    const pkg = JSON.parse(fs.readFileSync(file, 'utf-8'));
-    const deps = {
-      ...(pkg.dependencies || {}),
-      ...(pkg.devDependencies || {}),
-      ...(pkg.peerDependencies || {}),
-    };
-    return Boolean(deps['@sveltejs/kit'] || deps['@sveltejs/vite-plugin-svelte'] || deps.svelte);
-  } catch {
-    return false;
-  }
-}
-
 function fileIncludes(file, text) {
   try {
     return fs.readFileSync(file, 'utf-8').includes(text);

.agents/skills/impeccable/scripts/live/tanstack-adapter.mjs

@@ -19,6 +19,8 @@
 
 import fs from 'node:fs';
 import path from 'node:path';
+
+import { firstExistingFile, hasAnyDependency } from './frameworks/detect-utils.mjs';
 import { buildLiveScriptSrc } from './frameworks/script-src.mjs';
 
 export const TANSTACK_MARKER_OPEN = '{/* impeccable-live-tanstack-start */}';
@@ -42,8 +44,8 @@ const START_PACKAGES = [
 ];
 
 export function detectTanStackStartProject(cwd = process.cwd()) {
-  if (!packageHasTanStackStart(cwd)) return null;
-  const rootRoute = findRootRouteFile(cwd);
+  if (!hasAnyDependency(cwd, START_PACKAGES)) return null;
+  const rootRoute = firstExistingFile(cwd, ROOT_ROUTE_CANDIDATES);
   if (!rootRoute) return null;
 
   const ext = path.extname(rootRoute);
@@ -218,29 +220,6 @@ function isManagedComponent(content) {
   return String(content || '').includes('impeccable-live-tanstack');
 }
 
-function findRootRouteFile(cwd) {
-  for (const rel of ROOT_ROUTE_CANDIDATES) {
-    if (fs.existsSync(path.join(cwd, rel))) return rel;
-  }
-  return null;
-}
-
-function packageHasTanStackStart(cwd) {
-  const file = path.join(cwd, 'package.json');
-  if (!fs.existsSync(file)) return false;
-  try {
-    const pkg = JSON.parse(fs.readFileSync(file, 'utf-8'));
-    const deps = {
-      ...(pkg.dependencies || {}),
-      ...(pkg.devDependencies || {}),
-      ...(pkg.peerDependencies || {}),
-    };
-    return START_PACKAGES.some((name) => Boolean(deps[name]));
-  } catch {
-    return false;
-  }
-}
-
 function relativeImportSpecifier(fromFile, toFile) {
   const rel = path.posix.relative(
     path.posix.dirname(fromFile.split(path.sep).join('/')),

.agents/skills/impeccable/scripts/serve-question.mjs

@@ -95,9 +95,16 @@
  *   --stop --key K               kill a daemonized question.
  *   --update --key K --payload F deliver the next hand after a re-roll: the
  *              live page swaps to loading cards when the user re-rolls, and
- *              reloads into this new payload the moment it lands.
+ *              reloads into this new payload the moment it lands. Always the
+ *              same key the round started with; a second --start serves a new
+ *              URL and strands the open tab on a hand that never arrives.
  *
- *   node serve-question.mjs --payload question.json [--timeout 900] [--no-open] [--port 0]
+ * --timeout bounds the wait for a page to arrive, never the user's decision:
+ * once the page heartbeats, the server lives while the page does, and exits
+ * only after --idle-grace seconds (default 600) pass with no beat, wide
+ * enough to survive a closed laptop lid mid-decision.
+ *
+ *   node serve-question.mjs --payload question.json [--timeout 900] [--idle-grace 600] [--no-open] [--port 0]
  */
 import http from 'node:http';
 import fs from 'node:fs';
@@ -120,11 +127,13 @@ if (process.env.IMPECCABLE_QUESTION_DISABLED) {
 }
 // Headless self-detection, applied only where a browser is actually wanted.
 // --no-open means the caller opens the URL itself, and --wait / --stop /
-// --schema never open anything: --wait polls a daemon whose browser question
-// was already settled at --start, --stop kills one, --schema prints text. A
-// spurious exit 2 from those breaks the documented loop, which polls --wait
-// while it exits 3 and reads --schema before building a payload.
-const wantsBrowser = !hasFlag('no-open') && !hasFlag('wait') && !hasFlag('stop') && !hasFlag('schema');
+// --schema / --update never open anything: --wait polls a daemon whose
+// browser question was already settled at --start, --stop kills one,
+// --schema prints text, and --update hands the next round to a page that is
+// already open. A spurious exit 2 from those breaks the documented loop,
+// which polls --wait while it exits 3, reads --schema before building a
+// payload, and delivers re-rolled hands with --update.
+const wantsBrowser = !hasFlag('no-open') && !hasFlag('wait') && !hasFlag('stop') && !hasFlag('schema') && !hasFlag('update');
 if (wantsBrowser && !process.env.IMPECCABLE_QUESTION_FORCE) {
   const headless =
     process.env.CI ||
@@ -176,7 +185,20 @@ function printAnswer(raw) {
 }
 
 const payloadPath = arg('payload');
-const timeoutSec = Number(arg('timeout', '900'));
+// --timeout bounds only the wait for a page to open; 0 is the explicit
+// wait-forever. A negative or unparseable value takes the default, so a
+// typo cannot disarm the no-page exit and leak the daemon.
+const timeoutArg = Number(arg('timeout', '900'));
+const timeoutSec = Number.isFinite(timeoutArg) && timeoutArg >= 0 ? timeoutArg : 900;
+// How long the server (and the page's own delivery deadline) outlive the
+// last heartbeat; a zero, negative, or unparseable value takes the default.
+const idleGraceArg = Number(arg('idle-grace', '600'));
+const idleGraceMs = (Number.isFinite(idleGraceArg) && idleGraceArg > 0 ? idleGraceArg : 600) * 1000;
+// How long a delivered next hand may sit unclaimed before it means no page
+// is coming back: --wait reads it to keep a stalled page from counting as
+// closed mid-delivery, and the daemon reads it to survive until the page's
+// watch claims a hand delivered moments before the idle deadline.
+const NEXT_CLAIM_GRACE_MS = 10000;
 const portArg = Number(arg('port', '0'));
 const QUESTION_DIR = path.join(process.cwd(), '.impeccable', 'questions');
 const stateFile = (key) => path.join(QUESTION_DIR, `${key}.state.json`);
@@ -243,7 +265,19 @@ if (hasFlag('wait')) {
     }
     try {
       const state = JSON.parse(fs.readFileSync(stateFile(key), 'utf8'));
-      if (state.lastBeat && Date.now() - state.lastBeat > 15000) { sawClose = true; break; }
+      // A silent page is not a closed one while a freshly delivered next
+      // hand sits unclaimed: a stalled page stops beating by design and its
+      // watch reloads, beating again, within seconds of the file landing.
+      // The suppression is age-bound because a closed tab never claims the
+      // hand: a file still there after the grace means no page is coming.
+      const midDelivery = (() => {
+        try { if (Date.now() - fs.statSync(path.join(QUESTION_DIR, `${key}.next.json`)).mtimeMs < NEXT_CLAIM_GRACE_MS) return true; }
+        catch { /* nothing delivered */ }
+        // The claim deletes that file before the reloaded page can beat: the
+        // claim stamp the server persisted covers the same bounded gap.
+        return Boolean(state.claimedAt) && Date.now() - state.claimedAt < NEXT_CLAIM_GRACE_MS;
+      })();
+      if (!midDelivery && state.lastBeat && Date.now() - state.lastBeat > 15000) { sawClose = true; break; }
     } catch { /* state mid-write */ }
     await new Promise((r) => setTimeout(r, 1000));
   }
@@ -280,10 +314,33 @@ if (hasFlag('stop')) {
 if (hasFlag('update')) {
   const key = arg('key');
   if (!key || !payloadPath) { console.error('serve-question: --update needs --key and --payload'); process.exit(1); }
-  JSON.parse(fs.readFileSync(payloadPath, 'utf8'));
-  try { process.kill(JSON.parse(fs.readFileSync(stateFile(key), 'utf8')).pid, 0); }
-  catch { console.error('serve-question: no live question server for that key'); process.exit(2); }
-  fs.copyFileSync(payloadPath, path.join(QUESTION_DIR, `${key}.next.json`));
+  // A hand the server cannot load must fail here, at the sender: delivered
+  // anyway, the page would see ready:true for a round that never renders.
+  const nextRound = JSON.parse(fs.readFileSync(payloadPath, 'utf8'));
+  if (!nextRound || !Array.isArray(nextRound.options) || nextRound.options.length === 0) {
+    console.error('serve-question: --update payload needs an options array; nothing was delivered. Fix the payload and rerun --update on the same key.');
+    process.exit(1);
+  }
+  // Liveness mirrors --wait: a fresh page heartbeat is the primary proof, the
+  // kill probe is secondary, and EPERM means a sandbox blocked the signal,
+  // never a dead server. This is the documented re-roll delivery step, so a
+  // false "no live server" here strands the page mid-shuffle.
+  const live = (() => {
+    try {
+      const state = JSON.parse(fs.readFileSync(stateFile(key), 'utf8'));
+      if (state.lastBeat && Date.now() - state.lastBeat < 12000) return true;
+      try { process.kill(state.pid, 0); return true; }
+      catch (err) { return err.code === 'EPERM'; }
+    } catch { return false; }
+  })();
+  if (!live) { console.error('serve-question: no live question server for that key; the page it served is gone too. Re-present the round with --start and a fresh key, or fall back to the structured question tool.'); process.exit(2); }
+  const deliveredFile = path.join(QUESTION_DIR, `${key}.next.json`);
+  fs.copyFileSync(payloadPath, deliveredFile);
+  // The file's mtime is the delivery clock --wait's grace reads: stamp it
+  // here, because a copy that preserves the source payload's older mtime
+  // would start the grace already spent.
+  const deliveredAt = new Date();
+  fs.utimesSync(deliveredFile, deliveredAt, deliveredAt);
   console.log('next round delivered; the page reloads itself');
   process.exit(0);
 }
@@ -301,7 +358,8 @@ if (hasFlag('start')) {
   const logFd = fs.openSync(logFile, 'a');
   const child = spawn(process.execPath, [
     fileURLToPath(import.meta.url), '--payload', payloadPath, '--detached-serve', '--key', key,
-    '--timeout', String(timeoutSec), ...(hasFlag('open') ? [] : ['--no-open']),
+    '--timeout', String(timeoutSec), ...(arg('idle-grace') ? ['--idle-grace', arg('idle-grace')] : []),
+    ...(hasFlag('open') ? [] : ['--no-open']),
   ], { detached: true, stdio: ['ignore', logFd, logFd] });
   child.unref();
   fs.closeSync(logFd);
@@ -338,6 +396,13 @@ let localImages = [];
 // even when the round never rendered a toggle.
 let buildPathDefault = null;
 let liveBuildPath = null;
+// True between a collected re-roll or followup answer and the --update that
+// replaces the round: the window where GET / must serve the wait, not the
+// answered cards. The timestamp anchors the delivery deadline server-side,
+// so a native refresh re-enters the wait with the time already spent, never
+// with a fresh allowance.
+let awaitingNext = false;
+let awaitingNextSince = 0;
 
 function loadRound(json) {
   const parsed = JSON.parse(json);
@@ -387,6 +452,9 @@ function loadRound(json) {
     ? { value: parsed.buildPath.value, toggle: parsed.buildPath.toggle === true }
     : null;
   liveBuildPath = buildPathDefault?.value ?? null;
+  // Last: a round that failed to load anywhere above must leave the waiting
+  // window open, never resurrect the answered cards.
+  awaitingNext = false;
 }
 try { loadRound(raw); } catch (error) { console.error(`serve-question: ${error.message}`); process.exit(1); }
 const detachedKey = hasFlag('detached-serve') ? arg('key') : null;
@@ -394,7 +462,11 @@ const nextFile = () => detachedKey ? path.join(QUESTION_DIR, `${detachedKey}.nex
 
 const esc = (s) => String(s ?? '').replace(/[&<>"]/g, (c) => ({ '&': '&amp;', '<': '&lt;', '>': '&gt;', '"': '&quot;' }[c]));
 
-function page() {
+function page(waiting = false) {
+  // The delivery deadline survives refreshes: a waiting page gets whatever
+  // remains of the original allowance, so reloading cannot renew it. Spent
+  // means the page renders already stalled and never starts a heartbeat.
+  const waitBudgetMs = waiting ? Math.max(0, awaitingNextSince + idleGraceMs - Date.now()) : idleGraceMs;
   const flipChip = (label) => `<button type="button" class="chip flip" aria-label="Flip the card"><svg viewBox="0 0 24 24" aria-hidden="true"><path d="M12 4a8 8 0 1 1-8 8" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round"/><path d="M4 5.5V12h6.5" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round"/></svg><span>${label}</span></button>`;
   const expandChip = `<button type="button" class="chip expand" aria-label="Expand the image"><svg viewBox="0 0 24 24" aria-hidden="true"><path d="M4 9V4h5M20 15v5h-5M20 9V4h-5M4 15v5h5" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round"/></svg></button>`;
   // Structured anatomy: chips and one-line facts render when the payload
@@ -866,6 +938,7 @@ function page() {
      not a recommendation. */
   #canon { align-self: center; padding: 0 4px; font-family: var(--ks-mono); font-size: .66rem; letter-spacing: .08em; text-transform: uppercase; color: inherit; opacity: .45; background: transparent; border: none; border-bottom: 1px dotted currentColor; cursor: pointer; transition: opacity .2s ease; }
   #canon:hover { opacity: .85; }
+  #canon[disabled] { opacity: .18; cursor: default; }
   .card.skeleton .media { background: var(--ks-graphite); }
   .shimmer { width: 100%; height: 100%; background: linear-gradient(100deg, var(--ks-graphite) 35%, var(--ks-graphite-2) 50%, var(--ks-graphite) 65%); background-size: 220% 100%; animation: shimmer 1.4s linear infinite; }
   .card.skeleton .line { height: 11px; border-radius: 4px; background: linear-gradient(100deg, var(--ks-graphite) 35%, var(--ks-graphite-2) 50%, var(--ks-graphite) 65%); background-size: 220% 100%; animation: shimmer 1.4s linear infinite; }
@@ -877,6 +950,8 @@ function page() {
   @keyframes shimmer { from { background-position: 120% 0; } to { background-position: -80% 0; } }
   @media (prefers-reduced-motion: reduce) { .shimmer, .card.skeleton .line { animation: none; } }
   .done { display: flex; flex-direction: column; align-items: center; gap: 1rem; padding: 7rem 1rem; font-family: var(--ks-font-display); font-size: 1.4rem; color: var(--ks-champagne); text-align: center; }
+  .stall { width: 100%; display: flex; flex-direction: column; align-items: center; gap: 1.2rem; padding: 4.5rem 1rem; font-family: var(--ks-font-display); font-size: 1.4rem; color: var(--ks-champagne); text-align: center; }
+  .stall .choose { align-self: center; margin-top: 0; }
 </style>
 <div id="ambient" aria-hidden="true"></div>
 <div id="scrim" aria-hidden="true"></div>
@@ -945,11 +1020,22 @@ ${buildPath?.toggle ? `<div id="bp-confirm" role="dialog" aria-modal="true" aria
   // still gets the goodbye screen, never a loading hand nothing will resolve.
   const FOLLOWUP = ${payload.followup === true && Boolean(detachedKey) ? 'true' : 'false'};
   const beat = () => { try { navigator.sendBeacon('/heartbeat'); } catch { fetch('/heartbeat', { method: 'POST' }); } };
-  beat();
-  setInterval(beat, 5000);
+  ${waiting && waitBudgetMs <= 0 ? '' : 'beat();'}
+  const beatTimer = setInterval(beat, 5000);
+  // A dead server must fail loudly: awaiting a rejected fetch here used to
+  // swallow the click and never print the confirmation, so the user believed
+  // a choice had landed that no one would ever collect.
   async function answer(optionId) {
-    await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId, steer: steer() }) });
-    if (FOLLOWUP) { await awaitNextRound(); return; }
+    // Quiet at the click: a re-roll or canon posted while this pick's POST
+    // is in flight would overwrite the answer being collected.
+    document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+    try {
+      await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId, steer: steer() }) });
+    } catch {
+      document.body.innerHTML = '<div class="done">The question server went away before this choice could land.<br>Tell the agent your pick in the chat instead.</div>';
+      return;
+    }
+    if (FOLLOWUP) { await awaitNextRound(true); return; }
     document.body.innerHTML = '<div class="done"><svg viewBox="0 0 24 24" width="38" height="38" fill="oklch(84% 0.19 80.46)" aria-hidden="true"><path d="M5 2.5 L13.5 2.5 L5.5 21.5 L5 21.5 Q2.5 21.5 2.5 19 L2.5 5 Q2.5 2.5 5 2.5 Z"/><path d="M16.5 2.5 L19 2.5 Q21.5 2.5 21.5 5 L21.5 19 Q21.5 21.5 19 21.5 L8.5 21.5 Z"/></svg>Choice recorded. The agent is resuming; you can close this tab.</div>';
   }
   document.querySelectorAll('button.choose').forEach(b => b.addEventListener('click', () => answer(b.dataset.id)));
@@ -1381,15 +1467,62 @@ ${buildPath?.toggle ? `<div id="bp-confirm" role="dialog" aria-modal="true" aria
   document.addEventListener('keydown', (e) => { if (e.key === 'Escape' && !lightbox.hidden) closeLightbox(); });
   document.getElementById('canon')?.addEventListener('click', () => answer('canon'));
   const dealAgain = async (register) => {
-    await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId: 'reroll', steer: steer(), ...(register ? { register } : {}) }) });
-    await awaitNextRound();
+    // Quiet at the click, not after the fly-out: the POST round-trip plus
+    // the 700ms animation was a window where a second click posted another
+    // re-roll and renewed the delivery deadline.
+    document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+    try {
+      await fetch('/answer', { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ optionId: 'reroll', steer: steer(), ...(register ? { register } : {}) }) });
+    } catch {
+      document.body.innerHTML = '<div class="done">The question server went away before this choice could land.<br>Tell the agent your pick in the chat instead.</div>';
+      return;
+    }
+    await awaitNextRound(true);
   };
-  async function awaitNextRound() {
+  async function awaitNextRound(animate, budgetMs = ${idleGraceMs}) {
     const grid = document.querySelector('.grid');
+    let poll;
+    let misses = 0;
+    const shuffleStart = Date.now();
+    const stall = (message) => {
+      clearInterval(poll);
+      // A stalled page is an abandoned flow: keep heartbeating and the
+      // daemon never reaches its idle grace, so --wait spins on WAITING
+      // forever. Go silent and let the server reclaim itself. Reload must
+      // not undo that silence: an unconditional reload re-serves the same
+      // unresolved round and its fresh page beats again, so check for a
+      // delivered hand first and only reload when one exists. The re-roll
+      // buttons and the canon exit go too: a stalled page served already
+      // expired never disabled them, a re-roll would renew the deadline the
+      // stall just enforced, and a canon pick would overwrite a re-roll
+      // --wait already collected, closing the table under the agent.
+      clearInterval(beatTimer);
+      document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+      // Silence is for heartbeats only: a hand delivered after the deadline
+      // must still land without a click, so a beat-free watch keeps checking
+      // and reloads into it. /next-status never beats, so the daemon's idle
+      // grace still reclaims a flow nobody resumes.
+      const watch = setInterval(async () => {
+        try { if ((await (await fetch('/next-status')).json()).ready) { clearInterval(watch); location.reload(); } } catch { /* server gone; the screen already says so */ }
+      }, 1500);
+      grid.innerHTML = '<div class="stall"><p>' + message + '</p><button type="button" class="choose">Reload</button></div>';
+      grid.querySelector('.stall .choose').addEventListener('click', async () => {
+        try {
+          if ((await (await fetch('/next-status')).json()).ready) { location.reload(); return; }
+          grid.querySelector('.stall p').textContent = 'Still nothing to deal. Check the agent session, or answer in the chat instead.';
+        } catch {
+          grid.querySelector('.stall p').textContent = 'The question server went away. Ask the agent to restart it, or answer in the chat instead.';
+        }
+      });
+    };
+    // A refresh that lands after the delivery deadline has nothing left to
+    // wait for: stall before the heartbeat timer's first tick can fire, so
+    // the served page stays silent.
+    if (budgetMs <= 0) { stall('The next hand never arrived. Check the agent session, then reload.'); return; }
     const cardsNow = [...grid.querySelectorAll('.card')];
-    const g = grid.getBoundingClientRect();
-    const cx = g.left + g.width / 2, cy = g.top + g.height / 2;
-    if (!matchMedia('(prefers-reduced-motion: reduce)').matches) {
+    if (animate && !matchMedia('(prefers-reduced-motion: reduce)').matches) {
+      const g = grid.getBoundingClientRect();
+      const cx = g.left + g.width / 2, cy = g.top + g.height / 2;
       cardsNow.forEach((card, i) => {
         const r = card.getBoundingClientRect();
         card.style.transition = 'transform .5s cubic-bezier(.5,0,.75,0) ' + (i * 60) + 'ms, opacity .4s ease ' + (i * 60 + 120) + 'ms, filter .45s ease ' + (i * 60) + 'ms';
@@ -1401,17 +1534,35 @@ ${buildPath?.toggle ? `<div id="bp-confirm" role="dialog" aria-modal="true" aria
     }
     const cardHeight = cardsNow[0] ? cardsNow[0].getBoundingClientRect().height : 0;
     grid.innerHTML = cardsNow.map(() => '<article class="card skeleton"' + (cardHeight ? ' style="height:' + cardHeight + 'px"' : '') + '><div class="card-inner"><div class="face front"><div class="media"><div class="shimmer"></div></div><div class="body"><div class="line tier w40"></div><div class="line title w70"></div><div class="line w90"></div><div class="line w80"></div><div class="line w60"></div><div class="line button"></div></div></div></div></article>').join('');
-    document.querySelectorAll('.reroll-btn').forEach(b => b.setAttribute('disabled', ''));
-    const poll = setInterval(async () => {
+    // Canon goes quiet with the re-roll buttons: a pick posted mid-wait can
+    // never be collected once --wait has the re-roll, only close the table.
+    document.querySelectorAll('.reroll-btn, #canon').forEach(b => b.setAttribute('disabled', ''));
+    // The wait must be able to end: a dead server rejects every tick and a
+    // round nobody delivers stays ready:false forever, and both used to spin
+    // the skeletons indefinitely. Distinguish them, say so, and offer a way
+    // out. The delivery deadline is the server's own idle grace, so the page
+    // never gives up on a server that would still accept the hand.
+    poll = setInterval(async () => {
       try {
         const status = await (await fetch('/next-status')).json();
+        misses = 0;
         if (status.ready) { clearInterval(poll); location.reload(); }
-      } catch { /* server briefly busy */ }
+        else if (Date.now() - shuffleStart > budgetMs) stall('The next hand never arrived. Check the agent session, then reload.');
+      } catch {
+        misses += 1;
+        if (misses >= 8) stall('The question server went away. Ask the agent to restart it, or answer in the chat instead.');
+      }
     }, 1200);
   }
   document.getElementById('reroll')?.addEventListener('click', () => dealAgain());
   document.getElementById('reroll-safer')?.addEventListener('click', () => dealAgain('safer'));
   document.getElementById('reroll-bolder')?.addEventListener('click', () => dealAgain('bolder'));
+  // A native refresh must not resurrect an answered round: while the server
+  // holds a collected re-roll or followup pick with no replacement delivered,
+  // it serves the page in waiting mode and the refresh re-enters the same
+  // bounded wait, with only the time the original deadline has left, instead
+  // of showing dead cards whose heartbeat props the daemon forever.
+  ${waiting ? `awaitNextRound(false, ${waitBudgetMs});` : ''}
 </script>`;
 }
 
@@ -1419,14 +1570,32 @@ const server = http.createServer((req, res) => {
   if (req.method === 'GET' && req.url === '/') {
     const pending = nextFile();
     if (pending && fs.existsSync(pending)) {
-      try { loadRound(fs.readFileSync(pending, 'utf8')); fs.rmSync(pending); } catch { /* keep current round */ }
+      // A next file the round cannot load has to leave the disk either way:
+      // kept, /next-status stays ready:true and the waiting page reloads
+      // into the same failure without bound.
+      try { loadRound(fs.readFileSync(pending, 'utf8')); } catch { /* keep current round */ }
+      try { fs.rmSync(pending); } catch { /* already gone */ }
+      // The claim consumes the file the idle-exit hold reads, and the
+      // reloading page cannot beat until it has parsed: stamp the claim so
+      // the same bounded grace covers the gap between them. Persisted too,
+      // because --wait watches the same gap from outside this process and
+      // would otherwise read the stale beat as a closed page.
+      server.lastClaimAt = Date.now();
+      if (detachedKey) {
+        try {
+          const state = JSON.parse(fs.readFileSync(stateFile(detachedKey), 'utf8'));
+          state.claimedAt = server.lastClaimAt;
+          fs.writeFileSync(stateFile(detachedKey), JSON.stringify(state));
+        } catch { /* state file recreated on next beat */ }
+      }
     }
     res.writeHead(200, { 'content-type': 'text/html; charset=utf-8' });
-    res.end(page());
+    res.end(page(awaitingNext));
     return;
   }
   if (req.method === 'POST' && req.url === '/heartbeat') {
     res.writeHead(204); res.end();
+    server.lastBeatSeen = Date.now();
     if (detachedKey) {
       const now = Date.now();
       if (!server.lastBeatWrite || now - server.lastBeatWrite > 4000) {
@@ -1502,6 +1671,11 @@ const server = http.createServer((req, res) => {
         ...((chosen?.comp ?? chosen?.sketch) ? { comp: chosen.comp ?? chosen.sketch } : {}),
         ...(liveBuildPath && !isReroll ? { buildPath: liveBuildPath, buildPathFlipped: liveBuildPath !== (buildPathDefault?.value ?? null) } : {}),
       });
+      // The delivery deadline is single-issue: a duplicate answer racing the
+      // page's disable must not restamp the allowance already inherited.
+      const wasAwaiting = awaitingNext;
+      awaitingNext = (isReroll || followupOpen) && Boolean(detachedKey);
+      if (awaitingNext && !wasAwaiting) awaitingNextSince = Date.now();
       if (detachedKey) {
         fs.mkdirSync(QUESTION_DIR, { recursive: true });
         fs.writeFileSync(answerFile(detachedKey), answer + '\n');
@@ -1531,10 +1705,38 @@ server.listen(portArg, '127.0.0.1', () => {
   if (!hasFlag('no-open')) {
     openSystemBrowser(url);
   }
-  if (timeoutSec > 0) {
-    setTimeout(() => {
-      console.log('serve-question: timed out with no answer');
-      process.exit(2);
-    }, timeoutSec * 1000).unref?.();
-  }
+  // The timeout bounds the wait for a page, never the user's decision: an
+  // absolute guillotine counted from start used to kill the server under a
+  // still-open tab (a slow re-rolled round easily outlived it), leaving the
+  // page polling skeletons that could never resolve. Once the page beats,
+  // the server's lifetime tracks the beats, and it exits only after the idle
+  // grace passes with none, long enough to survive a closed laptop lid.
+  // --timeout 0 waits for a page forever, but the idle grace still applies
+  // once one has beat: a page that arrived and went silent is a closed tab,
+  // and no timeout setting should let that daemon leak.
+  const startedAt = Date.now();
+  const lifetime = setInterval(() => {
+    if (!server.lastBeatSeen) {
+      if (timeoutSec > 0 && Date.now() - startedAt > timeoutSec * 1000) {
+        console.log('serve-question: timed out with no answer');
+        process.exit(2);
+      }
+    } else if (Date.now() - server.lastBeatSeen > idleGraceMs) {
+      // A hand delivered moments before this deadline still gets its claim
+      // window: the stalled page's watch reloads into it and beats again
+      // within seconds, while a file unclaimed past the grace means no page
+      // is coming back (the same verdict --wait reads from its age). The
+      // claim itself holds the daemon too: GET / deletes the file before the
+      // reloaded page can beat, so a tick in that gap must not exit under
+      // the hand just claimed.
+      const pending = nextFile();
+      let deliveredAt = 0;
+      if (pending) { try { deliveredAt = fs.statSync(pending).mtimeMs; } catch { /* nothing delivered */ } }
+      if (Date.now() - Math.max(deliveredAt, server.lastClaimAt || 0) > NEXT_CLAIM_GRACE_MS) {
+        console.log('serve-question: the page stopped beating and never came back; exiting');
+        process.exit(2);
+      }
+    }
+  }, 2000);
+  lifetime.unref?.();
 });

.claude-plugin/marketplace.json

@@ -12,7 +12,7 @@
     {
       "name": "impeccable",
       "description": "Design fluency for frontend development. 1 skill with 23 commands (/impeccable polish, /impeccable audit, /impeccable critique, etc.) and curated anti-pattern detection.",
-      "version": "4.1.0",
+      "version": "4.1.2",
       "author": {
         "name": "Paul Bakaus",
         "email": "paul@paulbakaus.com"

.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "impeccable",
   "description": "Design fluency for frontend development. 1 skill with 23 commands (/impeccable polish, /impeccable audit, /impeccable critique, etc.) and curated anti-pattern detection.",
-  "version": "4.1.0",
+  "version": "4.1.2",
   "author": {
     "name": "Paul Bakaus",
     "email": "paul@paulbakaus.com"

.claude/agents/impeccable-asset-producer.md

@@ -8,17 +8,15 @@ maxTurns: 24
 ---
 # Impeccable Asset Producer
 
-You are the asset production agent for Impeccable craft.
-
-Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent agent gives you. The assets you create will be used to build a real site, so treat every raster as a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
+You are the asset production agent for Impeccable craft. Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent gives you. Every raster you create is a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
 
 ## Core Rule
 
-Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; if CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
+Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; when CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
 
 ## Decision Comps
 
-When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so your entire contract is this card; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; a card too thin to brief a comp is reported back, not padded from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (its regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world; a native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment is what keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
+When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so this card is your entire contract; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; report a card too thin to brief a comp, never pad it from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world. A native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
 
 ## Input Contract
 
@@ -30,42 +28,42 @@ Expect:
 - Required dimensions, format, transparency needs, and avoid list.
 - Notes on what should remain semantic HTML/CSS/SVG instead of raster.
 
-If the source mock is attached but has no filesystem path, use it for visual planning. Ask for a path only before cropping or writing assets.
+If the source mock is attached but has no filesystem path, use it for visual planning; ask for a path only before cropping or writing assets.
 
-Use defaults unless contradicted:
+Defaults unless contradicted:
 
 - `.webp` for opaque photos, backgrounds, and textures.
 - `.png` for transparent cutouts, seals, tickets, and illustrations.
-- Target production size or at least 2x display size when dimensions are known. Do not use small full-page mock crop size as the default shipping size.
-- Remove UI text, navigation, buttons, labels, and body copy by default.
+- Target production size, or at least 2x display size when dimensions are known. Never default to the small size of a full-page mock crop.
+- Remove UI text, navigation, buttons, labels, and body copy.
 - Keep physical marks only when the parent says they are part of the asset.
-- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic to the asset.
-- Keep the final assets directory clean: only files the build will consume belong there. Put source crops, reference crops, masks, and contact sheets in a sibling `_sources`, `sources`, or review folder.
+- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic.
+- Keep the final assets directory clean: only files the build will consume. Source crops, reference crops, masks, and contact sheets go in a sibling `_sources`, `sources`, or review folder.
 
-Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not block; choose defaults and report them.
+Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not; choose defaults and report them.
 
 ## Workflow
 
 1. Inventory the full approved mock or every assigned crop.
 2. Put each visual role in exactly one bucket:
    - `produce`: needs generation, image editing, cleanup, cutout work, or a clean plate before it can ship.
-   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source asset, a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
+   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source: a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
    - `semantic`: build in HTML/CSS/SVG/canvas, no raster output.
-3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, not asset grade, and a shipped crop, however close it looks, is how a beautiful comp turns into a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
+3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, and a shipped crop, however close it looks, is how a beautiful comp becomes a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
 4. Give the parent an execution order for the `produce` bucket.
-5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or semantic HTML/CSS/SVG recommendation if raster is wrong.
+5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or a semantic HTML/CSS/SVG recommendation when raster is wrong.
 6. Use the harness's native image tool by default when generation or editing is needed; otherwise use the skill's generate-image.mjs.
 
 7. Remove baked-in UI text, navigation, buttons, body copy, and mock chrome unless the text is part of the asset.
 8. Think through the final DOM/CSS representation before generating. If CSS will own radius, clipping, shadows, borders, perspective, responsive cropping, captions, or card frames, do not bake those into the bitmap.
-9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .claude/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt is embedded in the image itself, because the build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
+9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .claude/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt lives inside the image itself. The build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
 10. Compare each output against its source crop, opening every image by its workspace-relative path; sandboxed viewers reject absolute paths. If a review/QA tool is available, run it before the final manifest, then retry each major/fatal finding once before finalizing.
 
-Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed to make a reusable texture or background, classify it as crop-derived cleanup or clean-plate work.
+Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed first, classify it as crop-derived cleanup or clean-plate work.
 
-Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Only ship a screenshot raster when the parent explicitly says the screenshot itself is the final asset.
+Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Ship a screenshot raster only when the parent explicitly says the screenshot itself is the final asset.
 
-Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: name the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it should compose with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
+Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it composes with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
 
 ## Prompt Pattern
 
@@ -86,10 +84,10 @@ For transparent cutouts: use true alpha when the tool supports it; otherwise gen
 
 Return a complete manifest, grouped by `produce`, `direct`, and `semantic`. For each asset include: `id`, `source_crop`, `output_path` when applicable, `strategy`, `prompt_used` when applicable, `dimensions`, `format`, `transparency`, `deviations`, and `qa_status`.
 
-For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` must be a concrete build handoff, not a short explanation that no asset was produced. It should name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities that code owns.
+For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` is a concrete build handoff, not a note that no asset was produced: name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities code owns.
 
-`qa_status` must be `accepted`, `needs_parent_review`, or `blocked`. Use `accepted` only after visual comparison passes. Use `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. Use `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
+`qa_status` is `accepted`, `needs_parent_review`, or `blocked`. `accepted` only after visual comparison passes. `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
 
-End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal. Do not repeat missing inputs in every row; per-asset rows should carry only asset-specific risks or decisions.
+End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal; per-asset rows carry only asset-specific risks or decisions.
 
 Do not modify implementation code. Do not edit the approved mock. Do not produce final page copy. The parent craft agent owns implementation and final mock fidelity.

.claude/agents/impeccable-finish-reviewer.md

@@ -8,35 +8,36 @@ maxTurns: 30
 ---
 # Impeccable Finish Reviewer
 
-You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You do not edit anything; the parent agent applies your fixes.
+You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You edit nothing; the parent applies your fixes.
 
-You have no browser. Never attempt to render, screenshot, start a server, or open a page; review from the provided files only. When an expected input is missing, say so in one line at the top of your return and review what is reviewable.
+You have no browser. Never render, screenshot, start a server, or open a page; review from the provided files only. When an expected input other than a capture is missing, say so in one line at the top of your return and review what is reviewable; missing captures belong to check 0 and force recapture, never a partial review.
 
-A hard turn ceiling ends the run without warning; a run that ends before the five sections are written returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads into each turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
+A hard turn ceiling ends the run without warning; a run that ends before its contracted sections are written (five, or the single recapture section) returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads per turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
 
 ## Input Contract
 
-Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, which live in `.impeccable/review/` (on the web, `desktop.png` and `mobile.png`; on native, device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive); a screenshot path the calling brief names is authoritative when the file exists, and `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent; the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths and, on a comp-led build, the approved comp path (a code-led build has no approved comp; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing in this file that binds “the approved comp” binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet also carries the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor and judge every check in the platform's own conventions, the screenshots are device captures rather than browser viewports, and your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
+Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, in `.impeccable/review/` (web: `desktop.png` and `mobile.png`; native: device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive). A screenshot path the calling brief names is authoritative when the file exists; `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent. Also expect: the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); the PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths; on a comp-led build the approved comp path (a code-led build has none; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing here that binds "the approved comp" binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet adds the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor, judge every check in the platform's own conventions, treat the screenshots as device captures, and know your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
 
 ## Checks, in order
 
-1. **Persistence.** PRODUCT.md exists. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too, the surface brief naming the approved comp or an `approved` flag in its sidecar; comp-round comps with no recorded pick mean the approval point was skipped, and that is a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and they imply no approval whatever the build path; a code-led build has no comp round at all.
-2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element, and its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Two rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement, because medium is part of the promise. When no approved comp was supplied, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality, CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never actually renders, as contradicted on its face; imitation material is the single most reliable mark of machine-made design. A critique-reference comp, when one arrived on such a build, is provocation rather than spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is the question of what the image dared that the build did not, and the dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. In every material_fixes list, a fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
+0. **Evidence.** Before any other check, verify the required captures exist and every capture is valid. Required: the platform's full viewport set (web: `desktop.png` and `mobile.png`; native: one capture per shipped device class), plus every capture the calling brief names as required, a reported user viewport (`user-<width>.png`) included. Valid: no black or blank regions, content matching what the filename claims (a visit capture showing the About section is invalid), the document top visible where the file claims a full page, dimensions that make sense for the named viewport. A required capture that is absent fails exactly like one that is malformed: a viewport nobody captured is a viewport nobody inspected, and it cannot ship. When any capture fails, the whole review changes shape: return `disposition: recapture` as the first line, then one section, `recapture`, listing each missing or invalid file and what a valid capture of it shows, and stop. Never build a matrix on malformed evidence; a verdict derived from a broken capture launders the breakage into an approval, and the parent owes you a full re-review on valid captures, not a scoring round.
+1. **Persistence.** PRODUCT.md exists. On a comp-led build, `.impeccable/review/hero-repro.png` exists: the hero reproduction checkpoint's capture at the comp's own dimensions; its absence means the reproduction phase ran unproven, a material finding. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too: the surface brief naming the approved comp, or an `approved` flag in its sidecar. Comp-round comps with no recorded pick mean the approval point was skipped, a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and imply no approval whatever the build path; a code-led build has no comp round at all.
+2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element; its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Three rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement; medium is part of the promise. GROUND: the page field's value and temperature against the comp's, sampled from pixels on both sides when tooling allows rather than judged from memory, and read as the net on-screen result where a texture or tile paints over the base color; a ground warmer or cooler than the comp's is contradicted however faithfully the layout matches, and drift toward the rendition prior (warm cream on light grounds, blue-black slate on dark) is the direction to hunt. With no approved comp, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality (CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never renders) as contradicted on its face; imitation material is the single most reliable mark of machine-made design. GROUND narrows rather than lapses: with no comp to sample, a color OWN-WORLD names is the target and the same warmer-or-cooler judgment applies; when OWN-WORLD names none, there is no GROUND authority, and the review says so in place of a verdict, because a target the reviewer invents turns the check into taste. A critique-reference comp on such a build is provocation, not spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is what the image dared that the build did not, and dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. A fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
 3. **Ceiling.** Against the QUALITY BAR card: name the world's native devices the build left unused, frame, depth, lettering treatment, ornament density, motion. The card governs commitment and finish, never composition.
-4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped and that is a material fix ahead of any craft point. Then, for each of the five blocks, does the render keep the promise? Apply the memory test to the first viewport.
+4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped, a material fix ahead of any craft point. Then, for each of the five blocks: does the render keep the promise? Apply the memory test to the first viewport.
 5. **Truth.** Demonstration data authored and labeled synthetic; no invented commercial claims; unanswered claims present as marked placeholders, not omissions. Every image-native region of the approved comp shipped as a real asset, not a gradient standing in for one, and every produced asset visibly present in the screenshots; an asset applied at near-zero opacity or buried behind other paint is a compliance token, not a shipped material.
-6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp, because the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
+6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp: the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
 
 Do not run a second detector pass; mechanical findings belong to the parent's hooks.
 
 ## Disposition
 
-The first line of your return is `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. It is derived, never felt: rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
+The first line of your return is `disposition: recapture`, `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. These four words are the whole vocabulary; never invent another. The word is derived, never felt: recapture when the evidence check failed, rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
 
 ## Output Contract
 
-Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). Missing inputs are named in one line above the sections. No praise, no summary prose.
+Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). A recapture return replaces the five sections with the single `recapture` section from check 0. Missing inputs are named in one line above the sections. No praise, no summary prose.
 
 ## Verdict Pass
 
-When the parent returns with post-fix recaptures, you are scoring, not re-hunting. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths for this round; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open; unresolved or partial material findings can never recompute to ship.
+When the parent returns with post-fix recaptures, you are scoring, not re-hunting. Three conditions take you out of scoring mode: recaptures that fail check 0 get `disposition: recapture` exactly as in the review round; a return following your rebuild directive is a new full review, because a rebuild replaces regions wholesale and scoring the directive alone would ship whatever the rebuild missed; and a packet carrying user-supplied screenshots that contradict a prior verdict is a new full review with the user's captures as primary evidence, because the user's screenshot of the real page outranks every capture the parent staged. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open, in the same four-word vocabulary. Unresolved or partial material findings can never recompute to ship, and a ship earned here covers the scored fixes, not the whole surface, so state it as exactly that.

.claude/settings.json

@@ -1,9 +1,9 @@
 {
-  "description": "Impeccable design detector: immediate-tier checks after Edit/Write/MultiEdit on UI files, full-rule deep pass on Stop.",
+  "description": "Impeccable design detector: immediate-tier checks after Edit/Write on UI files, full-rule deep pass on Stop.",
   "hooks": {
     "PostToolUse": [
       {
-        "matcher": "Edit|Write|MultiEdit",
+        "matcher": "Edit|Write",
         "hooks": [
           {
             "type": "command",

.claude/skills/impeccable/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: impeccable
 description: Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, settings, onboarding, and empty states. Handles UX review, visual hierarchy, information architecture, cognitive load, accessibility, performance, responsive behavior, theming, anti-patterns, typography, fonts, spacing, layout, alignment, color, motion, micro-interactions, UX copy, error states, edge cases, i18n, and reusable design systems or tokens. Also use for bland designs that need to become bolder or more delightful, loud designs that should become quieter, live browser iteration on UI elements, or ambitious visual effects that should feel technically extraordinary. Not for backend-only or non-UI tasks.
-version: 4.1.0
+version: 4.1.2
 user-invocable: true
 argument-hint: "[shape · audit|critique · animate|bolder|colorize|delight|layout|overdrive|quieter|typeset · adapt|clarify|distill · harden|onboard|optimize|polish · init|document|extract|live] [target]"
 license: Apache 2.0

.claude/skills/impeccable/reference/craft-floor.md

@@ -38,6 +38,7 @@ Surface habits:
 - Monospace as a costume for "technical" rather than for code, data, or measurement.
 - A system display face (Impact, Arial Black, the platform sans) as the display voice of an own-world page. Source and self-host a face whose character matches the approved lettering; the closest installed font is a failure, not a fallback.
 - Unicode glyphs or emoji standing in for an icon system. Icons are drawn, from a real library or authored SVG, in one consistent stroke and weight.
+- Geometric masks standing in for organic contours. A circle, polygon, or radial-gradient cutout approximating a photographic subject's edge is the cheap version of the effect and reads worse than omitting it. Derive an alpha matte from the actual image, or produce a cut-out asset.
 - Light or dark picked by category. Pick it from the use scene: who, where, under what ambient light.
 
 The floor holds the mechanics; it never picks the direction. With every check green, spend the page on the committed world, and when torn between refined and committed, commit.

.claude/skills/impeccable/reference/degraded/asset-producer.md

@@ -3,17 +3,15 @@ This harness has no subagent capability, so you are running this role inline. St
 
 # Impeccable Asset Producer
 
-You are the asset production agent for Impeccable craft.
-
-Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent agent gives you. The assets you create will be used to build a real site, so treat every raster as a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
+You are the asset production agent for Impeccable craft. Your job is production cleanup, not new art direction. Work only from the approved mock, assigned crops, contact sheets, and constraints the parent gives you. Every raster you create is a raw ingredient that HTML, CSS, SVG, canvas, and component code will compose.
 
 ## Core Rule
 
-Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; if CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
+Do not redesign. Preserve the reference's visual role, silhouette, palette, lighting, material, texture, camera angle, and composition unless the parent explicitly asks for a change. Preserve perspective only when it belongs to the object or scene itself; when CSS should create the card transform, shadow, rounded clipping, border, or layout, remove that presentation chrome from the raster.
 
 ## Decision Comps
 
-When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so your entire contract is this card; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; a card too thin to brief a comp is reported back, not padded from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (its regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world; a native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment is what keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
+When the parent hands you a decision card packet instead of an approved mock, the job is one comp: one card, one file, written to the card's declared `comp` path the moment it renders. The parent runs several of you in parallel, one per card, so this card is your entire contract; generate first, plan never, because the file on disk is the deliverable and the decision page is waiting on it. Work from the card's structured fields and PRODUCT.md alone; report a card too thin to brief a comp, never pad it from imagination. Render the card's direction as a north-star comp at full fidelity: the requested surface's first viewport, prompt led by the surface's own structure (regions named in order with their scale relationships, never the world's atmosphere), fully committed in the card's own palette, type character, and material world. A native app or mobile-first surface is a portrait frame at its device viewport, never a landscape default. Every sibling renders at the same full fidelity in its own grammar, one surface, one aspect; equal commitment keeps the comparison honest. Real product name and real content only; never invent commercial claims, prices, benchmarks, or dates PRODUCT.md does not carry. Exclusions bind those claims, never a medium the card's own world has not excluded: a subject that lives in photographs keeps its photographs. Write the prompt sidecar beside the file. Return one line naming the path and any deviation, nothing more. Everything below this section is the asset-production job; none of it applies to a decision-comp run.
 
 ## Input Contract
 
@@ -25,42 +23,42 @@ Expect:
 - Required dimensions, format, transparency needs, and avoid list.
 - Notes on what should remain semantic HTML/CSS/SVG instead of raster.
 
-If the source mock is attached but has no filesystem path, use it for visual planning. Ask for a path only before cropping or writing assets.
+If the source mock is attached but has no filesystem path, use it for visual planning; ask for a path only before cropping or writing assets.
 
-Use defaults unless contradicted:
+Defaults unless contradicted:
 
 - `.webp` for opaque photos, backgrounds, and textures.
 - `.png` for transparent cutouts, seals, tickets, and illustrations.
-- Target production size or at least 2x display size when dimensions are known. Do not use small full-page mock crop size as the default shipping size.
-- Remove UI text, navigation, buttons, labels, and body copy by default.
+- Target production size, or at least 2x display size when dimensions are known. Never default to the small size of a full-page mock crop.
+- Remove UI text, navigation, buttons, labels, and body copy.
 - Keep physical marks only when the parent says they are part of the asset.
-- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic to the asset.
-- Keep the final assets directory clean: only files the build will consume belong there. Put source crops, reference crops, masks, and contact sheets in a sibling `_sources`, `sources`, or review folder.
+- Remove letterboxing, empty padding, baked card corners, borders, shadows, caption bands, and layout background unless the parent says those pixels are intrinsic.
+- Keep the final assets directory clean: only files the build will consume. Source crops, reference crops, masks, and contact sheets go in a sibling `_sources`, `sources`, or review folder.
 
-Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not block; choose defaults and report them.
+Ask blockers once, globally. Missing source path/crops or output directory blocks production. Exact dimensions, compression targets, retina variants, and format preferences do not; choose defaults and report them.
 
 ## Workflow
 
 1. Inventory the full approved mock or every assigned crop.
 2. Put each visual role in exactly one bucket:
    - `produce`: needs generation, image editing, cleanup, cutout work, or a clean plate before it can ship.
-   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source asset, a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
+   - `direct`: ships after format conversion, compression, or renaming because the parent supplied a real standalone source: a project file, stock, or prior production art. A crop from the approved mock is never `direct`, whatever its apparent size.
    - `semantic`: build in HTML/CSS/SVG/canvas, no raster output.
-3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, not asset grade, and a shipped crop, however close it looks, is how a beautiful comp turns into a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
+3. Crops from the mock are binding visual references, never shipping pixels: a full-page mock's effective resolution is reference grade, and a shipped crop, however close it looks, is how a beautiful comp becomes a blurry site. Every mock-derived asset goes through `produce` as a clean regeneration.
 4. Give the parent an execution order for the `produce` bucket.
-5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or semantic HTML/CSS/SVG recommendation if raster is wrong.
+5. For produced assets, choose the least inventive strategy: image-to-image clean plate, faithful regeneration from crop reference, transparent cutout, texture/pattern reconstruction, stock/project source, or a semantic HTML/CSS/SVG recommendation when raster is wrong.
 6. Use the harness's native image tool by default when generation or editing is needed; otherwise use the skill's generate-image.mjs.
 
 7. Remove baked-in UI text, navigation, buttons, body copy, and mock chrome unless the text is part of the asset.
 8. Think through the final DOM/CSS representation before generating. If CSS will own radius, clipping, shadows, borders, perspective, responsive cropping, captions, or card frames, do not bake those into the bitmap.
-9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .claude/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt is embedded in the image itself, because the build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
+9. Save outputs non-destructively in the requested project directory, and leave the intent with the file: after every generation, run `node .claude/skills/impeccable/scripts/embed-prompt.mjs <asset> --prompt "<the prompt used>"` so the prompt lives inside the image itself. The build thread composes what you made and needs to know what it is looking at, and the embedding survives copies where sidecars get lost.
 10. Compare each output against its source crop, opening every image by its workspace-relative path; sandboxed viewers reject absolute paths. If a review/QA tool is available, run it before the final manifest, then retry each major/fatal finding once before finalizing.
 
-Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed to make a reusable texture or background, classify it as crop-derived cleanup or clean-plate work.
+Use `texture/pattern extraction` only when the source region is already clean enough to sample as texture. If UI, cards, labels, headings, body copy, or footer chrome must be removed first, classify it as crop-derived cleanup or clean-plate work.
 
-Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Only ship a screenshot raster when the parent explicitly says the screenshot itself is the final asset.
+Use `semantic` for dashboards, charts, controls, screenshots of whole UI sections, data widgets, card chrome, app frames, icon toolbars, logos, wordmarks, and anything the final implementation can render crisply in HTML/CSS/SVG/canvas. Ship a screenshot raster only when the parent explicitly says the screenshot itself is the final asset.
 
-Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: name the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it should compose with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
+Semantic does not mean ignored. For every semantic role, write a concrete implementation handoff for the parent craft agent: the DOM/component layers, CSS-owned visual treatment, SVG/canvas/icon-library pieces, responsive behavior, and which nearby produced raster assets it composes with. For logos and icons, prefer inline SVG/vector or icon-library implementation unless the parent provides a production logo raster.
 
 ## Prompt Pattern
 
@@ -81,10 +79,10 @@ For transparent cutouts: use true alpha when the tool supports it; otherwise gen
 
 Return a complete manifest, grouped by `produce`, `direct`, and `semantic`. For each asset include: `id`, `source_crop`, `output_path` when applicable, `strategy`, `prompt_used` when applicable, `dimensions`, `format`, `transparency`, `deviations`, and `qa_status`.
 
-For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` must be a concrete build handoff, not a short explanation that no asset was produced. It should name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities that code owns.
+For each semantic row include `id`, `implementation`, `notes`, and `qa_status`. The `implementation` is a concrete build handoff, not a note that no asset was produced: name the likely HTML/CSS/SVG/canvas/icon/component pieces and the visual responsibilities code owns.
 
-`qa_status` must be `accepted`, `needs_parent_review`, or `blocked`. Use `accepted` only after visual comparison passes. Use `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. Use `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
+`qa_status` is `accepted`, `needs_parent_review`, or `blocked`. `accepted` only after visual comparison passes. `needs_parent_review` for cut-off subjects, unwanted borders or rounded-card chrome, letterboxing, baked semantic text, low-resolution output, perspective that should have been CSS, missing transparency, or drift from the crop. `blocked` when inputs, permissions, image capability, or asset source quality prevent a credible result.
 
-End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal. Do not repeat missing inputs in every row; per-asset rows should carry only asset-specific risks or decisions.
+End with `execution_order`, `blockers`, and `assumptions` sections. Keep blockers global and minimal; per-asset rows carry only asset-specific risks or decisions.
 
 Do not modify implementation code. Do not edit the approved mock. Do not produce final page copy. The parent craft agent owns implementation and final mock fidelity.

.claude/skills/impeccable/reference/degraded/finish-reviewer.md

@@ -3,35 +3,36 @@ This harness has no subagent capability, so you are running this role inline. St
 
 # Impeccable Finish Reviewer
 
-You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You do not edit anything; the parent agent applies your fixes.
+You are the finishing reviewer for an Impeccable build: fresh eyes on a done artifact, outside the build thread's attention gravity. You edit nothing; the parent applies your fixes.
 
-You have no browser. Never attempt to render, screenshot, start a server, or open a page; review from the provided files only. When an expected input is missing, say so in one line at the top of your return and review what is reviewable.
+You have no browser. Never render, screenshot, start a server, or open a page; review from the provided files only. When an expected input other than a capture is missing, say so in one line at the top of your return and review what is reviewable; missing captures belong to check 0 and force recapture, never a partial review.
 
-A hard turn ceiling ends the run without warning; a run that ends before the five sections are written returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads into each turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
+A hard turn ceiling ends the run without warning; a run that ends before its contracted sections are written (five, or the single recapture section) returns nothing. Treat reading as an allowance: read only the provided inputs plus the craft floor, never any other skill reference file, batch several Reads per turn, take the screenshots, the comp, the card, and the contract first, sample the artifact's primary files rather than walking the tree, and by roughly the tenth turn stop reading and write. Name whatever went unread in the line above the sections.
 
 ## Input Contract
 
-Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, which live in `.impeccable/review/` (on the web, `desktop.png` and `mobile.png`; on native, device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive); a screenshot path the calling brief names is authoritative when the file exists, and `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent; the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths and, on a comp-led build, the approved comp path (a code-led build has no approved comp; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing in this file that binds “the approved comp” binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet also carries the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor and judge every check in the platform's own conventions, the screenshots are device captures rather than browser viewports, and your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
+Expect: the original request; the confirmed user answers; the artifact path(s); the screenshots the parent captured, in `.impeccable/review/` (web: `desktop.png` and `mobile.png`; native: device-class names such as `phone.png` and `tablet.png`, suffixed per OS on adaptive). A screenshot path the calling brief names is authoritative when the file exists; `.impeccable/review/` is where to look when the brief names none or a named path is missing, never a filename you invent. Also expect: the direction contract (THESIS, OWN-WORLD, STORY, FIRST VIEWPORT, FORM); the PRODUCT.md path; existing hook or detector findings; the chosen world's QUALITY BAR card paths; on a comp-led build the approved comp path (a code-led build has none; it passes the chosen decision comp as a separate critique-reference input, labeled as such, and nothing here that binds "the approved comp" binds it); and the skill's `reference/craft-floor.md` path. On a native (`ios` / `android` / `adaptive`) build the packet adds the platform reference path(s) (`reference/ios.md` / `reference/android.md`) and a line saying no detector ran: read the platform reference alongside the craft floor, judge every check in the platform's own conventions, treat the screenshots as device captures, and know your floor check is the build's only slop gate. When the harness can view images, open the screenshots, the comp, and the card first, and inventory the comp's salient elements in your own words before reading the direction contract or any builder-authored summary: a review anchored on the contract inherits whatever the builder's abstraction dropped.
 
 ## Checks, in order
 
-1. **Persistence.** PRODUCT.md exists. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too, the surface brief naming the approved comp or an `approved` flag in its sidecar; comp-round comps with no recorded pick mean the approval point was skipped, and that is a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and they imply no approval whatever the build path; a code-led build has no comp round at all.
-2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element, and its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Two rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement, because medium is part of the promise. When no approved comp was supplied, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality, CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never actually renders, as contradicted on its face; imitation material is the single most reliable mark of machine-made design. A critique-reference comp, when one arrived on such a build, is provocation rather than spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is the question of what the image dared that the build did not, and the dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. In every material_fixes list, a fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
+0. **Evidence.** Before any other check, verify the required captures exist and every capture is valid. Required: the platform's full viewport set (web: `desktop.png` and `mobile.png`; native: one capture per shipped device class), plus every capture the calling brief names as required, a reported user viewport (`user-<width>.png`) included. Valid: no black or blank regions, content matching what the filename claims (a visit capture showing the About section is invalid), the document top visible where the file claims a full page, dimensions that make sense for the named viewport. A required capture that is absent fails exactly like one that is malformed: a viewport nobody captured is a viewport nobody inspected, and it cannot ship. When any capture fails, the whole review changes shape: return `disposition: recapture` as the first line, then one section, `recapture`, listing each missing or invalid file and what a valid capture of it shows, and stop. Never build a matrix on malformed evidence; a verdict derived from a broken capture launders the breakage into an approval, and the parent owes you a full re-review on valid captures, not a scoring round.
+1. **Persistence.** PRODUCT.md exists. On a comp-led build, `.impeccable/review/hero-repro.png` exists: the hero reproduction checkpoint's capture at the comp's own dimensions; its absence means the reproduction phase ran unproven, a material finding. When DESIGN.md predates this build (an extension or redesign), it matches the built world; on a new world it is written after this review by the documenter, so its absence here is not a finding. When comp-round comps exist under `.impeccable/mocks/`, an approval record exists too: the surface brief naming the approved comp, or an `approved` flag in its sidecar. Comp-round comps with no recorded pick mean the approval point was skipped, a material finding. Files under `.impeccable/mocks/decision/` are exempt: they are the direction round's dealt hand, produced before any comp round, and imply no approval whatever the build path; a code-led build has no comp round at all.
+2. **Fidelity.** Against your own element inventory of the approved comp, never against the contract's summary of it: topology, reading order, focal scale, overlaps and z-order, density, signature geometry, the primary action's treatment (a CTA the comp physically works, dissolves, or stamps is a signature element; its plain-rectangle rendition is contradicted), navigation items and icons, headline levels and scale relationships. Classify every salient element: match, acceptable adaptation, missing, contradicted, or added without approval. Three rows are mandatory in every matrix. TYPE: the display lettering's character, compression, width, weight, contrast, terminals, against the comp's; a face of a different character is contradicted however the layout matches. MATERIAL: an element rendered as flat CSS or clean vector where the comp shows painted, textured, dimensional, or photographic material is contradicted regardless of placement; medium is part of the promise. GROUND: the page field's value and temperature against the comp's, sampled from pixels on both sides when tooling allows rather than judged from memory, and read as the net on-screen result where a texture or tile paints over the base color; a ground warmer or cooler than the comp's is contradicted however faithfully the layout matches, and drift toward the rendition prior (warm cream on light grounds, blue-black slate on dark) is the direction to hunt. With no approved comp, TYPE and MATERIAL do not lapse: judge them against the contract's OWN-WORLD and the world's real materials, and treat faked physicality (CSS bevels, embossing, stamped-metal or chalk effects imitating a material the page never renders) as contradicted on its face; imitation material is the single most reliable mark of machine-made design. GROUND narrows rather than lapses: with no comp to sample, a color OWN-WORLD names is the target and the same warmer-or-cooler judgment applies; when OWN-WORLD names none, there is no GROUND authority, and the review says so in place of a verdict, because a target the reviewer invents turns the check into taste. A critique-reference comp on such a build is provocation, not spec: no element matrix, no adaptation citations, no asset obligations; its one contribution is what the image dared that the build did not, and dares worth adopting enter material_fixes as ordinary ordered fixes. An adaptation counts as intentional only when it cites the user answer, surface brief, accessibility need, or product truth that forced it; an uncited deviation is a defect. A missing signature element, a changed topology, or content added without approval fails fidelity and outranks every craft point in material_fixes. When MATERIAL is contradicted on the focal element, or contradiction is the page rather than the exception, stop ordering repairs: make the first material fix a rebuild directive naming the comp regions to re-derive and the assets to produce; a list of patches against a rejected page launders the rejection into an approval. A fix that requires producing an asset says so explicitly ("produce: <region> as a raster asset"), never phrased as a style adjustment the parent will answer with CSS. The comp is the spec for composition, topology, element inventory, density, lettering character, and material; it is not a pixel spec for semantics, accessibility, or responsive reflow, and that allowance covers translation, never replacement.
 3. **Ceiling.** Against the QUALITY BAR card: name the world's native devices the build left unused, frame, depth, lettering treatment, ornament density, motion. The card governs commitment and finish, never composition.
-4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped and that is a material fix ahead of any craft point. Then, for each of the five blocks, does the render keep the promise? Apply the memory test to the first viewport.
+4. **Contract, promise by promise.** First verify FORM carries the seed key the concept roll printed; a contract with no seed key, or one the parent cannot corroborate, means the roll was skipped, a material fix ahead of any craft point. Then, for each of the five blocks: does the render keep the promise? Apply the memory test to the first viewport.
 5. **Truth.** Demonstration data authored and labeled synthetic; no invented commercial claims; unanswered claims present as marked placeholders, not omissions. Every image-native region of the approved comp shipped as a real asset, not a gradient standing in for one, and every produced asset visibly present in the screenshots; an asset applied at near-zero opacity or buried behind other paint is a compliance token, not a shipped material.
-6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp, because the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
+6. **Floor.** Read the craft floor's Refuse list and hold the screenshots against it: kickers and eyebrows, hard offset shadows outside a neobrutalist world, glyph icons, system display faces, gradient text, side stripes, and the rest. A banned element is a material fix even when it matches nothing in the comp: the builder loaded the same ban before writing it, and fidelity to a comp cannot authorize what the floor refuses. The parent's hook findings cover this mechanically where hooks run; this check exists because hookless harnesses reach you with none, and the last two live sessions shipped five kickers past a reviewer that never looked.
 
 Do not run a second detector pass; mechanical findings belong to the parent's hooks.
 
 ## Disposition
 
-The first line of your return is `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. It is derived, never felt: rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
+The first line of your return is `disposition: recapture`, `disposition: rebuild`, `disposition: fix`, or `disposition: ship`. These four words are the whole vocabulary; never invent another. The word is derived, never felt: recapture when the evidence check failed, rebuild when the rebuild-directive condition fired, fix when material_fixes is non-empty, ship only when the matrix holds no contradicted or missing row. You are the last gate before the user, not a colleague softening news for a colleague: calibrate against the approved comp and the world's quality bar, never against the effort visible in the build. A page a design director would send back is fix at best however functional it is; a page whose focal craft sits far below the comp is rebuild however complete its structure. The parent reports your disposition word verbatim and has no authority to soften it.
 
 ## Output Contract
 
-Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). Missing inputs are named in one line above the sections. No praise, no summary prose.
+Return the disposition line first, then exactly five sections: `persistence` (pass/fail with specifics), `fidelity` (the element matrix: match, adaptation, missing, contradicted, or added without approval per salient element, adaptations citing their evidence, or "faithful"), `ceiling` (unused native devices, or "reached"), `material_fixes` (ordered, most material first, fidelity failures ahead of craft, each one line tied to a check or contract promise, at most eight), and `keep` (one line naming what must not be diluted while fixing). A recapture return replaces the five sections with the single `recapture` section from check 0. Missing inputs are named in one line above the sections. No praise, no summary prose.
 
 ## Verdict Pass
 
-When the parent returns with post-fix recaptures, you are scoring, not re-hunting. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths for this round; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open; unresolved or partial material findings can never recompute to ship.
+When the parent returns with post-fix recaptures, you are scoring, not re-hunting. Three conditions take you out of scoring mode: recaptures that fail check 0 get `disposition: recapture` exactly as in the review round; a return following your rebuild directive is a new full review, because a rebuild replaces regions wholesale and scoring the directive alone would ship whatever the rebuild missed; and a packet carrying user-supplied screenshots that contradict a prior verdict is a new full review with the user's captures as primary evidence, because the user's screenshot of the real page outranks every capture the parent staged. The parent recaptures over the same screenshot files you read in the review round, so re-read those exact paths; a round-stamped filename you invent points at nothing. The parent's narration of what was fixed is not evidence; a claimed fix you cannot see in the recaptures is unresolved. For each material fix from your review, one line: resolved, partial, or unresolved, tied to what the new screenshots visibly show; a fix answered mechanically, positions moved but the quality the finding named still absent, is partial at best. Then name at most three regressions the fix batch itself introduced, judged by the same matrix rules, and nothing else; no new hunt, no new checks. Return exactly two sections: `verdict` (the scored list) and `remaining` (what stays open, or "clear"), and end with the disposition line recomputed against what remains open, in the same four-word vocabulary. Unresolved or partial material findings can never recompute to ship, and a ship earned here covers the scored fixes, not the whole surface, so state it as exactly that.

.claude/skills/impeccable/reference/hooks.md

@@ -2,9 +2,9 @@
 
 Manage the **design detector hook** for the current project.
 
-The hook runs the impeccable design detector on direct file edits to design-relevant files (`.tsx`, `.jsx`, `.html`, `.vue`, `.svelte`, `.astro`, `.css`, `.scss`, `.sass`, `.less`, `.ts`, `.js`). Claude Code, Codex, and GitHub Copilot use a post-tool-use hook and push a short system reminder into the agent's context after the edit; findings get a correction prompt, pending issues get a re-nudge, and clean UI-ish files get a short ack unless quiet mode is on (`hook.quiet` in config). Plain `.ts` and `.js` files are still scanned, but stay quiet unless the detector finds something. Cursor uses `preToolUse` to block bad proposed writes before they land and stays silent when it allows a clean write.
+The hook runs the impeccable design detector on direct file edits to design-relevant files (`.tsx`, `.jsx`, `.html`, `.vue`, `.svelte`, `.astro`, `.css`, `.scss`, `.sass`, `.less`, `.ts`, `.js`). Claude Code, Codex, and GitHub Copilot use a post-tool-use hook and push a short system reminder into the agent's context after the edit; findings get a correction prompt, pending issues get a re-nudge, and clean UI-ish files get a short ack unless quiet mode is on (`hook.quiet` in config). Plain `.ts` and `.js` files are still scanned, but stay quiet unless the detector finds something. Cursor uses `preToolUse` to block bad proposed writes before they land and stays silent when it allows a clean write. Grok Build fires the same PostToolUse scan to mark touched files, then surfaces findings on Stop `additionalContext`. Do not expect a Grok per-edit reminder: Grok discards that stdout.
 
-The detector rules run in two tiers. The per-edit hook surfaces only the immediate tier: mechanical, unambiguous problems worth interrupting an edit for, such as broken images, overflowing or clipped content, contrast and legibility failures, gradient text, glow shadows, and design-system drift. Everything else (copy cadence, palette and typography taste, layout rhythm) is deferred to a deep pass on the `Stop` hook event, which runs the full rule set over every UI file touched in the session and surfaces the remaining findings once, deduplicated against what the per-edit pass already reported. A session with nothing left to report stops silently. Set `hook.perEditRules` to `"all"` in `.impeccable/config.json` to restore the full rule set on every edit. The Stop deep pass is wired for Claude Code and Codex, which both dispatch a native `Stop` hook event. Cursor does not get one (its stop hook is not consistently dispatched; the pre-write gate covers it), and GitHub Copilot's stop-style events do not feed context back to the model, so they keep the full detector per edit.
+The detector rules run in two tiers. The per-edit hook surfaces only the immediate tier: mechanical, unambiguous problems worth interrupting an edit for, such as broken images, overflowing or clipped content, contrast and legibility failures, gradient text, glow shadows, and design-system drift. Everything else (copy cadence, palette and typography taste, layout rhythm) is deferred to a deep pass on the `Stop` hook event, which runs the full rule set over every UI file touched in the session and surfaces the remaining findings once, deduplicated against what the per-edit pass already reported. A session with nothing left to report stops silently. Set `hook.perEditRules` to `"all"` in `.impeccable/config.json` to restore the full rule set on every edit. The Stop deep pass is wired for Claude Code, Codex, and Grok Build, which dispatch a native `Stop` hook event. Cursor does not get one (its stop hook is not consistently dispatched; the pre-write gate covers it), and GitHub Copilot's stop-style events do not feed context back to the model, so they keep the full detector per edit. Grok also fires an observe-only Stop with `reason: "shutdown"` after `end_turn`; skip that one, scan only `end_turn`.
 
 Every hook is a mechanical pass. The reflexes no scanner catches live in [craft-floor.md](craft-floor.md), which the skill loads before it edits UI, so they apply whether or not a hook is wired. A session with no automatic hook gets one `MANUAL_DETECTOR_REQUIRED` directive from `context.mjs` asking for a single detector run at the end.
 
@@ -14,7 +14,7 @@ Declare server-side template extensions under **`detector.extensions`** when the
 
 Manual `npx impeccable detect` scans use the same project filter config by default: `detector.ignoreRules`, `detector.ignoreFiles`, `detector.ignoreValues`, and `detector.designSystem.enabled`. `hook.enabled` only controls automatic hook execution, not manual CLI scans. Use `npx impeccable detect --no-config ...` for a raw detector run that ignores project config/context. Use `npx impeccable ignores ...` for direct CLI CRUD on the same detector ignores.
 
-Supported harnesses: Claude Code (`.claude/settings.local.json` in the project, which is gitignored so the hook stays machine-local; a hook you move into the shared `settings.json` is honored in place too), Codex (`.codex/hooks.json` in the project), Cursor (`.cursor/hooks.json` in the project), and GitHub Copilot (`.github/hooks/impeccable.json` in the project, a team-shared committed file that both the Copilot CLI and the cloud agent read). For the Copilot CLI, repo-level hooks fire once `.github/hooks/impeccable.json` is committed to the repository's default branch.
+Supported harnesses: Claude Code (`.claude/settings.local.json` in the project, which is gitignored so the hook stays machine-local; a hook you move into the shared `settings.json` is honored in place too), Codex (`.codex/hooks.json` in the project), Cursor (`.cursor/hooks.json` in the project), Grok Build (`.grok/hooks/impeccable.json` in the project; requires `/hooks-trust` or `--trust`), and GitHub Copilot (`.github/hooks/impeccable.json` in the project, a team-shared committed file that both the Copilot CLI and the cloud agent read). For the Copilot CLI, repo-level hooks fire once `.github/hooks/impeccable.json` is committed to the repository's default branch.
 
 On **Cursor**, `preToolUse` checks proposed Write/Edit/Shell write content and denies only when the real detector finds an issue. The denial message is visible to the agent as the tool error, so the agent can reconsider before the bad write lands.
 
@@ -44,7 +44,7 @@ The first argument is the action. Defaults to `status`.
    ```
 
 3. If `<action>` is `off`, follow up with a one-line note: "Done. New edits will not trigger the design hook in this project until you run `/impeccable hooks on`."
-4. If `<action>` is `on`, follow up with: "Done. The design hook will fire after the next Edit/Write/MultiEdit on a UI file."
+4. If `<action>` is `on`, follow up with: "Done. The design hook will fire after the next Edit/Write on a UI file."
 5. If `<action>` is `ignore-value`, `ignore-file`, or `ignore-rule`, just print the script output. The default scope is shared `.impeccable/config.json`; add `--local` only when the user explicitly asks for a private exception.
 6. If `<action>` is `status`, just print the script output. Do not add commentary unless the user asked a follow-up question.
 

.claude/skills/impeccable/reference/new-work.md

@@ -1,34 +1,34 @@
 # New visual work
 
-Use this flow when making a new surface or replacing a visual identity. PRODUCT.md owns product truth. DESIGN.md owns durable visual decisions. A surface brief keeps strategy that belongs only to one route or artifact. Complete [init.md](init.md) first when PRODUCT.md is missing; a missing DESIGN.md does not route back to init.
+Use this flow for a new surface or a replacement visual identity. PRODUCT.md owns product truth. DESIGN.md owns durable visual decisions. A surface brief keeps strategy that belongs to one route or artifact. Complete [init.md](init.md) first when PRODUCT.md is missing; a missing DESIGN.md does not route back to init.
 
 ## 1. Decide what is already true
 
 Read DESIGN.md, representative code, tokens, components, and assets.
 
 - **Redesign:** preserve product truth, content, function, constraints, and explicit brand commitments; replace the old visual world rather than polishing it. The old look is evidence of what the subject is, not authority over what it becomes.
-- **Established world:** inherit it. A missing DESIGN.md does not erase a coherent identity already present in code; document that identity instead of inventing a replacement.
-- **Incomplete brand:** preserve confirmed assets and recognizable traits, then help the user expand the system for this new surface.
+- **Established world:** inherit it. A missing DESIGN.md does not erase a coherent identity already in code; document that identity instead of inventing a replacement.
+- **Incomplete brand:** preserve confirmed assets and recognizable traits, then expand the system with the user for this surface.
 - **No visual authority:** create a new world with the user.
 
-A section, component, feature, or state inside an established surface inherits that surface. Do not turn a local addition into a new identity exercise.
+A section, component, feature, or state inside an established surface inherits that surface. Never turn a local addition into a new identity exercise.
 
 ## 2. Ask what will change the work
 
 Ask one round of two or three related questions through the structured question tool when available. Skip settled facts; a precise request may need only a compact confirmation.
 
-- **Persuade:** clarify who must act, what they should believe, and which real proof, content, or assets can earn that belief.
-- **Operate:** clarify the task, information, important states, frequency, and constraints.
-- **Read:** clarify the reader's question, source material, structure, and wayfinding.
-- **Experience:** clarify what leads, how exploration unfolds, and which interaction or transition matters.
+- **Persuade:** who must act, what they should believe, which real proof, content, or assets earn that belief.
+- **Operate:** the task, information, important states, frequency, constraints.
+- **Read:** the reader's question, source material, structure, wayfinding.
+- **Experience:** what leads, how exploration unfolds, which interaction or transition matters.
 
-Across modes, ask what success looks like, what must remain untouched, and what would make a polished result feel wrong. Do not ask for CSS values or canned aesthetic lanes.
+Across modes, ask what success looks like, what must remain untouched, and what would make a polished result feel wrong. Never ask for CSS values or canned aesthetic lanes.
 
 ## 3. Choose the right amount of invention
 
 ### Extend an existing surface
 
-Inherit its world and composition. Resolve only the new purpose, content, hierarchy, states, interaction, and how the addition joins the surrounding experience. Do not run a concept tournament or change DESIGN.md unless the user approves a durable system change.
+Inherit its world and composition. Resolve only the new purpose, content, hierarchy, states, interaction, and how the addition joins the surrounding experience. No concept tournament, and no DESIGN.md change unless the user approves a durable system change.
 
 ### Create a whole surface inside an established world
 
@@ -36,21 +36,21 @@ Keep the visual system fixed. Derive five to seven materially different structur
 
 `node .claude/skills/impeccable/scripts/concept-seed.mjs --scope surface --mode <mode>`
 
-The script deals three of your structures to the table; the dice decide which three reach the user, so the ranking rut stays broken while the user still holds a real choice. Present the three dealt structures on the decision page as full cards of equal salience, the dealt lead carrying kicker THE ROLL, with steer and re-roll; the user locks one in. No canon card and no pick card at surface scope: the world is settled, so every card visualizes composition, not identity. With image generation available and a comp-led default (the build-path paragraph below: `.impeccable/config.json`, the toggle handles the exception), each card declares a `comp` under `.impeccable/mocks/decision/`, generated after serving in reading order under the comp discipline in [visualize.md](visualize.md); anchor each of these comps on the established identity by passing a captured screenshot of a representative existing page as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`) beside a prompt that leads with the new surface's structure and names DESIGN.md's palette, type, and component character, because a prose paraphrase of a design system drifts where a pixel reference does not. Without image generation, or under a code-led default, each card instead carries a `wireframe` layout schematic (see `serve-question.mjs --schema`) that the page draws itself. Locking a card is the approval and sets the build path: a locked comp builds comp-led with that comp as the approved comp, discharging [visualize.md](visualize.md)'s three-option round with no second approval point; a locked wireframe builds code-led, its ambition carried by the direction contract. Never run the script for a local extension or a precisely specified narrow request; shape those directly.
+The script deals three of your structures; the dice pick which three reach the user, breaking the ranking rut while the user keeps a real choice. Present them on the decision page as full cards of equal salience, the dealt lead under kicker THE ROLL, with steer and re-roll; the user locks one. No canon card and no pick card at surface scope: the world is settled, so every card visualizes composition, not identity. With image generation and a comp-led default (`.impeccable/config.json`; the build-path paragraph below), each card declares a `comp` under `.impeccable/mocks/decision/`, generated after serving, in reading order, under [visualize.md](visualize.md)'s comp discipline. Anchor each comp on the established identity: pass a screenshot of a representative existing page as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`) with a prompt that leads with the new surface's structure and names DESIGN.md's palette, type, and component character; prose paraphrases of a design system drift, pixel references do not. Without image generation, or under a code-led default, each card carries a `wireframe` schematic (`serve-question.mjs --schema`) the page draws itself. Locking a card is the approval and sets the build path: a locked comp builds comp-led with that comp as the approved comp, discharging [visualize.md](visualize.md)'s three-option round with no second approval point; a locked wireframe builds code-led, its ambition carried by the direction contract. Never run the script for a local extension or a precisely specified narrow request; shape those directly.
 
 ### Create or replace the visual world
 
-1. Name the product's unique mechanism in one sentence, the audience's real scene, its cultural home, and what this first surface must prove. Note the page this category always ships and its predictable opposite; name both as the rut and keep them out of the seven-candidate list. A brief that paints its own picture, a product name, a titled artifact, a governing metaphor, adds its literal reading to the rut: spend at most one candidate on it and derive the rest from elsewhere in the audience's world.
-2. From that cultural world, list seven concrete visual systems, artifacts, places, or rituals the audience knows by heart, each with one line on why it resonates and can carry the mechanism, ordered by resonance. The audience's world includes its graphic and screen traditions, not only its physical objects: the notation, publications, identity programs, data graphics, and interfaces it reads daily; a nameable abstract system (a school of poster, a documentation standard) is as concrete a candidate as any artifact. What would this thing look like as a physical object; what did its world look like before the web? Near-duplicates count once. When more than three of the seven share one material family, the derivation stopped at the subject's most obvious artifact; dig until the list spans at least three families.
+1. Name the product's unique mechanism in one sentence, the audience's real scene, its cultural home, and what this first surface must prove. Note the page this category always ships and its predictable opposite; both are the rut, kept out of the seven-candidate list. A brief that paints its own picture, a product name, a titled artifact, a governing metaphor, adds its literal reading to the rut: spend at most one candidate on it and derive the rest from elsewhere in the audience's world.
+2. From that cultural world, list seven concrete visual systems, artifacts, places, or rituals the audience knows by heart, each with one line on why it resonates and can carry the mechanism, ordered by resonance. The audience's world includes its graphic and screen traditions, not only its physical objects: the notation, publications, identity programs, data graphics, and interfaces it reads daily. A nameable abstract system (a school of poster, a documentation standard) is as concrete a candidate as any artifact. What would this thing look like as a physical object; what did its world look like before the web? Near-duplicates count once. When more than three of the seven share one material family, the derivation stopped at the subject's most obvious artifact; dig until the list spans at least three families.
 3. Turn that material into complete directions: each joins a reusable visual world to a concrete first-surface experience.
-4. Run `node .claude/skills/impeccable/scripts/concept-seed.mjs --scope direction --mode <mode>` and follow what it prints. This step has no substitute and no skip condition: on a new or replacement world, writing artifact code before this script has run and its assignment is acknowledged is a contract violation, whatever the harness, the model, or the time pressure, because the roll is the mechanism that keeps every run from converging on the category default. The script assigns which direction gets built and deals catalog challengers. Fuse each challenger before judging it: the challenger supplies the form and its system grammar, the product supplies every fact, and clarity wins conflicts. Weigh fused challengers against the assigned direction on exactly two axes, audience identification and product clarity; losing to strong grounded material is a valid outcome, and beating a thin or tool-monoculture list is the point. The weighing closes with a verdict per challenger, decided before any borrowing is considered: wins (beats the assigned direction on both axes; it becomes the build candidate), competitive (holds one axis; it stays a full alternate), or declined (loses both). A declined challenger is not spent: name the one discipline of its system the assigned direction lacks, and raise the assigned direction to match before presenting it. A donation transfers ambition and system discipline (a palette's total commitment, a grid's density courage, a form's structural honesty), never the challenger's clothes; a motif lifted from a declined world is a costume note, not a raise, and one world owns the page. Write each raise into the presented direction as its own line, named for its donor; a raise nobody can read did not happen.
-5. Present one direction, fully committed and already raised by the hand it beat, its raises visible as named lines: its world, first viewport, visitor path, signature interaction, cross-surface reach, and honest risk. Alongside it, route each dealt challenger by its verdict: winning and competitive challengers are full alternates carrying their QUALITY BAR cards and one-line case, while declined challengers render demoted, compact and quiet, each carrying its verdict plus what the direction kept from it, never full-size and never silently dropped, each still adoptable on request. The verdict informs the user's choice, it never pre-empts it; the demoted row is the hand's proof of judgment, showing why the dealt worlds made the presented direction better. A hand holds at most three full-card challengers: when the roll deals more, the three strongest join the hand and the rest wait in the re-roll pool, noted in one line; dropping a challenger from the hand itself takes a named product-truth failure, disclosed. Add one card for your own top-ranked grounded candidate when it is not the assigned direction, kicker IMPECCABLE’S PICK, same anatomy as every card, with an honest risk line naming its familiarity when true: the strongest grounded direction is often the one most runs in this category land on, and the user deciding that trade is the point of showing it. Familiar and effective is a legitimate destination, not a failure of nerve; the pick card and the standing exit serve it at two depths. One pick card, never two, never a ranked list: the rest of your grounded candidates stay yours, because a lineup of them hands selection back to a taste function and invites the safest card. The pick never takes the lead position, and when the dice assign your top candidate there is no pick card; the assigned card notes it also topped your list. Add re-roll with an optional one-line steer, offered in three registers: plain (a fresh hand, same spread), safer (the familiar register: your remaining conventional grounded candidates plus the canon against named competitors), and bolder (foreign forms only, at full commitment). A register is the user's steering on the familiar-to-bold axis, never yours to pre-select; when the answer carries one, re-run the seed with `--register <value>` and the next `--reroll` round, and follow what it prints. A user saying "bolder" or "safer" while a direction round is open means these registers, never the bolder or harden commands. The two channels share this structure and differ only in richness: cards and boards on the decision page, names and one-liners through the structured tool; the structured tool's option list carries the assigned direction, the pick, the winning and competitive challengers, and the standing exit as its last option, while declined challengers fold into the assigned option's description as their kept lines, so the raise survives the text channel too.
+4. Run `node .claude/skills/impeccable/scripts/concept-seed.mjs --scope direction --mode <mode>` and follow what it prints. No substitute, no skip: on a new or replacement world, writing artifact code before this script has run and its assignment is acknowledged is a contract violation, whatever the harness, the model, or the time pressure; the roll is what keeps every run from converging on the category default. The script assigns the direction to build and deals catalog challengers. Fuse each challenger before judging it: the challenger supplies the form and its system grammar, the product supplies every fact, clarity wins conflicts. Weigh fused challengers against the assigned direction on exactly two axes, audience identification and product clarity. Losing to strong grounded material is a valid outcome; beating a thin or tool-monoculture list is the point. Close with a verdict per challenger, decided before any borrowing: wins (beats the assigned direction on both axes; becomes the build candidate), competitive (holds one axis; stays a full alternate), or declined (loses both). A declined challenger is not spent: name the one discipline of its system the assigned direction lacks, and raise the assigned direction to match before presenting it. A donation transfers ambition and system discipline (a palette's total commitment, a grid's density courage, a form's structural honesty), never the challenger's clothes; a lifted motif is a costume note, not a raise, and one world owns the page. Write each raise into the presented direction as its own line, named for its donor; a raise nobody can read did not happen.
+5. Present one direction, fully committed and already raised by the hand it beat, raises visible as named lines: world, first viewport, visitor path, signature interaction, cross-surface reach, honest risk. Route each challenger by verdict: winning and competitive challengers are full alternates with their QUALITY BAR cards and one-line case; declined challengers render demoted, compact and quiet, each carrying its verdict and what the direction kept from it, never full-size, never silently dropped, still adoptable on request. The verdict informs the user's choice, never pre-empts it; the demoted row is the hand's proof of judgment. A hand holds at most three full-card challengers: when the roll deals more, the three strongest join and the rest wait in the re-roll pool, noted in one line; dropping a challenger from the hand itself takes a named product-truth failure, disclosed. Add one card for your own top-ranked grounded candidate when it is not the assigned direction, kicker IMPECCABLE’S PICK, same anatomy as every card, with an honest risk line naming its familiarity when true: the strongest grounded direction is often where most runs in this category land, and the user deciding that trade is the point of showing it. Familiar and effective is a legitimate destination, not a failure of nerve; the pick card and the standing exit serve it at two depths. One pick card, never two, never a ranked list: a lineup of your candidates hands selection back to a taste function and invites the safest card. The pick never takes the lead position; when the dice assign your top candidate there is no pick card, and the assigned card notes it topped your list. Add re-roll with an optional one-line steer, in three registers: plain (a fresh hand, same spread), safer (your remaining conventional grounded candidates plus the canon against named competitors), bolder (foreign forms only, at full commitment). The register is the user's steering on the familiar-to-bold axis, never yours to pre-select; when the answer carries one, re-run the seed with `--register <value>` and the next `--reroll` round, and follow what it prints. A user saying "bolder" or "safer" while a direction round is open means these registers, never the bolder or harden commands. The two channels share this structure and differ only in richness: cards and boards on the decision page, names and one-liners through the structured tool, whose option list carries the assigned direction, the pick, the winning and competitive challengers, and the standing exit last; declined challengers fold into the assigned option's description as their kept lines, so the raise survives the text channel.
 
-The standing exit: every direction round offers one quiet, permanent alternative, the category standard, played straight. It is the user's door, never yours: never recommend it, never weigh it against the roll, never let it soften the dealt directions; the counterweights bind the unchosen default, not the chosen one. When the user takes it, in the canon action, a safer-steer, or plain words asking for the familiar or competitor-like path, convention becomes the commitment: ask once for two or three products this should sit alongside, make their craft level the bar, and execute the canon at full fidelity, without irony or smuggled quirk. A standing preference gets recorded as a brand commitment in PRODUCT.md. Re-roll eliminates every direction already shown, grounded and challenger alike; after two consecutive re-rolls, ask what quality is missing. You may re-roll on your own only on named factual grounds, when the assigned direction cannot carry the product's truth or task; taste is never grounds. The user may re-roll freely, and a user- or brief-pinned direction beats the roll, always. Present the decision visually: write an options payload with the assigned direction leading and its raised lines included, the pick card when one exists, the dealt challengers as alternates carrying their QUALITY BAR cards plus each challenger's verdict and kept line, re-roll with its safer and bolder registers, steer, plus canon enabled, and `buildPath` carrying the recorded default with `toggle: true` whenever image generation exists (the build-path paragraph below owns the details); a degraded roll with no challengers still uses the page, as a single text-only card with re-roll. Give every card the same anatomy, thesis, palette, materials, first viewport, honest risk, and the challengers' case lines (run the script with `--schema` for the exact shape); the page renders identity from these fields, routes declined challengers to a demoted row on its own, and a challenger's catalog image rides as labeled inspiration, never as the promise of the build. Author `canonCard` too: the category standard as one honest card with the same anatomy; the page keeps it subordinate, and the counterweights still bind you. Run `node .claude/skills/impeccable/scripts/serve-question.mjs --start --payload <file>` (run it with `--schema` first for the exact payload shape). It daemonizes, prints the page URL and a key, and exits immediately; now open that URL for the user, in-app browser first, then the system opener, then showing the URL. Collect the choice with `--wait --key <key>`, repeating while it exits 3; the ANSWER prints as JSON. Exit 4 means the page was closed without an answer: re-present once through the structured question tool, and with no answer there either, proceed unattended with the assigned direction and state the assumptions. A harness that can leave a shell blocked in the background may instead run the script without `--start` and let it auto-open and block. The fallback to the structured question tool is never yours to predict: run the script, and only exit code 2 from starting it routes the decision there; treat that exit as the fallback, never as an error to retry.
+The standing exit: every direction round offers one quiet, permanent alternative, the category standard, played straight. It is the user's door, never yours: never recommend it, never weigh it against the roll, never let it soften the dealt directions; the counterweights bind the unchosen default, not the chosen one. When the user takes it (the canon action, a safer-steer, or plain words asking for the familiar or competitor-like path), convention becomes the commitment: ask once for two or three products this should sit alongside, make their craft level the bar, and execute the canon at full fidelity, without irony or smuggled quirk. Record a standing preference as a brand commitment in PRODUCT.md. Re-roll eliminates every direction already shown, grounded and challenger alike; after two consecutive re-rolls, ask what quality is missing. Re-roll on your own only on named factual grounds, when the assigned direction cannot carry the product's truth or task; taste is never grounds. The user may re-roll freely, and a user- or brief-pinned direction beats the roll, always. Present the decision visually: write an options payload with the assigned direction leading, its raised lines included; the pick card when one exists; the dealt challengers as alternates with their QUALITY BAR cards, verdicts, and kept lines; re-roll with its safer and bolder registers; steer; canon enabled; and `buildPath` carrying the recorded default with `toggle: true` whenever image generation exists (details in the build-path paragraph below). A degraded roll with no challengers still uses the page, as a single text-only card with re-roll. Give every card the same anatomy: thesis, palette, materials, first viewport, honest risk, and the challengers' case lines (`--schema` prints the exact shape); the page renders identity from these fields, demotes declined challengers to their row on its own, and a challenger's catalog image rides as labeled inspiration, never the promise of the build. Author `canonCard` too: the category standard as one honest card, same anatomy; the page keeps it subordinate, and the counterweights still bind you. Run `node .claude/skills/impeccable/scripts/serve-question.mjs --start --payload <file>` (`--schema` first for the payload shape). It daemonizes, prints the page URL and a key, and exits; open that URL for the user, in-app browser first, then the system opener, then showing the URL. Collect the choice with `--wait --key <key>`, repeating while it exits 3; the ANSWER prints as JSON. An ANSWER of `{"optionId":"reroll"}` keeps the server alive and the page open on a loading hand: rerun concept-seed with the same `--scope` and `--mode` plus `--from <seed-key> --reroll <n>` (1 on the first re-roll, counting up), build the next payload, deliver it with `--update --key <same key> --payload <file>`, then return to `--wait` on that key. Never `--start` a second server or fall back to chat here: either strands the open tab on a hand that never arrives. Exit 4 means the page closed unanswered: re-present once through the structured question tool, and with no answer there either, proceed unattended with the assigned direction and state the assumptions. A harness that can leave a shell blocked in the background may run the script without `--start` and let it auto-open and block. Never predict the fallback: run the script, and only exit code 2 from starting it routes the decision to the structured tool; that exit is the fallback, never an error to retry.
 
-When image generation exists, every card also declares a `comp` path under `.impeccable/mocks/decision/`, the canon card included. Where the harness sandboxes its shell, start the page through the least-sandboxed command path it offers: a sandboxed shell cannot bind the board's port, and the first-attempt failure costs a retry every session. Serve the page first, then produce the comps; the page shimmer-waits per slot and the user may answer before they land. Each card's image is that direction's north-star comp at full fidelity, produced under the comp discipline in [visualize.md](visualize.md): the requested surface's first viewport, structure-led prompt, real product name and real content, no invented commercial claims, in that card's own palette, type character, and material world, committed all the way; visualize.md's self-checks bind decision comps identically. Generation takes the same time at any fidelity, so an unfinished draft pays draft quality for comp cost; fairness between cards comes from equal fidelity in each card's own grammar, one surface, one aspect, never from shared unfinishedness. The frame's aspect is the surface's own: a native app or mobile-first surface comps portrait at its device viewport, a desktop web surface landscape, and the decision page adapts to either, so a phone screen comped landscape is a broken frame, not a neutral default. Produce in the order the user reads, the assigned card, then the pick, then the full-card hand, then canon, each file written with its prompt sidecar the moment it is done, so a re-roll's spend front-loads onto the cards read first; declined challengers get no comp, their catalog thumb is their face. When the harness runs subagents in parallel, fan the set out as one agent per card: each spawn is the shipped asset producer with a single-comp packet, that card's fields, PRODUCT.md, the shared frame, and the card's declared path, up to four in flight at once. A slot still empty when its agent returns is regenerated inline, and a slot still empty when the user answers is dropped without ceremony; no other supervision is owed. Without parallel subagents, generate in the main thread after serving, in the same reading order, and let the harness's own generation display carry the progress; the wait for the answer follows the last file. The chosen card's comp is not spent by the choice: on a comp-led build it enters the comp round as compositional option one, and on a code-led build it returns at the finish review as the critique reference, what the image dared that the build did not. The unchosen comps stay in `.impeccable/mocks/decision/` as the round's spent hand; they carry no approval and imply none. With no image generation, the cards carry their identity in palette chips and facts, and that page is complete, not a lesser version; the page then also demotes every challenger's catalog art to a labeled thumbnail on its own, because salience must encode the verdict, never the accident of which cards have images.
+When image generation exists, every card also declares a `comp` path under `.impeccable/mocks/decision/`, the canon card included. Where the harness sandboxes its shell, start the page through the least-sandboxed command path it offers: a sandboxed shell cannot bind the board's port, and the first-attempt failure costs a retry every session. Serve the page first, then produce the comps; the page shimmer-waits per slot and the user may answer before they land. Each card's image is that direction's north-star comp at full fidelity under [visualize.md](visualize.md)'s comp discipline: the requested surface's first viewport, structure-led prompt, real product name and real content, no invented commercial claims, in that card's own palette, type character, and material world, committed all the way. Generation takes the same time at any fidelity, so an unfinished draft pays comp cost for draft quality; fairness between cards is equal fidelity in each card's own grammar, one surface, one aspect, never shared unfinishedness. The frame's aspect is the surface's own: portrait at device viewport for a native app or mobile-first surface, landscape for desktop web; the decision page adapts to either, and a phone screen comped landscape is a broken frame, not a neutral default. Produce in reading order, the assigned card, then the pick, then the full-card hand, then canon, each file written with its prompt sidecar the moment it is done, so a re-roll's spend front-loads onto the cards read first; declined challengers get no comp, their catalog thumb is their face. With parallel subagents, fan out one agent per card: each spawn is the shipped asset producer with a single-comp packet, that card's fields, PRODUCT.md, the shared frame, and the card's declared path, up to four in flight. Regenerate inline any slot still empty when its agent returns; drop without ceremony any slot still empty when the user answers. No other supervision is owed. Without parallel subagents, generate in the main thread after serving, same order, and let the harness's own generation display carry the progress; the wait for the answer follows the last file. The chosen card's comp is not spent by the choice: comp-led, it enters the comp round as compositional option one; code-led, it returns at the finish review as the critique reference, what the image dared that the build did not. Unchosen comps stay in `.impeccable/mocks/decision/` as the round's spent hand; they carry no approval and imply none. With no image generation, cards carry their identity in palette chips and facts, and that page is complete, not a lesser version; the page then also demotes every challenger's catalog art to a labeled thumbnail on its own, because salience must encode the verdict, never the accident of which cards have images.
 
-The execution contract, comp-led or code-led, is a workflow preference, not a per-surface decision, so no round asks it: the recorded default rides every round and the page's toggle handles the exception. Read the default from `.impeccable/config.json` (`buildPath`), with the gitignored `.impeccable/config.local.json` winning where one machine differs from the team's committed value; with neither, comp-led is the default whenever image generation exists. Author every direction and surface payload with `buildPath: { "value": <default>, "toggle": true }`; the page renders a footer toggle with the trade stated beside it, and the ANSWER returns `buildPath` plus `buildPathFlipped`. A flipped value binds that session only and is never written back, with one exception, and it is the only thing inside a round that earns a question about this preference (init records it up front on projects that get the chance): when `buildPathFlipped` comes back true on a project that records no `buildPath` at all, ask once after the round closes whether to keep it as the standing default. Either answer ends in a write to `.impeccable/config.json`; the answer picks the value, never whether to record one. Yes writes the flipped value, and "no, just this once" writes the value they flipped away from, which is the standing default they just confirmed by declining. Ask on the flip and never on the untouched default, because a user who left the toggle alone has told you nothing. A declined offer nothing writes down is an offer the next session makes again. When the user asks in words to change the standing default, update the file without asking. **Comp-led**: the chosen card's comp is law, generated before building when it does not exist yet, and the finish review audits the build against it; boldest composition on the table, fix rounds expected; comp-led makes the comp non-optional, no silent skipping. **Code-led**: no comp of this page and no apology for it; the QUALITY BAR boards still calibrate finish, and the ambition moves into the written contract, the FIRST VIEWPORT block plus a named signature interaction and motion grammar, which the finish reviewer audits in behavior; code-led is not a discount on commitment, the direction still lands fully committed in code. A code-led round still declares each card's comp path as a flip reserve: when the user flips the toggle to comp mid-round, `--wait` returns once with BUILD PATH FLIPPED while the page shimmers the slots; generate each open card's comp into its declared path then, lead first, and wait again. The flip back is free, and a comp that already rendered rides at the finish review as the critique reference. Without image generation there is no toggle and no choice: code-led is the only path, stated in one line rather than asked. The old two-card execution-contract round is retired; `followup: true` remains the general mechanism for delivering any later round over the same table via `--update`.
+The execution contract, comp-led or code-led, is a workflow preference, not a per-surface decision; no round asks it. The recorded default rides every round and the page's toggle handles the exception. Read the default from `.impeccable/config.json` (`buildPath`), the gitignored `.impeccable/config.local.json` winning where one machine differs from the team's committed value; with neither, comp-led is the default whenever image generation exists. Author every direction and surface payload with `buildPath: { "value": <default>, "toggle": true }`; the page renders a footer toggle with the trade stated beside it, and the ANSWER returns `buildPath` plus `buildPathFlipped`. A flipped value binds that session only and is never written back, with one exception, the only question this preference ever earns inside a round (init records it up front on projects that get the chance): when `buildPathFlipped` comes back true on a project that records no `buildPath` at all, ask once after the round closes whether to keep it as the standing default. Either answer writes `.impeccable/config.json`; the answer picks the value, never whether to record one. Yes writes the flipped value; "no, just this once" writes the value they flipped away from, the standing default they just confirmed by declining. Ask on the flip, never on the untouched default: a user who left the toggle alone told you nothing. A declined offer nothing writes down is an offer the next session makes again. When the user asks in words to change the standing default, update the file without asking. **Comp-led**: the chosen card's comp is law, generated before building when it does not yet exist, and the finish review audits the build against it; boldest composition on the table, fix rounds expected, and the comp is non-optional, no silent skipping. **Code-led**: no comp of this page and no apology for it; the QUALITY BAR boards still calibrate finish, and the ambition moves into the written contract, the FIRST VIEWPORT block plus a named signature interaction and motion grammar, which the finish reviewer audits in behavior; code-led is not a discount on commitment. A code-led round still declares each card's comp path as a flip reserve: when the user flips the toggle to comp mid-round, `--wait` returns once with BUILD PATH FLIPPED while the page shimmers the slots; generate each open card's comp into its declared path then, lead first, and wait again. The flip back is free, and a comp that already rendered rides at the finish review as the critique reference. Without image generation there is no toggle and no choice: code-led is the only path, stated in one line rather than asked. The old two-card execution-contract round is retired; `followup: true` remains the general mechanism for delivering any later round over the same table via `--update`.
 
 Catalog worlds are working systems, not mood references. When one survives, carry its palette and material, type and composition, topology, controls and state, and responsive rules into the product. When the source is itself an interface language, commit to its native grammar across navigation, content, controls, and states. Open the QUALITY BAR board and hero for the world you build the moment the choice lands, even if you viewed another card earlier; the ANSWER line names the chosen card's images (when the harness only reads files or runs sandboxed, download them into the workspace and open the relative path; sandboxed viewers reject absolute paths outside it). They set the craft level the build must reach, a rendered reference's finish, commitment, and art direction, never the composition; your surface serves this product.
 
@@ -64,13 +64,13 @@ Pick a color strategy before picking colors: Restrained (neutrals plus one accen
 
 Choose faces like objects from the subject's world, in the mode's register. Operate and Read surfaces are well served by system stacks and workhorse UI faces; Persuade and Experience surfaces want faces with a point of view, and these training-data defaults mean you stopped looking: Fraunces, Playfair Display, Cormorant, Lora, Crimson, Newsreader, Syne, Space Grotesk, Space Mono, IBM Plex, Inter-as-display, DM Sans, DM Serif, Outfit, Plus Jakarta Sans, Instrument Sans. Naming one of these faces anyway requires a reason no other face could satisfy, and a subject association is never that reason: books wanting a serif, bookshops wanting hand-lettering, and tech wanting a mono are the associations the list exists to break.
 
-Calibration: AI-generated interfaces cluster around a few looks regardless of subject: warm cream ground, high-contrast serif display, and a terracotta or signal-red accent; near-black with one neon accent and glowing edges; broadsheet-editorial hairlines, italic display serif, and small tracked mono labels. All are legitimate when the brief calls for them. Where the brief leaves the aesthetic free, landing in one means the self-check failed: if someone could guess your aesthetic from the category alone, or from category-plus-avoidance, rework until neither answer is obvious. Energy is not the enemy of trust: a brief's negative constraints (no gamification, no hype) rule out those devices, not exuberance, and adjectives describing the product's behavior (quiet support, calm coaching) do not dictate the surface's energy. A bookish, warm, or child-facing subject does not soften the calibration: book cloth, thread, jackets, endpapers, and shelf ephemera span the whole saturated spectrum, and cream paper is the smallest corner of that world; landing on cream plus serif for a book subject is the default wearing the subject's clothes. A brief-pinned world pins the world, not its softest rendition: the pinned world's full material range stays in play, and a rendition that matches what any model ships for that world failed the self-check at execution rather than selection.
+Calibration: AI-generated interfaces cluster around a few looks regardless of subject: warm cream ground, high-contrast serif display, and a terracotta or signal-red accent; near-black with one neon accent and glowing edges; broadsheet-editorial hairlines, italic display serif, and small tracked mono labels. All are legitimate when the brief calls for them. Where the brief leaves the aesthetic free, landing in one means the self-check failed: if someone could guess your aesthetic from the category alone, or from category-plus-avoidance, rework until neither answer is obvious. Energy is not the enemy of trust: a brief's negative constraints (no gamification, no hype) rule out those devices, not exuberance, and adjectives describing the product's behavior (quiet support, calm coaching) do not dictate the surface's energy. A bookish, warm, or child-facing subject does not soften the calibration: book cloth, thread, jackets, endpapers, and shelf ephemera span the whole saturated spectrum, and cream paper is the smallest corner of that world; landing on cream plus serif for a book subject is the default wearing the subject's clothes. A brief-pinned world pins the world, not its softest rendition: the pinned world's full material range stays in play, and a rendition matching what any model ships for that world failed the self-check at execution rather than selection.
 
 Your measured rendition prior: warm, bookish, family, and child-facing subjects come out as cream grounds, serif display with italic accents, and lamplight, even when the assigned direction never asked for them. Treat that first palette as already spent. Before writing code, reread your OWN-WORLD block: when it says cream, paper, parchment, ivory, or lamplight for a Persuade surface the brief did not pin, the rendition failed and you rework it from the world's saturated materials first. The same subject renders as bookcloth, thread, jacket, and endpaper color on other models; nothing about the subject requires your default.
 
 ## 5. Record the decision
 
-Before code, state the chosen direction as a contract in the artifact's opening comment, five short blocks, 150 words at most, in a form that survives the production build: an HTML comment in the emitted markup, never only a templating-frontmatter comment, placed as the first child of the document's body in the root layout, never inside a slotted or child component (some compilers, Astro among them, strip a slot's leading comment while keeping deeper ones). After the first production build, grep the built output for the seed key; a contract the build erased is a contract nobody can audit. THESIS: the one idea this surface owns and the category-default arrangement it refuses. OWN-WORLD: the palette and component language, specific enough to be recognizable with all content removed. STORY: what the visitor understands, believes, and does. FIRST VIEWPORT: the exact composition, what is where and at what scale, and where the primary action sits. FORM: the chosen form, its position on your ordered list, and the seed key the script printed. Close the comment with one more line, FINISH: the run's exit condition, verbatim "unreviewed and undocumented is unfinished; this build ends with the finish review, the verdict, and DESIGN.md". The comment tops the artifact you re-open on every edit, the one reminder that survives a long build: a page that looks complete with the FINISH line undischarged is not done, it is abandoned at the finish line. If a block reads like a mood, the direction is not decided yet; the finishing review audits the render against this contract.
+Before code, state the chosen direction as a contract in the artifact's opening comment, five short blocks, 150 words at most, in a form that survives the production build: an HTML comment in the emitted markup, never only a templating-frontmatter comment, placed as the first child of the document's body in the root layout, never inside a slotted or child component (some compilers, Astro among them, strip a slot's leading comment while keeping deeper ones). After the first production build, grep the built output for the seed key; a contract the build erased is a contract nobody can audit. THESIS: the one idea this surface owns and the category-default arrangement it refuses. OWN-WORLD: the palette and component language, specific enough to be recognizable with all content removed. STORY: what the visitor understands, believes, and does. FIRST VIEWPORT: the exact composition, what is where and at what scale, and where the primary action sits. FORM: the chosen form, its position on your ordered list, and the seed key the script printed. Close with one more line, FINISH: the run's exit condition, verbatim "unreviewed and undocumented is unfinished; this build ends with the finish review, the verdict, DESIGN.md, and every shipping raster carrying its provenance". The comment tops the artifact you re-open on every edit, the one reminder that survives a long build: a page that looks complete with the FINISH line undischarged is not done, it is abandoned at the finish line. If a block reads like a mood, the direction is not decided yet; the finishing review audits the render against this contract.
 
 On a new or replacement world, DESIGN.md is written at finish, from the built world, by the shipped documenter (section 7); a rulebook written before the build gets defended against reality instead of describing it, and hands the design-system detector an unstable target. A new world shipped with no DESIGN.md is still an incomplete run. An ordinary extension does not rewrite DESIGN.md.
 
@@ -82,18 +82,18 @@ If the work establishes durable strategy for a route or artifact, read its exist
 
 Keep the brief small: scope and visitor mode; audience, job, action/task, proof/content, and constraints; chosen direction and memorable moment; unresolved decisions. Do not copy global product truth or DESIGN.md tokens into it.
 
-On a comp-led build, whenever any image generation is available, a harness-native tool or the API fallback context.mjs reports, the locked direction is visualized before it is built, never skipped: load [visualize.md](visualize.md) and follow it, three compositional options put before the user for approval, the chosen card's decision comp plus two variations. This step is proven to produce the most compositional and ambitious work. On a code-led build the comp round is skipped by contract, never by drift: the ambition it would have carried lives in the direction contract's FIRST VIEWPORT block and named signature interaction, and the finish reviewer audits those promises in behavior.
+On a comp-led build, whenever any image generation is available (a harness-native tool or the API fallback context.mjs reports), the locked direction is visualized before it is built, never skipped: load [visualize.md](visualize.md) and follow it, three compositional options put before the user for approval, the chosen card's decision comp plus two variations. This step is proven to produce the most compositional and ambitious work. On a code-led build the comp round is skipped by contract, never by drift: the ambition it would have carried lives in the direction contract's FIRST VIEWPORT block and named signature interaction, and the finish reviewer audits those promises in behavior.
 
 For `shape`, return the selected direction to [shape.md](shape.md) and stop before persistence or implementation.
 
 ## 6. Build with full commitment
 
-When an approved comp exists, the comp is king, and the build happens in phases. Phase one is reproduction: rebuild the comp at its own breakpoint until a screenshot at the comp's width and height overlaps it near pixel-perfectly, materials, components, elevation, assets, and implied design language included. Exactly three concessions exist: fonts (the closest obtainable face), icons (exact match unless the user already chose an icon library), and genuine defects in the generated comp such as spelling errors. Everything else must match, and models systematically believe their HTML, CSS, and SVG recreation succeeded when it did not, so the overlap comparison is the authority, never your conviction: set the screenshot beside the freshly reopened comp image at identical dimensions after every region, never beside your memory of it, and when a region keeps losing that comparison, stop recreating it in code and produce it as a rendered asset composited into the page. The comp also outranks every written record of it: when the recorded brief or inventory commits to less than the comp shows, a softer texture, a sparser field, a sculpted plate reduced to flat CSS, correct the record upward to the comp; qualifiers like subtle, restrained, and low-contrast, and counts rounded down to a comfortable fraction, are how approved materials die between approval and build. A produced material must then survive to the screen: a texture buried under a nearly opaque color wash ships the wash, not the material, so judge every material by the screenshot beside the comp, never by the stylesheet. Only when reproduction holds does phase two begin: static regions that should live become animated or interactive, reveals and motion are added, then responsiveness across the surface's devices. Where the comp does not cover the whole surface, continue building the remainder inside the comp's recorded world and design language; a component the comp never shows inherits the recorded system's corner language, line weights, and materials, and may not introduce container styles, border weights, or chrome the comp never uses.
+When an approved comp exists, the comp is king, and the build happens in phases. The comp is a spatial contract, not a mood board: only the user can downgrade its authority, in explicit words, and difficulty never infers a downgrade. Phase one is reproduction: rebuild the comp at its own breakpoint until a screenshot at the comp's width and height overlaps it near pixel-perfectly, materials, components, elevation, assets, and implied design language included. Exactly three concessions exist: fonts (the closest obtainable face), icons (exact match unless the user already chose an icon library), and genuine defects in the generated comp such as spelling errors. Everything else must match, and models systematically believe their HTML, CSS, and SVG recreation succeeded when it did not, so the overlap comparison is the authority, never your conviction: set the screenshot beside the freshly reopened comp image at identical dimensions after every region, never beside your memory of it, and when a region keeps losing that comparison, stop recreating it in code and produce it as a rendered asset composited into the page. The comp also outranks every written record of it: when the recorded brief or inventory commits to less than the comp shows, a softer texture, a sparser field, a sculpted plate reduced to flat CSS, correct the record upward to the comp; qualifiers like subtle, restrained, and low-contrast, and counts rounded down to a comfortable fraction, are how approved materials die between approval and build. A produced material must then survive to the screen: a texture buried under a nearly opaque color wash ships the wash, not the material, so judge every material by the screenshot beside the comp, never by the stylesheet. Every color the brief records gets that comparison by number, not by eye: sample the build screenshot's ground, dominant fields, and accents the same way each record was taken (an interior patch average where the record is an average, both end colors where the record is a gradient) and set each value against its recorded counterpart (sampled from the comp itself when the brief lacks one), and when a texture or tile paints over a base token, measure the net on-screen value, because the eye files a drifted color under the same color word and the number is what catches it. Judge the gap like a colorist, not a diff tool: a difference with a color name (warmer, grayer, darker than the record) is drift to fix, while a few digits of render and compression noise are the same color. Only when reproduction holds does phase two begin: static regions that should live become animated or interactive, reveals and motion are added, then responsiveness across the surface's devices. Where the comp does not cover the whole surface, continue building the remainder inside the comp's recorded world and design language; a component the comp never shows inherits the recorded system's corner language, line weights, and materials, and may not introduce container styles, border weights, or chrome the comp never uses.
 
 Build the assigned direction, not a safer interpretation of it. The form supplies structure, reading order, component conventions, and native motion; the product supplies every fact. Commit every atom: nav, buttons, inputs, and links are rebuilt in the form's vocabulary, and a stock component inside a committed form is a lapse. Land the first build fully committed; committing is the hard part, and the passes that follow exist to make the committed thing clear and effective, never to dilute it. In unattended work, the safe rendition is the known risk.
 
 - **The first viewport is a thesis, not a header.** Demonstrate the mechanism immediately, at the scale the form has in life; do not trap the concept inside a standard hero or card shell. The memory test: if someone left after one viewport, what would they describe an hour later? If the honest answer is a mood, the concept has not committed yet.
-- **Prove the hero before building past it.** When an approved comp exists, render the first viewport, capture it, and set it beside the comp's first viewport before any later section: the hero carries the run's ambition, and every following section inherits its shortfall. Judge scale and density as quantities, a field at a tenth of the comp's coverage or type at half its weight is a different design, and a five-minute retry here is what a rebuild verdict at the finish costs when this check is skipped.
+- **Prove the hero before building past it.** When an approved comp exists, render the first viewport, capture it at the comp's own pixel dimensions, and set it beside the comp's first viewport before any later section: the hero carries the run's ambition, and every following section inherits its shortfall. Save that capture as `.impeccable/review/hero-repro.png` (create the directory); the finish reviewer verifies it exists, so a skipped checkpoint is a visible checkpoint. Judge scale and density as quantities, a field at a tenth of the comp's coverage or type at half its weight is a different design, and a five-minute retry here is what a rebuild verdict at the finish costs when this check is skipped.
 - **Prove, don't claim.** Show the subject doing its job: the interface at work, the mechanism dramatized, specifics a competitor could not copy-paste. Sections that restate a claim in different words add length, not substance. Demonstration data is design material: author it at full fidelity and label it synthetic; claims stay uninventable.
 - **Author the assets; never substitute chrome.** Great surfaces live on carefully made content: names, entries, copy, covers, thumbnails, textures. In greenfield work every blank the ask round left open is yours to author at production fidelity; content is authorable, claims are labelable, no section is omittable. An unanswered commercial claim ships as a clearly marked placeholder on the user's replacement list. When image generation exists, producing the design's imagery is part of building, at the scale the composition needs: a viewport that wants atmosphere gets a full-bleed layered scene, and a library of small centered subjects standardized for tidiness forecloses it. Gradients, glass, and generic icon tiles where an authored asset belongs are the gap wearing chrome; icons drawn in the world's own grammar are the remedy, not the target.
 - **Build the form's web leverage.** When the chosen world names a technique (canvas, WebGL, view transitions, generative motion), build the technique itself, not a static imitation of it; the graceful fallback serves constrained clients, it is not the default experience.
@@ -105,8 +105,18 @@ Preserve semantics, accessibility, performance, responsiveness, project conventi
 
 ## 7. Inspect and finish
 
-Inspect the surface's target sizes in one batched screenshot round: desktop and mobile on the web; on a native platform (`ios` / `android` / `adaptive`), the shipped device classes per OS, captured from the simulator or emulator the way the platform reference's Verifying the build section describes. Critique the render against the user's request and the direction contract, fix material gaps, and confirm with one final round; two rounds is the ceiling, and fixes batch between them rather than earning per-tweak screenshots. When an approved comp exists, the critique is a side-by-side: view the comp region and the build region together, the hero and each section as its own crop at legible scale, never one full-page thumbnail, which hides exactly the failures that matter, crude controls, wrong lettering character, flattened material, behind a superficially similar section order. On a Persuade surface, verify the mode did its job: a first-time visitor should know what this is, why it matters, and what to do within seconds, in the form's own vocabulary.
+Inspect the surface's target sizes in one batched screenshot round: desktop and mobile on the web; on a native platform (`ios` / `android` / `adaptive`), the shipped device classes per OS, captured from the simulator or emulator the way the platform reference's Verifying the build section describes. When the harness reports the user's actual viewport (an in-app browser's size, a named resolution), add that width to the set: the width that breaks is the one the user sees first. Critique the render against the user's request and the direction contract, fix material gaps, and confirm with one final round; two rounds is the ceiling, and fixes batch between them rather than earning per-tweak screenshots. When an approved comp exists, the critique is a side-by-side: view the comp region and the build region together, the hero and each section as its own crop at legible scale, never one full-page thumbnail, which hides exactly the failures that matter, crude controls, wrong lettering character, flattened material, behind a superficially similar section order. On a Persuade surface, verify the mode did its job: a first-time visitor should know what this is, why it matters, and what to do within seconds, in the form's own vocabulary.
 
-After the second inspection round the build thread's polishing is over: no further defect hunts, micro-edit scripts, or rebuilds here; whatever remains ships through the handoffs, where a fresh context does the finding better and cheaper. On the web, where this harness runs no design hook, run `node .claude/skills/impeccable/scripts/detect.mjs --json` on the changed targets once here, fix what is mechanical, and pass the remaining findings to the reviewer; a hookless web build that skips this ships every tell the hook exists to catch. A native platform skips the detector entirely: it reads HTML and CSS and has no verdict on native code, so the reviewer's floor check is the only slop gate and the input packet says so. Capture the screenshots into `.impeccable/review/`, one file per captured viewport (on the web, `desktop.png` and `mobile.png`; on native, one per device class, such as `phone.png` and `tablet.png`, suffixed per OS on adaptive), creating that directory when the harness does not; the paths you pass the reviewer are its spec, and that directory is where it looks when a passed path is missing. Then spawn the shipped finish reviewer, `impeccable-finish-reviewer` (`impeccable_finish_reviewer` in codex; `/impeccable-finish-reviewer` in Cursor; on GitHub Copilot say "Use the impeccable-finish-reviewer agent"), with the original request, confirmed answers, the artifact path, the screenshot paths, its direction contract, existing hook findings, the QUALITY BAR card and approved comp paths (on a code-led build there is no approved comp; the chosen decision comp rides in that slot as the critique reference, named as such), the craft-floor reference path, and on a native platform the platform reference path(s), [ios.md](ios.md) / [android.md](android.md), both on adaptive, plus one line saying no detector ran, so the reviewer judges in the platform's conventions rather than the web's. The reviewer has no browser; screenshots you fail to pass are checks it cannot run. Never read the shipped agents' definition files before spawning; the harness loads them at spawn, and you owe only the input packet. Wait on any agent with one long timeout rather than a loop of short polls, and spend the wait on the next independent step. Verify its return carries the five contract sections; on an empty or thrashed return, respawn once with the same inputs before doing anything else. This review never runs inside the build thread and never inherits it: spawn the reviewer fresh, with no forked conversation history (`fork_turns: 0` in codex); a reviewer that inherits your transcript inherits your framing, your optimism, and your abstractions, and everything it needs travels in the inputs above. Only a harness whose tool surface has no subagent capability at all substitutes a fresh in-thread pass after stepping fully out of the build context, run from [degraded/finish-reviewer.md](degraded/finish-reviewer.md), and a substituted or failed-and-replaced review is disclosed in one line at finish, never silently. When the reviewer's first material fix is a rebuild directive, fidelity failed wholesale rather than in patches, so skip the fix batch and execute the rebuild immediately: re-derive the named regions, produce the named assets, and send the result back for a verdict, telling the user what is happening rather than asking permission to fix a failure. The user is consulted only when a second rebuild directive arrives, both verdicts on the table, or when rebuilding would discard content the user approved. Otherwise apply the material fixes in one batch, rebuild once, and recapture the same viewports over the same files. A recapture measures positions, loading, and overflow; it cannot measure whether a fix reached the quality the finding named, so send the recaptured screenshots back to the same reviewer for a verdict scoring every material fix resolved, partial, or unresolved (through the harness's agent continuation; without one, run the scoring fresh from [degraded/finish-reviewer.md](degraded/finish-reviewer.md)'s Verdict Pass). Fixes scored partial or unresolved get another batch, recapture, and verdict. Two rounds is the budget an unattended run ends at; an attended session's ceiling belongs to the user, so when the second verdict still lists open items, put the table in front of them and let them choose between shipping as it stands and funding another round. Whoever is deciding, stop the moment a round resolves nothing, and the reviewer's findings are the only list you work from, never your own re-opened hunt. Report the final verdict table to the user as it stands, open items included, under the reviewer's own disposition word: a table with open material findings is never announced as a pass, and never under a softer label than the reviewer wrote. Do not run a second detector.
+A capture is evidence only when it is valid, and you validate before you send. Settle or disable entrance motion first: an element hidden by animation timing reads as a missing element and gets fixed into a regression. Capture full-page shots from the document top. Capture the comp comparison at the comp's own pixel dimensions. Then open every file once and confirm it shows what its name claims: no black or blank regions, no wrong section behind a right filename, no half-loaded state. A malformed capture sent onward costs the whole round; the reviewer answers it with `disposition: recapture` and nothing it reviewed binds.
 
-Then spawn the shipped documenter, `impeccable-documenter` (`impeccable_documenter` in codex), with the project root, the artifact path, the direction contract, PRODUCT.md, the [document.md](document.md) reference path, and the boundary to write at; it records DESIGN.md and the sidecar from the built world, ground truth over intention; without subagents the pass runs from [degraded/documenter.md](degraded/documenter.md). A clean detector pass is not finished; finished is the contract kept, the comp honored, the review closed, and the system recorded.
+After the second inspection round the build thread's polishing is over: no further defect hunts, micro-edit scripts, or rebuilds here; whatever remains ships through the handoffs, where a fresh context does the finding better and cheaper. On the web, where this harness runs no design hook, run `node .claude/skills/impeccable/scripts/detect.mjs --json` on the changed targets once here, fix what is mechanical, and pass the remaining findings to the reviewer; a hookless web build that skips this ships every tell the hook exists to catch. A native platform skips the detector entirely: it reads HTML and CSS and has no verdict on native code, so the reviewer's floor check is the only slop gate and the input packet says so. Capture the screenshots into `.impeccable/review/`, one file per captured viewport (on the web, `desktop.png` and `mobile.png`, plus `user-<width>.png` whenever the user's viewport joined the inspected set; on native, one per device class, such as `phone.png` and `tablet.png`, suffixed per OS on adaptive), creating that directory when the harness does not; the paths you pass the reviewer are its spec, every viewport you inspected is named required in the packet, and that directory is where it looks when a passed path is missing.
+
+Then spawn the shipped finish reviewer, `impeccable-finish-reviewer` (`impeccable_finish_reviewer` in codex; `/impeccable-finish-reviewer` in Cursor; on GitHub Copilot say "Use the impeccable-finish-reviewer agent"), with the original request, confirmed answers, the artifact path, the screenshot paths, the direction contract, existing hook findings, the QUALITY BAR card and approved comp paths (a code-led build has no approved comp; the chosen decision comp rides in that slot as the critique reference, named as such), the craft-floor reference path, and on a native platform the platform reference path(s), [ios.md](ios.md) / [android.md](android.md), both on adaptive, plus one line saying no detector ran, so the reviewer judges in the platform's conventions rather than the web's. The reviewer has no browser; screenshots you fail to pass are checks it cannot run. Never read the shipped agents' definition files before spawning; the harness loads them at spawn, and you owe only the input packet. Wait on any agent with one long timeout rather than a loop of short polls, and spend the wait on the next independent step. Verify the return carries the five contract sections (a recapture return carries one, its recapture list); on an empty or thrashed return, respawn once with the same inputs. This review never runs inside the build thread and never inherits it: spawn the reviewer fresh, with no forked conversation history (`fork_turns: 0` in codex); a reviewer that inherits your transcript inherits your framing, your optimism, and your abstractions, and everything it needs travels in the inputs above. Only a harness with no subagent capability at all substitutes a fresh in-thread pass after stepping fully out of the build context, run from [degraded/finish-reviewer.md](degraded/finish-reviewer.md), and a substituted or failed-and-replaced review is disclosed in one line at finish, never silently.
+
+Act on the disposition word; there are exactly four. **recapture**: the evidence failed, not the build. Recapture what the return names under the capture-validity rules, then run a full review over the new evidence. A review conducted on invalid evidence binds nothing, and a verdict pass may never follow it. **rebuild**: fidelity failed wholesale, not in patches. Skip the fix batch and execute the rebuild immediately: re-derive the named regions, produce the named assets, and send the result back for a fresh full review, never a verdict pass; a rebuild replaces regions wholesale, so the whole matrix runs again over the recaptures. Tell the user what is happening rather than asking permission to fix a failure. Consult the user only on a second rebuild directive, both verdicts on the table, or when rebuilding would discard content the user approved. **ship**: nothing is owed; report the verdict at its scope and continue to the documenter. **fix**: apply the material fixes in one batch, rebuild once, and recapture the same viewports over the same files. A recapture measures positions, loading, and overflow; it cannot measure whether a fix reached the quality the finding named, so send the recaptured screenshots back to the same reviewer for a verdict scoring every material fix resolved, partial, or unresolved (through the harness's agent continuation; without one, run the scoring fresh from [degraded/finish-reviewer.md](degraded/finish-reviewer.md)'s Verdict Pass). Fixes scored partial or unresolved get another batch, recapture, and verdict. Two rounds is the budget an unattended run ends at; an attended session's ceiling belongs to the user, so when the second verdict still lists open items, put the table in front of them and let them choose between shipping as it stands and funding another round. Whoever decides, stop the moment a round resolves nothing, and the reviewer's findings are the only list you work from, never your own re-opened hunt. Do not run a second detector.
+
+A rebuild and a fix round share one asset rule: a raster either round creates or replaces is still asset work under [visualize.md](visualize.md)'s Produce section and keeps its **provenance** like every build raster, and a raster the round abandons is deleted in the same batch. Before either round's result goes back for review or verdict, run `node .claude/skills/impeccable/scripts/embed-prompt.mjs --scan <asset-dir...>` over the directories the artifact's rasters ship from and clear every file it reports by embedding what it is missing: the exact generation prompt for a produced raster, the origin for a sourced, stock, or pre-existing one. The scan only reads; deletion is reserved for rasters the round abandoned, never for a file the scan flagged.
+
+Report the final verdict under the reviewer's own disposition word and at its actual scope. A verdict pass scores the listed fixes and nothing else: "the reviewer scored all three fixes resolved" is a claim it supports, "no material issues remain" is not. A table with open material findings is never announced as a pass, never softened, and never dressed as whole-surface approval when only a fix list was scored. When the user answers a ship with evidence against it, their own screenshot, a named mismatch with the comp, that evidence outranks every capture you made: put their material in the packet and spawn a fresh reviewer for a new full review. Patching inline and self-certifying is how a rejected page ships twice.
+
+Then spawn the shipped documenter, `impeccable-documenter` (`impeccable_documenter` in codex), with the project root, the artifact path, the direction contract, PRODUCT.md, the [document.md](document.md) reference path, and the boundary to write at; it records DESIGN.md and the sidecar from the built world, ground truth over intention; without subagents the pass runs from [degraded/documenter.md](degraded/documenter.md). The documenter runs after the last correction lands: when any fix round follows the documentation, re-run the documenter over the changed surface, because a DESIGN.md describing a layout that no longer exists turns defects into system guidance. A clean detector pass is not finished; finished is the contract kept, the comp honored, the review closed, and the system recorded.

.claude/skills/impeccable/reference/visualize.md

@@ -1,51 +1,55 @@
 # Visualize: Direction Comps & Asset Production
 
-Load this from [new-work.md](new-work.md) on a comp-led build, when image generation is available (a harness-native tool or the API fallback context.mjs reports). A code-led execution contract skips this file by design, not by drift: its ambition lives in the written direction contract and is audited in behavior, so do not load it for a code-led round. PRODUCT.md and DESIGN.md are preconditions. New-work has already resolved the visual world; this file must not reopen it. A surface-scope structure round that already put three visualized cards before the user (new-work.md, established world) has discharged this round: the locked card’s comp is the approved comp, so record the approval and continue at After approval; generate nothing new.
+Load this from [new-work.md](new-work.md) on a comp-led build, when image generation is available (a harness-native tool or the API fallback context.mjs reports). A code-led contract skips this file by design, not by drift; do not load it then. PRODUCT.md and DESIGN.md are preconditions. New-work has already resolved the visual world; this file must not reopen it. A surface-scope structure round that already put three visualized cards before the user (new-work.md, established world) has discharged this round: the locked card's comp is the approved comp, so record the approval and continue at After approval; generate nothing new.
 
-The purpose of a probe is to test composition, narrative, hierarchy, density, focal moment, signature use, and image requirements. It is not a second identity workshop. Keep DESIGN.md's palette, typography direction, material language, component character, imagery stance, and motion grammar fixed.
+A probe tests composition, narrative, hierarchy, density, focal moment, signature use, and image requirements. It is not a second identity workshop. Keep DESIGN.md's palette, typography direction, material language, component character, imagery stance, and motion grammar fixed.
 
 ## Generate three compositional options
 
-Render three distinct high-fidelity north-star comps of the requested surface, with whatever generation capability exists, saved under `.impeccable/mocks/` so they survive the session. Comp at the surface's own viewport: portrait at device size for a native app or mobile-first surface, desktop landscape otherwise; a phone screen comped landscape misstates the composition before anything gets built against it. Comps are the build thread's own work, never delegated: the thread that writes the comp prompts holds the direction's full context, and it has already seen every comp when the build starts. Open every image you produce or reference by its workspace-relative path, never an absolute one: sandboxed viewers reject absolute paths, and everything under the project root has a relative path. Base them on the real content and the surface concepts already developed with the user. On an established world, anchor every comp on the real identity: capture a screenshot of a representative existing page and pass it as a reference image (the harness image tool’s input image, or `generate-image.mjs --ref`); the prompt then leads with the new surface’s structure while the reference carries palette, type, and component character, because DESIGN.md words alone drift where a pixel reference does not. Name what the reference contributes and what it must not: chrome, palette, type, and component character carry over; the reference page’s own content does not, so a banner, hero, or card lifted verbatim from the reference is the reference leaking, not fidelity. Three is the number: one comp invites rubber-stamping, and the spread between three is what surfaces the composition worth building. The chosen card's decision comp is the first of the three: it already renders this direction at full fidelity under this file's discipline, so this round generates two more that vary what the first held fixed, and all three go to the approval point together. Only a round that arrives with no decision comp, a degraded roll, an identity-mode page, a direction pinned without the decision round, renders all three here.
+Render three distinct high-fidelity north-star comps of the requested surface, saved under `.impeccable/mocks/` so they survive the session. Comp at the surface's own viewport: portrait at device size for a native app or mobile-first surface, desktop landscape otherwise; a phone screen comped landscape misstates the composition before anything is built against it. Comps are the build thread's own work, never delegated: the thread that writes the prompts holds the direction's full context and has seen every comp when the build starts. Open every image by its workspace-relative path; sandboxed viewers reject absolute paths, and everything under the project root has a relative one. Base the comps on real content and the surface concepts already developed with the user. On an established world, anchor every comp on the real identity: capture a screenshot of a representative existing page and pass it as a reference image (the harness image tool's input image, or `generate-image.mjs --ref`); the prompt leads with the new surface's structure while the reference carries palette, type, and component character, because DESIGN.md words alone drift where a pixel reference does not. Name what the reference contributes and what it must not: chrome, palette, type, and component character carry over; the reference page's own content does not, and a banner, hero, or card lifted verbatim is the reference leaking, not fidelity. Three is the number: one comp invites rubber-stamping; the spread between three surfaces the composition worth building. The chosen card's decision comp is the first of the three: it already renders this direction at full fidelity under this discipline, so generate two more that vary what the first held fixed, and send all three to the approval point together. Only a round arriving with no decision comp (a degraded roll, an identity-mode page, a direction pinned without the decision round) renders all three here.
 
-- A comp is a designed surface, not a picture of the subject. Lead the generation prompt with the surface's own structure, whatever regions this design actually has, named in order with their scale relationships; a page with no navigation states that instead of inventing one, and an unconventional surface states its unconventional skeleton. A prompt that leads with the world's atmosphere gets a vignette back: the model paints the fish market instead of the fish market's website. Self-check every render: if it could hang as a poster, or reads as a photograph or scene with some text on it, it is not a comp; regenerate with the layout scaffold stated more literally.
-- The inverse is also a failure: a surface with none of its subject in it. The subject appears as the content the regions exist to hold; the world dresses the frame and never displaces what the frame exists to show. The deletion usually rides in on the prompt's exclusion list, so exclusions bind invented claims, and a medium ban belongs to the committed imagery stance, never to caution. Before accepting a render, point at the subject: a render that depicts everything about the world and nothing of the subject fails however faithful its atmosphere, so regenerate with the subject's content named region by region.
-- A comp is judged as the shipped screen: the visitor's job must be readable from the image alone. Name the surface's mode from the render with no caption; a render whose mode cannot be read back is art direction without a surface, so regenerate with the visitor's job as the prompt's spine.
-- Commitment is depth, not coverage. The world enters through one dominant move plus the material, type, and spacing that support it, and the remaining regions hold still so that move can be read; a region that simply does its job in the world's own grammar carries the direction further than a region performing the concept. The check cuts competition, never content: a quieted region keeps its information and stops performing. Where the direction names a focal moment, a second element competing with it at the same scale means the comp is shouting; where it names none, several regions performing the concept at once is the same shout. Regenerate keeping the strongest move and quieting the rest. Busy is louder, not bolder.
+- A comp is a designed surface, not a picture of the subject. Lead the prompt with the surface's own structure: the regions this design has, named in order with their scale relationships; a page with no navigation says so instead of inventing one, and an unconventional surface states its unconventional skeleton. A prompt that leads with atmosphere gets a vignette back: the model paints the fish market instead of the fish market's website. Self-check every render: if it could hang as a poster, or reads as a photograph with some text on it, it is not a comp; regenerate with the layout scaffold stated more literally.
+- The inverse is also a failure: a surface with none of its subject in it. The subject appears as the content the regions hold; the world dresses the frame and never displaces what the frame shows. The deletion usually rides in on the prompt's exclusion list, so exclusions bind invented claims, and a medium ban belongs to the committed imagery stance, never to caution. Before accepting a render, point at the subject; a render that depicts everything about the world and nothing of the subject fails however faithful its atmosphere. Regenerate with the subject's content named region by region.
+- Judge a comp as the shipped screen: the visitor's job must be readable from the image alone. Name the surface's mode from the render with no caption; a render whose mode cannot be read back is art direction without a surface. Regenerate with the visitor's job as the prompt's spine.
+- Commitment is depth, not coverage. The world enters through one dominant move plus the material, type, and spacing that support it; the remaining regions hold still so that move can be read. A region that simply does its job in the world's grammar carries the direction further than a region performing the concept. The check cuts competition, never content: a quieted region keeps its information and stops performing. A second element competing with the named focal moment at the same scale means the comp is shouting; with no named focal moment, several regions performing the concept at once is the same shout. Regenerate keeping the strongest move and quieting the rest. Busy is louder, not bolder.
 - When the user shortlisted multiple concepts, spread the three across them.
 - When one direction is committed, vary the structural uncertainty an image can resolve: topology, sequence, density, hierarchy, focal composition, or interaction framing.
-- Show enough beyond the opening moment to prove the concept can govern the whole requested surface.
+- Show enough beyond the opening moment to prove the concept can govern the whole surface.
 - Do not generate a palette artifact, ask new atmosphere questions, introduce a different type voice, or invent a new motif. If the committed world cannot support the concept, return to the concept shortlist rather than changing the world.
 
-Treat each comp as a direction test, not a screenshot specification. Core UI text, responsive behavior, accessibility, semantics, and interaction states remain implementation responsibilities.
+Each comp is a direction test, not a screenshot specification. Core UI text, responsive behavior, accessibility, semantics, and interaction states remain implementation responsibilities.
 
 ## One approval point
 
-Show the three together on the decision page (`serve-question.mjs`, one option per comp with the comp as its hero), or in the harness only when it renders images inline; a text-only surface does not count as display. Ask what should carry forward, what feels false to the world, and whether the selected surface concept should be approved, combined, revised, or rejected. Then stop and wait. A structured simulated user counts as attended and receives the same question.
+Show the three together on the decision page (`serve-question.mjs`, one option per comp with the comp as its hero), or in the harness only when it renders images inline; a text-only surface does not count as display. Ask what should carry forward, what feels false to the world, and whether the selected concept should be approved, combined, revised, or rejected. Then stop and wait. A structured simulated user counts as attended and receives the same question.
 
 Do not begin code until the user approves a direction or explicitly delegates the choice. If they delegate, choose using the task brief, PRODUCT.md, and DESIGN.md, and state the evidence. Approval refines the task concept; it does not modify DESIGN.md.
 
-This approval point has no substitute and no skip condition. When the structured question tool errors, fall back to the decision page; only after both fail may you treat the choice as delegated, and a delegated pick is still recorded exactly as an approval is and disclosed in your first reply, not your last. The finish reviewer treats a build whose comp round produced comps with no recorded approval as carrying a material finding; decision comps under `.impeccable/mocks/decision/` are the direction round's hand, not comp-round output, and imply no approval on their own.
+This approval point has no substitute and no skip condition. When the structured question tool errors, fall back to the decision page; only after both fail may you treat the choice as delegated, and a delegated pick is recorded exactly as an approval is and disclosed in your first reply, not your last. The finish reviewer treats comp-round comps with no recorded approval as a material finding; decision comps under `.impeccable/mocks/decision/` are the direction round's hand, not comp-round output, and imply no approval on their own.
 
-After approval, record the choice where tools can find it: the approved comp's path goes in the surface brief, and the approved comp's `.json` prompt sidecar gains `"approved": true` (every comp generated through `generate-image.mjs` has one; create it if a native tool didn't). The sidecar travels with the mocks folder, so the approval survives sessions and machines that never see the brief. Then summarize the composition and the parts of the comp that must not be literalized, return to new-work.md, record the direction contract from the approved surface concept, and build.
+After approval, record the choice where tools can find it: the approved comp's path goes in the surface brief, and its `.json` prompt sidecar gains `"approved": true` (every comp generated through `generate-image.mjs` has one; create it if a native tool didn't). The sidecar travels with the mocks folder, so the approval survives sessions and machines that never see the brief. Summarize the composition and the parts of the comp that must not be literalized, return to new-work.md, record the direction contract from the approved concept, and build.
 
 ## Inventory implementation fidelity
 
-Before building, read the approved comp as a design system and record it in the brief: component grammar, corner language, line weights, elevation treatment, and the type ramp, because everything the comp does not show gets built from this record, and without it the fallback is the model's stock kit of square boxes, 1px grids, bento cells, and hard shadows. Then inventory the comp's major visible ingredients in writing (a short table in the surface brief or working notes; the finish reviewer audits shipped assets against it) and choose an implementation medium for each: semantic HTML/CSS/SVG, existing project asset, generated raster, sourced raster, icon library, canvas/WebGL, or accepted omission. The same written inventory names the comp's compositional commitments: navigation items and icons, headline levels and their scale relationship, signature geometry such as seams, masks, and overlaps, and each section's arrangement and density. The primary action gets its own row with its own medium: when the comp dissolves, stamps, erodes, or otherwise physically works the main CTA, that treatment is signature material on the page's most important element, and shrinking it to a border trick or a few decorative pixels is the compliance-token version of commitment. An element never written down is the element the build silently drops, and the direction contract's 150 words cannot carry this list, so this inventory is where it lives.
+Before building, read the approved comp as a design system and record it in the brief: component grammar, corner language, line weights, elevation treatment, and the type ramp. Everything the comp does not show gets built from this record; without it the fallback is the model's stock kit of square boxes, 1px grids, bento cells, and hard shadows. Then inventory the comp's major visible ingredients in writing (a short table in the surface brief or working notes; the finish reviewer audits shipped assets against it) and choose an implementation medium for each: semantic HTML/CSS/SVG, existing project asset, generated raster, sourced raster, icon library, canvas/WebGL, or accepted omission. The same inventory names the comp's compositional commitments: navigation items and icons, headline levels and their scale relationship, signature geometry such as seams, masks, and overlaps, and each section's arrangement and density. The primary action gets its own row with its own medium: when the comp dissolves, stamps, erodes, or otherwise physically works the main CTA, that treatment is signature material on the page's most important element, and shrinking it to a border trick is the compliance-token version of commitment. An element never written down is the element the build silently drops; the direction contract's 150 words cannot carry this list, so it lives here.
 
-The medium column is where an approved design most often dies, so it obeys a gate: the medium is decided by what the comp region shows, never by what feels buildable in the current stack. A human figure, a product object, machinery, or any material with lighting and depth is raster whatever the stack, and so is any texture by that name alone: woven cloth, paper grain, fabric, leather, brushed metal need no depth argument, because a CSS gradient or layered background is not a texture medium and "layered CSS textures" is not a medium at all. Writing "silhouette" for a photographic figure, or "CSS" for a sculpted panel's finish or a cotton field's weave, is not a medium choice, it is the quiet deletion of the approved design, and it is how a comp full of physical material becomes a flat page with the same section order. Style does not move this boundary: a comp region with perspective, shading, figure drawing, or dense mechanical detail is illustration however line-drawn it looks, and no build session can author illustration as vectors, so it regenerates as raster like any photograph. Authored SVG covers what a session can specify exactly, diagrams with countable elements, controls, flat shape systems, and it ends where drawing skill begins; an instruction-manual world does not convert its illustrations into diagrams, it makes them line-art illustrations. Produce such regions by regenerating them cleanly, with the approved comp and its embedded prompt as the reference for a fresh render at asset resolution; never crop pixels out of the comp itself, whose effective resolution sits far below asset grade. Dropping an image-native region instead of producing it is a scope decision the user makes at the approval point, never a silent flattening after it. Generated imagery is a material, not a claim: evidence rules bind assertions, specs, testimonials, and photographs presented as real, never render fidelity, so "no photography on hand" forbids fake proof, not an illustrated hero.
+The record is sampled, never estimated: read the comp's page **ground**, each dominant field, and each accent's actual hex from its pixels (ImageMagick, Python with PIL, any pixel-reading tool on the machine) and write the values into the same record. Take a flat field from any interior pixel, a textured or grainy one as the average of an interior patch (crop a swatch, scale it to one pixel), and a gradient as its two end colors; never sample an edge, where antialiasing blends neighbors into colors the design never chose. An adjective is a direction, not a record: cream covers everything from near-white to beige, charcoal a third of the value scale, and wherever no number pins a color, the rendition prior picks the spot. Sampled values supersede the palette chips on the decision and composition cards: those were authored before this comp existed, and a chip that disagrees with the comp's pixels is a draft the approval retired.
 
-The gate runs both ways: precise geometry, hard-edged shape systems, diagrams, expressive motion, shaders, and anything interactive are vector and GPU territory (SVG, canvas, WebGL), where a raster flattens what should move, scale, and respond, and code executed safely and professionally remains first-class there. A field or texture built from many small elements carries a quantity commitment either way: write down its approximate density and coverage ("thousands of glyphs over two-thirds of the fold, dense at the top fading into the path"), because a field rebuilt at a tenth of its density passes every checklist and still is not the design. TYPE rows carry the same discipline: name the face's compression class, and render one headline word against the comp before building on it; a visibly wider or lighter silhouette means the face is wrong, and every section built on it inherits the miss. Raster is for what the world paints; code is for what the world draws, animates, or reacts with, and choosing code there is ambition, not economy. Every `produce` entry is produced before the build ships, through the asset producer or in the current thread; an inventory with unproduced entries is an unfinished build, and this gate is where imagery-free pages come from when it is skipped.
+The medium column is where an approved design most often dies, so it obeys a gate: the medium is decided by what the comp region shows, never by what feels buildable in the current stack. A human figure, a product object, machinery, or any material with lighting and depth is raster whatever the stack; so is any texture by name alone: woven cloth, paper grain, fabric, leather, brushed metal need no depth argument, because a CSS gradient is not a texture medium and "layered CSS textures" is not a medium at all. Writing "silhouette" for a photographic figure, or "CSS" for a sculpted panel's finish, is not a medium choice; it is the quiet deletion of the approved design, and it is how a comp full of physical material becomes a flat page with the same section order. Style does not move this boundary: a comp region with perspective, shading, figure drawing, or dense mechanical detail is illustration however line-drawn it looks, and no build session can author illustration as vectors, so it regenerates as raster like any photograph. Authored SVG covers what a session can specify exactly (diagrams with countable elements, controls, flat shape systems) and ends where drawing skill begins; an instruction-manual world keeps its illustrations as line-art illustrations, not diagrams. Produce such regions by regenerating them cleanly, with the approved comp and its embedded prompt as the reference for a fresh render at asset resolution; never crop pixels out of the comp itself, whose effective resolution sits far below asset grade. Dropping an image-native region is a scope decision the user makes at the approval point, never a silent flattening after it. Generated imagery is a material, not a claim: evidence rules bind assertions, specs, testimonials, and photographs presented as real, never render fidelity; "no photography on hand" forbids fake proof, not an illustrated hero.
+
+The gate runs both ways: precise geometry, hard-edged shape systems, diagrams, expressive motion, shaders, and anything interactive are vector and GPU territory (SVG, canvas, WebGL), where a raster flattens what should move, scale, and respond. A field or texture built from many small elements carries a quantity commitment either way: write down its approximate density and coverage ("thousands of glyphs over two-thirds of the fold, dense at the top fading into the path"), because a field rebuilt at a tenth of its density passes every checklist and still is not the design. TYPE rows carry the same discipline: name the face's compression class, and render one headline word against the comp before building on it; a visibly wider or lighter silhouette means the face is wrong, and every section built on it inherits the miss. Raster is for what the world paints; code is for what the world draws, animates, or reacts with, and choosing code there is ambition, not economy. Every `produce` entry is produced before the build ships, through the asset producer or in the current thread; an inventory with unproduced entries is an unfinished build, and this gate is where imagery-free pages come from when it is skipped.
 
 Pay special attention to the dominant composition, signature use, image-native content, second-fold system, and any interaction the still image only implies.
 
-Treat the comp as a north star, not something to trace, and know what that allows: translation into semantic, responsive, accessible code, never recomposition. Keeping the palette and mood while redrawing the topology is a second art direction, not an adaptation. Do not rasterize core UI text or controls. Do not substitute a different visual driver after approval without asking.
+The comp is a north star, not something to trace, and know what that allows: translation into semantic, responsive, accessible code, never recomposition. Keeping the palette and mood while redrawing the topology is a second art direction, not an adaptation. Do not rasterize core UI text or controls. Do not substitute a different visual driver after approval without asking.
 
 ## Produce only the assets the build needs
 
-Generation context is part of the asset: a build composed by a thread that never saw the prompts places assets it does not understand. So prefer generating build-critical imagery in the build thread when the budget allows, and when a subagent produces assets instead, every asset must carry its prompt, and the builder reads those prompts before composing a single one of them. The carrier is uniform across harnesses: after generating any image with any tool, native or `generate-image.mjs` (which does it automatically), run `node .claude/skills/impeccable/scripts/embed-prompt.mjs <image> --prompt "<the prompt used>"` so the intent lives inside the file itself and survives copies between machines and harnesses; `--read` recovers it from any impeccable-generated image.
+Generation context is part of the asset: a build composed by a thread that never saw the prompts places assets it does not understand. Prefer generating build-critical imagery in the build thread when the budget allows; when a subagent produces assets instead, every asset carries its prompt, and the builder reads those prompts before composing. The carrier is uniform across harnesses: after generating any image with any tool, native or `generate-image.mjs` (which does it automatically), run `node .claude/skills/impeccable/scripts/embed-prompt.mjs <image> --prompt "<prompt>"` with the exact string the generation tool received, pasted whole, so the intent lives inside the file and survives copies between machines and harnesses; a summary reconstructed from memory records an asset that was never made. `--read` recovers the prompt from any impeccable-generated image, and `--scan <dir>` lists every raster in a directory still missing one. The embedded prompt plus the asset's row in the written inventory is the raster's **provenance**, and every raster the artifact references carries it; a sourced, stock, or pre-existing raster with no generation prompt embeds its origin instead.
 
-When the harness runs subagents, spawn the shipped asset producer every time, even when the inventory's produce bucket looks empty: its manifest is the independent second opinion on your media, and runs that skipped the spawn are the runs whose cotton became CSS. An honestly empty manifest costs one cheap spawn; a wrongly empty produce bucket costs the build its materials. Use the producer, `impeccable-asset-producer` (`impeccable_asset_producer` in codex; `/impeccable-asset-producer` in Cursor; on GitHub Copilot say "Use the impeccable-asset-producer agent"): give it the approved comp, output paths, required dimensions and formats, transparency needs, crop notes, and what must remain semantic code. Otherwise produce the minimum required assets in the current thread by the book: load [degraded/asset-producer.md](degraded/asset-producer.md) and follow it inline, with whatever generation exists, the native tool or generate-image.mjs.
+Provenance is owed for the run, not the build phase: a raster created or replaced later, in a fix batch or a reviewer's rebuild, is produced under this same section, prompt embedded and inventory row added, because the inventory is how the next thread knows what ships. A raster a fix abandons or supersedes is deleted from the assets directory in the same batch; an unreferenced raster with no record is a provenance leak, not a spare.
+
+When the harness runs subagents, spawn the shipped asset producer every time, even when the inventory's produce bucket looks empty: its manifest is the independent second opinion on your media, and the runs that skipped the spawn are the runs whose cotton became CSS. An honestly empty manifest costs one cheap spawn; a wrongly empty produce bucket costs the build its materials. Use `impeccable-asset-producer` (`impeccable_asset_producer` in codex; `/impeccable-asset-producer` in Cursor; on GitHub Copilot say "Use the impeccable-asset-producer agent"): give it the approved comp, output paths, required dimensions and formats, transparency needs, crop notes, and what must remain semantic code. Without subagents, produce the minimum required assets in the current thread by the book: load [degraded/asset-producer.md](degraded/asset-producer.md) and follow it inline, with whatever generation exists.
 
 Convert images with a converter context.mjs reported at boot (the IMAGE_TOOLS line); probe only when it reported none, at most once per session, never per image.
 

.claude/skills/impeccable/scripts/context.mjs

@@ -1013,6 +1013,27 @@ async function fetchLatestSkillVersion() {
   }
 }
 
+// Destroy fetch's global undici dispatcher before process.exit(): a live
+// keep-alive socket trips a libuv assertion on Windows/Node 24 after a
+// successful boot (nodejs/node#56645, issue #573).
+async function destroyFetchDispatcher() {
+  const dispatcher = globalThis[Symbol.for('undici.globalDispatcher.1')];
+  if (dispatcher && typeof dispatcher.destroy === 'function') {
+    try { await dispatcher.destroy(); } catch { /* exit regardless */ }
+  }
+}
+
+// Drain the boot payload before process.exit(): a live pipe that has not
+// flushed yet is truncated when Node tears down (issue #573 review). Then
+// close fetch so Windows teardown does not abort on the keep-alive socket.
+async function finishCli(output) {
+  await new Promise((resolve) => {
+    process.stdout.write(output, () => resolve());
+  });
+  await destroyFetchDispatcher();
+  process.exit(0);
+}
+
 // Two instructions used to sit in one directive: ask, and "if they agree, run
 // it". Nothing gated the second on an answer, and the same sentence said to
 // continue without waiting, so a run that could never establish agreement was
@@ -1159,8 +1180,7 @@ async function cli() {
     appendImageToolsDirective(parts);
     appendStalenessDirective(parts, ctx, cliOptions);
     if (updateDirective) parts.push(updateDirective);
-    process.stdout.write(parts.join('\n\n---\n\n') + '\n');
-    process.exit(0);
+    await finishCli(parts.join('\n\n---\n\n') + '\n');
   }
   const parts = [`# PRODUCT.md\n\n${ctx.product.trim()}`];
   if (ctx.hasDesign) {
@@ -1206,7 +1226,7 @@ async function cli() {
     }
   }
   if (updateDirective) parts.push(updateDirective);
-  process.stdout.write(parts.join('\n\n---\n\n') + '\n');
+  await finishCli(parts.join('\n\n---\n\n') + '\n');
 }
 
 function parseCliOptions(args) {

.claude/skills/impeccable/scripts/detector/browser/injected/index.mjs

@@ -626,7 +626,7 @@ if (IS_BROWSER) {
       if (currentStyle.filter && currentStyle.filter !== 'none') reasons.add('filter');
       if (currentStyle.backdropFilter && currentStyle.backdropFilter !== 'none') reasons.add('backdrop filter');
 
-      const solidBg = parseRgb(currentStyle.backgroundColor);
+      const solidBg = parseRgb(currentStyle.backgroundColor) || parseAnyColor(currentStyle.backgroundColor);
       if (solidBg && solidBg.a >= 0.95 && (!bgImage || bgImage === 'none')) break;
       current = current.parentElement;
     }
@@ -688,7 +688,7 @@ if (IS_BROWSER) {
       // starve the url()-backed texts this mode exists to sample.
       if (options.imageOnly && !reasons.includes('image background')) continue;
 
-      const textColor = parseRgb(style.color);
+      const textColor = parseRgb(style.color) || parseAnyColor(style.color);
       const fontSize = parseFloat(style.fontSize) || 16;
       const fontWeight = parseInt(style.fontWeight) || 400;
       const isLargeText = fontSize >= WCAG_LARGE_TEXT_PX || (fontSize >= WCAG_LARGE_BOLD_TEXT_PX && fontWeight >= 700);
@@ -985,7 +985,7 @@ if (IS_BROWSER) {
         return sample;
       }
     }
-    const bg = parseRgb(style.backgroundColor);
+    const bg = parseRgb(style.backgroundColor) || parseAnyColor(style.backgroundColor);
     if (bg && bg.a > 0.05) return { status: 'sampled', color: bg, method: 'solid-background' };
     return { status: 'unresolved', reason: 'no readable background' };
   }
@@ -1115,7 +1115,7 @@ if (IS_BROWSER) {
     }
 
     const style = getComputedStyle(el);
-    const textColor = parseRgb(style.color) || candidate.textColor;
+    const textColor = parseRgb(style.color) || parseAnyColor(style.color) || candidate.textColor;
     if (!textColor) return { ...candidate, status: 'unresolved', confidence: 'none', reason: 'unreadable text color' };
 
     const rect = getDirectTextRect(el) || el.getBoundingClientRect();

.claude/skills/impeccable/scripts/detector/cli/main.mjs

@@ -105,6 +105,13 @@ function formatFindings(findings, jsonMode) {
 // `optionsFor` maps a local path to scan options carrying that path's own
 // project design system (or base options when null). Falls back to a plain
 // object so direct/legacy callers still work.
+async function detectLocalFile(filePath, options) {
+  if (HTML_EXTENSIONS.has(path.extname(filePath).toLowerCase())) {
+    return detectHtml(filePath, options);
+  }
+  return detectText(fs.readFileSync(filePath, 'utf-8'), filePath, options);
+}
+
 async function handleStdin(optionsFor = () => ({})) {
   const resolve = typeof optionsFor === 'function' ? optionsFor : () => optionsFor;
   const chunks = [];
@@ -114,9 +121,7 @@ async function handleStdin(optionsFor = () => ({})) {
     const parsed = JSON.parse(input);
     const fp = parsed?.tool_input?.file_path;
     if (fp && fs.existsSync(fp)) {
-      const options = resolve(fp);
-      return HTML_EXTENSIONS.has(path.extname(fp).toLowerCase())
-        ? detectHtml(fp, options) : detectText(fs.readFileSync(fp, 'utf-8'), fp, options);
+      return detectLocalFile(fp, resolve(fp));
     }
   } catch { /* not JSON */ }
   return detectText(input, '<stdin>', resolve(null));
@@ -374,16 +379,10 @@ async function detectCli() {
           }
 
           for (const file of files) {
-            const ext = path.extname(file).toLowerCase();
             // Each file resolves its own project design system (cached by root),
             // so a scan spanning sibling projects applies the right rules per file.
             const fileOptions = scanOptionsFor(file);
-            let fileFindings;
-            if (HTML_EXTENSIONS.has(ext)) {
-              fileFindings = await detectHtml(file, fileOptions);
-            } else {
-              fileFindings = detectText(fs.readFileSync(file, 'utf-8'), file, fileOptions);
-            }
+            const fileFindings = await detectLocalFile(file, fileOptions);
             // Annotate findings with import context
             const importers = importedByMap.get(file);
             if (importers && importers.size > 0) {
@@ -396,13 +395,8 @@ async function detectCli() {
           }
         } else if (stat.isFile()) {
           if (shouldIgnoreDetectionFile(resolved, process.cwd(), detectionConfig)) continue;
-          const ext = path.extname(resolved).toLowerCase();
           const fileOptions = scanOptionsFor(resolved);
-          if (HTML_EXTENSIONS.has(ext)) {
-            allFindings.push(...await detectHtml(resolved, fileOptions));
-          } else {
-            allFindings.push(...detectText(fs.readFileSync(resolved, 'utf-8'), resolved, fileOptions));
-          }
+          allFindings.push(...await detectLocalFile(resolved, fileOptions));
         }
       }
     } finally {

.claude/skills/impeccable/scripts/detector/design-system.mjs

@@ -13,6 +13,11 @@ const FALLBACK_DIRS = ['.agents/context', 'docs'];
 // CLI can't import (separate tree). `.git` and `package.json` are the common
 // boundaries; `.impeccable` is our own project marker.
 const PROJECT_ROOT_MARKERS = ['.git', 'package.json', '.impeccable'];
+// Monorepo-root recognition, mirroring context.mjs's isMonorepoRoot: declared
+// workspace globs (package.json `workspaces`, pnpm-workspace.yaml `packages:`)
+// or a marker file beside apps/ or packages/ children.
+const MONOREPO_MARKER_FILES = ['pnpm-workspace.yaml', 'turbo.json', 'nx.json', 'lerna.json'];
+const MONOREPO_FALLBACK_PROJECT_DIRS = ['apps', 'packages'];
 const COLOR_CHANNEL_TOLERANCE = 6;
 // Shadow blacks at different alphas are different tokens (0.28 vs 0.55 is the
 // difference between a documented shadow and drift), so shadow matching cannot
@@ -575,14 +580,179 @@ function designSystemStartDir(targetPath, cwd = process.cwd()) {
   }
 }
 
+// Same two groups as context.mjs's readProjectPatternGroups: Impeccable
+// projectRoots govern any path they match (positive or negated); package-manager
+// globs only apply to paths the Impeccable group does not match.
+function readWorkspacePatternGroups(dir) {
+  const impeccable = [];
+  for (const name of ['config.json', 'config.local.json']) {
+    const roots = safeReadJson(path.join(dir, '.impeccable', name))?.projectRoots;
+    if (Array.isArray(roots)) {
+      impeccable.push(...roots.filter(entry => typeof entry === 'string' && entry.trim()).map(entry => entry.trim()));
+    }
+  }
+  const pkg = [];
+  const workspaces = safeReadJson(path.join(dir, 'package.json'))?.workspaces;
+  if (Array.isArray(workspaces)) pkg.push(...workspaces);
+  else if (Array.isArray(workspaces?.packages)) pkg.push(...workspaces.packages);
+  const lernaPackages = safeReadJson(path.join(dir, 'lerna.json'))?.packages;
+  if (Array.isArray(lernaPackages)) pkg.push(...lernaPackages);
+  try {
+    let inPackages = false;
+    for (const line of fs.readFileSync(path.join(dir, 'pnpm-workspace.yaml'), 'utf-8').split(/\r?\n/)) {
+      const trimmed = stripInlineYamlComment(line).trim();
+      if (!trimmed || trimmed.startsWith('#')) continue;
+      const flow = trimmed.match(/^packages:\s*\[(.*)\]\s*$/);
+      if (flow) {
+        pkg.push(...flow[1].split(',').map(entry => entry.trim().replace(/^['"]|['"]$/g, '')).filter(Boolean));
+        break;
+      }
+      if (/^packages:\s*$/.test(trimmed)) { inPackages = true; continue; }
+      if (!inPackages) continue;
+      const item = trimmed.match(/^-\s*(.+)$/);
+      if (item) pkg.push(item[1].trim().replace(/^['"]|['"]$/g, ''));
+      else if (/^[A-Za-z0-9_-]+:\s*/.test(trimmed)) break;
+    }
+  } catch { /* no pnpm-workspace.yaml */ }
+  return [impeccable, pkg];
+}
+
+function readWorkspacePatterns(dir) {
+  return readWorkspacePatternGroups(dir).flat();
+}
+
+function isMonorepoRoot(dir) {
+  if (readWorkspacePatterns(dir).some(pattern => !String(pattern).trim().startsWith('!'))) return true;
+  if (!MONOREPO_MARKER_FILES.some(file => fs.existsSync(path.join(dir, file)))) return false;
+  return MONOREPO_FALLBACK_PROJECT_DIRS.some(name => {
+    try {
+      return fs.readdirSync(path.join(dir, name), { withFileTypes: true }).some(entry => entry.isDirectory());
+    } catch {
+      return false;
+    }
+  });
+}
+
+function monorepoOwnsPath(root, boundaryDir) {
+  const rel = path.relative(root, boundaryDir);
+  if (!rel || rel.startsWith('..') || path.isAbsolute(rel)) return false;
+  const relSegments = rel.split(path.sep).filter(Boolean);
+
+  function normalizeWorkspacePattern(pattern) {
+    return String(pattern || '')
+      .trim()
+      .replace(/^['"]|['"]$/g, '')
+      .replace(/^\.\//, '')
+      .replace(/\/+$/, '');
+  }
+
+  function escapeRegExp(s) {
+    return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+  }
+
+  function segmentMatches(patternSegment, relSegment) {
+    if (patternSegment === '*') return true;
+    if (!patternSegment.includes('*')) return patternSegment === relSegment;
+    const re = new RegExp(`^${escapeRegExp(patternSegment).replace(/\\\*/g, '[^/]*')}$`);
+    return re.test(relSegment);
+  }
+
+  function matchGlobSegments(patternSegments, relSegments) {
+    function rec(pi, ri) {
+      if (pi === patternSegments.length) return ri === relSegments.length;
+      if (patternSegments[pi] === '**') {
+        if (pi === patternSegments.length - 1) return true;
+        for (let k = ri; k <= relSegments.length; k++) {
+          if (rec(pi + 1, k)) return true;
+        }
+        return false;
+      }
+      if (ri >= relSegments.length) return false;
+      if (!segmentMatches(patternSegments[pi], relSegments[ri])) return false;
+      return rec(pi + 1, ri + 1);
+    }
+    return rec(0, 0);
+  }
+
+  // Negations like !packages/excluded must also cover nested dirs under that path.
+  function matchesNegation(pattern) {
+    const patternSegments = normalizeWorkspacePattern(pattern).split('/').filter(Boolean);
+    if (!patternSegments.length) return false;
+    if (patternSegments.includes('**')) return matchGlobSegments(patternSegments, relSegments);
+    if (relSegments.length < patternSegments.length) return false;
+    for (let i = 0; i < patternSegments.length; i++) {
+      if (!segmentMatches(patternSegments[i], relSegments[i])) return false;
+    }
+    return true;
+  }
+
+  // Positive globs identify workspace packages at exact depth (`*` is a direct
+  // child). A nested package.json under that package is still owned: the
+  // ancestor directory of glob length must itself be a package.
+  function positiveOwns(pattern) {
+    const patternSegments = normalizeWorkspacePattern(pattern).split('/').filter(Boolean);
+    if (!patternSegments.length) return false;
+    if (patternSegments.includes('**')) return matchGlobSegments(patternSegments, relSegments);
+    if (relSegments.length < patternSegments.length) return false;
+    for (let i = 0; i < patternSegments.length; i++) {
+      if (!segmentMatches(patternSegments[i], relSegments[i])) return false;
+    }
+    if (relSegments.length === patternSegments.length) return true;
+    const ancestorDir = path.join(root, ...relSegments.slice(0, patternSegments.length));
+    return fs.existsSync(path.join(ancestorDir, 'package.json'));
+  }
+
+  function groupOwns(rawPatterns) {
+    const patterns = rawPatterns.map(normalizeWorkspacePattern).filter(Boolean);
+    if (!patterns.length) return null;
+    const excluded = patterns.some((pattern) => (
+      pattern.startsWith('!') && matchesNegation(pattern.slice(1))
+    ));
+    const included = patterns.filter((pattern) => !pattern.startsWith('!')).some(positiveOwns);
+    if (!excluded && !included) return null;
+    if (excluded) return false;
+    return true;
+  }
+
+  const [impeccable, pkg] = readWorkspacePatternGroups(root);
+  const fromImpeccable = groupOwns(impeccable);
+  if (fromImpeccable !== null) return fromImpeccable;
+  const fromPkg = groupOwns(pkg);
+  if (fromPkg !== null) return fromPkg;
+  if ([...impeccable, ...pkg].some((pattern) => !normalizeWorkspacePattern(pattern).startsWith('!'))) {
+    return false;
+  }
+  return relSegments.length >= 2 && MONOREPO_FALLBACK_PROJECT_DIRS.includes(relSegments[0]);
+}
+
+// Both forms of the home directory. The walk compares path strings, and a
+// symlinked home (e.g. /home -> /var/home) never string-matches the physical
+// paths a cwd-resolved target produces, which would let the post-boundary walk
+// sail through $HOME and inherit from it.
+function homeDirForms() {
+  const homeDir = path.resolve(os.homedir());
+  const forms = new Set([homeDir]);
+  try {
+    forms.add(fs.realpathSync(homeDir));
+  } catch { /* keep the logical form only */ }
+  return forms;
+}
+
 // Walk up from `startDir` to the directory that governs the target's design
 // system, mirroring skill/scripts/context.mjs's project-boundary semantics:
 //
 //   - A directory carrying a DESIGN.md (directly or in a fallback dir) IS the
 //     design root — that's where the rules live.
 //   - A directory carrying a project marker (.git / package.json / .impeccable)
-//     but no DESIGN.md is a project BOUNDARY: the walk stops with no design
-//     system, so a sibling project never inherits a parent's or cwd's rules.
+//     but no DESIGN.md is a project BOUNDARY. A nested package.json inherits
+//     the ancestor DESIGN.md only when that ancestor's workspace declarations
+//     include the path (negations win; a nested package under a matched
+//     workspace still inherits). Marker-only roots (turbo/nx/lerna/pnpm
+//     with no globs) still own apps/<name> and packages/<name>. A stray nested
+//     package that matches no glob does not inherit. This is detect's
+//     contamination contract, not skill-context's repoRoot fallback for
+//     excluded paths. A nested separate repository (.git with no workspace
+//     declaration) still inherits nothing (issue #570).
 //   - Reaching the home directory / filesystem root with neither means no
 //     design system at all — never process.cwd()'s.
 //
@@ -590,15 +760,33 @@ function designSystemStartDir(targetPath, cwd = process.cwd()) {
 // runs out. This is the fix for cross-project contamination.
 export function findDesignRoot(startDir) {
   let dir = path.resolve(startDir);
-  const homeDir = path.resolve(os.homedir());
+  const homeDirs = homeDirForms();
+  let boundary = null;
   while (true) {
-    if (resolveDesignMdPath(dir)) return { dir, hasDesign: true };
-    if (PROJECT_ROOT_MARKERS.some((marker) => fs.existsSync(path.join(dir, marker)))) {
-      return { dir, hasDesign: false };
+    if (!boundary && resolveDesignMdPath(dir)) return { dir, hasDesign: true };
+    if (boundary) {
+      // Past the boundary the walk only looks for the monorepo root that owns
+      // the workspace path (workspace globs including negations, or marker-only
+      // apps/packages fallback). Monorepo-root before .git, same order as
+      // context.mjs: a workspace root carrying its own .git is still recognized,
+      // while a .git that declares no workspaces is a separate repository and
+      // stops the walk with nothing inherited. The home directory is never an
+      // owning root, same as context.mjs's findMonorepoRoot, which stops at
+      // homeDir before its monorepo check.
+      if (!homeDirs.has(dir) && isMonorepoRoot(dir)) {
+        if (monorepoOwnsPath(dir, boundary.dir)) return { dir, hasDesign: !!resolveDesignMdPath(dir) };
+        return boundary;
+      }
+      if (fs.existsSync(path.join(dir, '.git'))) return boundary;
+    } else if (PROJECT_ROOT_MARKERS.some((marker) => fs.existsSync(path.join(dir, marker)))) {
+      boundary = { dir, hasDesign: false };
+      // A boundary that is itself a monorepo root, or a separate repository
+      // with its own .git, inherits nothing from above.
+      if (isMonorepoRoot(dir) || fs.existsSync(path.join(dir, '.git'))) return boundary;
     }
-    if (dir === homeDir) return null;
+    if (homeDirs.has(dir)) return boundary;
     const parent = path.dirname(dir);
-    if (parent === dir) return null;
+    if (parent === dir) return boundary;
     dir = parent;
   }
 }

.claude/skills/impeccable/scripts/detector/detect-antipatterns-browser.js

@@ -773,14 +773,22 @@ function extractColorFunctionTokens(value) {
 function parseGradientColors(bgImage) {
   if (!bgImage || !bgImage.includes('gradient')) return [];
   const colors = [];
+  const tokenSpans = [];
+  let from = 0;
   // Stops arrive in whatever syntax the author wrote and the browser kept.
   // A dark ground painted as `linear-gradient(oklch(...), oklch(...))` used
   // to read as a gradient with no stops at all.
   for (const token of extractColorFunctionTokens(bgImage)) {
+    const start = bgImage.indexOf(token, from);
+    if (start < 0) break;
+    tokenSpans.push({ start, end: start + token.length });
+    from = start + token.length;
     const c = parseAnyColor(token);
     if (c) colors.push(c);
   }
   for (const m of bgImage.matchAll(/#([0-9a-f]{6}|[0-9a-f]{3})\b/gi)) {
+    // Nested hex inside color-mix is an ingredient, not a stop (issue #578).
+    if (tokenSpans.some(s => m.index >= s.start && m.index < s.end)) continue;
     const h = m[1];
     if (h.length === 6) {
       colors.push({ r: parseInt(h.slice(0,2),16), g: parseInt(h.slice(2,4),16), b: parseInt(h.slice(4,6),16), a: 1 });
@@ -1955,20 +1963,19 @@ function scanCssTextForGlow(content) {
   return results;
 }
 
-// Decorative grid or line-field backgrounds drawn with hairline
+// Decorative two-axis grid backgrounds drawn with hairline
 // linear-gradient layers tiled by a fixed pixel cell. Shared by the HTML
 // pattern pass and the regex source engine so standalone CSS, component
 // styles, and inline styles receive the same coverage. Both signals must
 // co-occur in one declaration block; unrelated rules must not add up across
-// the file. Returns [{ index, snippet }], capped at one finding per source to
-// match the page-level HTML check's existing behavior.
+// the file. A single hairline is a line, divider, or rail, not a grid, even
+// when tiled by a 2D px cell. Returns [{ index, snippet }], capped at one
+// finding per source to match the page-level HTML check's existing behavior.
 function scanCssTextForGridBackground(content) {
   const hairlineRe = /\b\d{1,3}px\s*,\s*transparent\s+\d{1,3}px/gi;
   const invertedHairlineRe = /transparent\s+calc\(100%\s*-\s*\d{1,3}px\)/gi;
   const sizeDeclPxRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\b/i;
-  const sizeDeclPxPairRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\s+\d{1,3}px/i;
   const shorthandPxAnyRe = /\/\s*\d{1,3}px\b/;
-  const shorthandPxPairRe = /\/\s*\d{1,3}px\s+\d{1,3}px/;
   const bgDeclRe = /\bbackground(?:-image)?\s*:\s*([^;{}"']*)/gi;
   const blockRe = /\{([^{}]*)\}|style\s*=\s*"([^"]*)"|style\s*=\s*'([^']*)'/gi;
   let blk;
@@ -1985,13 +1992,10 @@ function scanCssTextForGridBackground(content) {
     }
     if (hairlineCount === 0) continue;
     const hasPxCell = sizeDeclPxRe.test(block) || shorthandPxAnyRe.test(bgJoined);
-    const hasPxPairCell = sizeDeclPxPairRe.test(block) || shorthandPxPairRe.test(bgJoined);
-    if ((hairlineCount >= 2 && hasPxCell) || hasPxPairCell) {
+    if (hairlineCount >= 2 && hasPxCell) {
       return [{
         index: blk.index,
-        snippet: hairlineCount >= 2
-          ? 'two-axis grid-line gradient background'
-          : 'px-tiled hairline line-field background',
+        snippet: 'two-axis grid-line gradient background',
       }];
     }
   }
@@ -3986,7 +3990,7 @@ function checkElementAIPaletteDOM(el) {
   }
 
   // Check for neon text (vivid cyan/purple color on dark background)
-  const textColor = parseRgb(style.color);
+  const textColor = parseRgb(style.color) || parseAnyColor(style.color);
   if (textColor && hasChroma(textColor, 80)) {
     const hue = getHue(textColor);
     const isAIPalette = (hue >= 160 && hue <= 200) || (hue >= 260 && hue <= 310);
@@ -7281,7 +7285,7 @@ if (IS_BROWSER) {
       if (currentStyle.filter && currentStyle.filter !== 'none') reasons.add('filter');
       if (currentStyle.backdropFilter && currentStyle.backdropFilter !== 'none') reasons.add('backdrop filter');
 
-      const solidBg = parseRgb(currentStyle.backgroundColor);
+      const solidBg = parseRgb(currentStyle.backgroundColor) || parseAnyColor(currentStyle.backgroundColor);
       if (solidBg && solidBg.a >= 0.95 && (!bgImage || bgImage === 'none')) break;
       current = current.parentElement;
     }
@@ -7343,7 +7347,7 @@ if (IS_BROWSER) {
       // starve the url()-backed texts this mode exists to sample.
       if (options.imageOnly && !reasons.includes('image background')) continue;
 
-      const textColor = parseRgb(style.color);
+      const textColor = parseRgb(style.color) || parseAnyColor(style.color);
       const fontSize = parseFloat(style.fontSize) || 16;
       const fontWeight = parseInt(style.fontWeight) || 400;
       const isLargeText = fontSize >= WCAG_LARGE_TEXT_PX || (fontSize >= WCAG_LARGE_BOLD_TEXT_PX && fontWeight >= 700);
@@ -7640,7 +7644,7 @@ if (IS_BROWSER) {
         return sample;
       }
     }
-    const bg = parseRgb(style.backgroundColor);
+    const bg = parseRgb(style.backgroundColor) || parseAnyColor(style.backgroundColor);
     if (bg && bg.a > 0.05) return { status: 'sampled', color: bg, method: 'solid-background' };
     return { status: 'unresolved', reason: 'no readable background' };
   }
@@ -7770,7 +7774,7 @@ if (IS_BROWSER) {
     }
 
     const style = getComputedStyle(el);
-    const textColor = parseRgb(style.color) || candidate.textColor;
+    const textColor = parseRgb(style.color) || parseAnyColor(style.color) || candidate.textColor;
     if (!textColor) return { ...candidate, status: 'unresolved', confidence: 'none', reason: 'unreadable text color' };
 
     const rect = getDirectTextRect(el) || el.getBoundingClientRect();

.claude/skills/impeccable/scripts/detector/engines/regex/detect-text.mjs

@@ -42,6 +42,7 @@ function shouldRunPageAnalyzers(content, filePath) {
 }
 
 const JS_SOURCE_EXTS = new Set(['.js', '.jsx', '.ts', '.tsx', '.mjs', '.cjs']);
+const STYLESHEET_EXTS = new Set(['.css', '.scss', '.sass', '.less']);
 const REGEX_PREFIX_KEYWORDS = new Set(['await', 'case', 'default', 'delete', 'do', 'else', 'in', 'instanceof', 'new', 'of', 'return', 'throw', 'typeof', 'void', 'yield']);
 const BLOCK_BRACE_PREFIX_KEYWORDS = new Set(['do', 'else', 'finally', 'try']);
 
@@ -256,6 +257,153 @@ function stripCssComments(content) {
   return content.replace(/\/\*[\s\S]*?\*\//g, comment => comment.replace(/[^\n]/g, ' '));
 }
 
+function blankHtmlComments(text) {
+  return text.replace(/<!--[\s\S]*?-->/g, comment => comment.replace(/[^\n]/g, ' '));
+}
+
+function blankCssLineCommentsInStyleBlocks(text) {
+  const re = /<style\b[^>]*>([\s\S]*?)<\/style>/gi;
+  let output = '';
+  let lastIndex = 0;
+  let match;
+  while ((match = re.exec(text)) !== null) {
+    const inner = match[1];
+    const openLength = match[0].length - inner.length - '</style>'.length;
+    output += text.slice(lastIndex, match.index);
+    output += match[0].slice(0, openLength);
+    output += blankCssLineComments(inner);
+    output += match[0].slice(openLength + inner.length);
+    lastIndex = re.lastIndex;
+  }
+  return output + text.slice(lastIndex);
+}
+
+function blankHtmlAndCssCommentsOutsideScripts(text) {
+  const re = /<script\b[^>]*>[\s\S]*?<\/script>/gi;
+  let output = '';
+  let lastIndex = 0;
+  let match;
+  while ((match = re.exec(text)) !== null) {
+    output += blankCssLineCommentsInStyleBlocks(stripCssComments(blankHtmlComments(text.slice(lastIndex, match.index))));
+    output += match[0];
+    lastIndex = re.lastIndex;
+  }
+  return output + blankCssLineCommentsInStyleBlocks(stripCssComments(blankHtmlComments(text.slice(lastIndex))));
+}
+
+function blankCssLineComments(text) {
+  let output = '';
+  let state = 'code';
+  let urlDepth = 0;
+  for (let i = 0; i < text.length; i++) {
+    const char = text[i];
+    const next = text[i + 1];
+    if (state === 'line') {
+      if (char === '\n') {
+        output += '\n';
+        state = 'code';
+      } else {
+        output += ' ';
+      }
+      continue;
+    }
+    if (state === 'single' || state === 'double') {
+      output += char;
+      if (char === '\\' && next) {
+        output += next;
+        i++;
+      } else if ((state === 'single' && char === "'") || (state === 'double' && char === '"')) {
+        state = 'code';
+      }
+      continue;
+    }
+    const prev = output.length ? output[output.length - 1] : '';
+    if (char === '/' && next === '/' && urlDepth === 0 && prev !== ':' && prev !== '(' && prev !== '\\') {
+      output += '  ';
+      i++;
+      state = 'line';
+      continue;
+    }
+    if (char === "'") state = 'single';
+    else if (char === '"') state = 'double';
+    if (char === '(') {
+      const behind = output.replace(/\s+$/, '');
+      if (urlDepth > 0 || /url$/i.test(behind)) urlDepth++;
+    } else if (char === ')' && urlDepth) {
+      urlDepth--;
+    }
+    output += char;
+  }
+  return output;
+}
+
+function findAstroFrontmatterClose(text) {
+  if (!text.startsWith('---')) return -1;
+  let cursor = text.indexOf('\n');
+  if (cursor === -1) return -1;
+  cursor += 1;
+  while (cursor < text.length) {
+    if (text[cursor - 1] === '\n' && text.startsWith('---', cursor)) {
+      let end = cursor + 3;
+      while (text[end] === ' ' || text[end] === '\t') end++;
+      if (end >= text.length || text[end] === '\n' || text[end] === '\r') return cursor - 1;
+    }
+    const char = text[cursor];
+    const next = text[cursor + 1];
+    if (char === "'" || char === '"') {
+      const close = findQuotedStringEnd(text, cursor, char);
+      if (close === -1) return -1;
+      cursor = close + 1;
+      continue;
+    }
+    if (char === '`') {
+      const close = findTemplateLiteralEnd(text, cursor);
+      if (close === -1) return -1;
+      cursor = close + 1;
+      continue;
+    }
+    if (char === '/' && next === '/') {
+      const lineEnd = text.indexOf('\n', cursor);
+      if (lineEnd === -1) return -1;
+      cursor = lineEnd;
+      continue;
+    }
+    if (char === '/' && next === '*') {
+      const commentEnd = text.indexOf('*/', cursor + 2);
+      if (commentEnd === -1) return -1;
+      cursor = commentEnd + 2;
+      continue;
+    }
+    if (char === '/' && next !== '/' && next !== '*') {
+      const close = findRegexLiteralEnd(text, cursor);
+      if (close !== -1) {
+        cursor = close + 1;
+        continue;
+      }
+    }
+    cursor++;
+  }
+  return -1;
+}
+
+function blankAstroFrontmatterComments(text) {
+  const close = findAstroFrontmatterClose(text);
+  if (close === -1) return text;
+  return stripJsComments(text.slice(0, close)) + text.slice(close);
+}
+
+function blankCommentsForMatchers(text, ext) {
+  if (PAGE_ANALYZER_EXTS.has(ext)) {
+    const withFrontmatter = ext === '.astro' ? blankAstroFrontmatterComments(text) : text;
+    return blankHtmlAndCssCommentsOutsideScripts(withFrontmatter);
+  }
+  if (STYLESHEET_EXTS.has(ext)) {
+    const withoutBlocks = stripCssComments(text);
+    return ext === '.css' ? withoutBlocks : blankCssLineComments(withoutBlocks);
+  }
+  return text;
+}
+
 function firstOverusedGoogleFont(text) {
   return extractGoogleFontFamilies(text).find(f => OVERUSED_FONTS.has(f)) || '';
 }
@@ -1028,14 +1176,13 @@ function detectText(content, filePath, options = {}) {
   const ext = extFromFilePath(filePath);
   const commentStrippedSource = JS_SOURCE_EXTS.has(ext) ? stripJsComments(content, {
     jsx: ext === '.js' || ext === '.jsx' || ext === '.tsx',
-  }) : content;
+  }) : blankCommentsForMatchers(content, ext);
   const source = stripCssInJsComments(commentStrippedSource, ext);
   const lines = source.split('\n');
 
   // Run regex matchers on the full file content (catches Tailwind classes, inline styles)
   // Enable block context for CSS files where related properties span multiple lines
-  const cssLike = new Set(['.css', '.scss', '.sass', '.less']);
-  findings.push(...runRegexMatchers(lines, filePath, 0, cssLike.has(ext) || null, {
+  findings.push(...runRegexMatchers(lines, filePath, 0, STYLESHEET_EXTS.has(ext) || null, {
     profile,
     phase: 'source',
   }));
@@ -1050,7 +1197,7 @@ function detectText(content, filePath, options = {}) {
     scanCssTextForPseudoStripe(text).map(hit =>
       finding(hit.id, filePath, hit.snippet, lineOffset + text.slice(0, hit.index).split('\n').length));
 
-  if (cssLike.has(ext)) {
+  if (STYLESHEET_EXTS.has(ext)) {
     findings.push(...scanInsetStripeCss(content, filePath));
     findings.push(...pseudoStripeFindings(content, 0));
   }
@@ -1078,7 +1225,8 @@ function detectText(content, filePath, options = {}) {
     }, () => extractStyleBlocks(content, ext))
     : extractStyleBlocks(content, ext);
   for (const block of styleBlocks) {
-    const blockLines = block.content.split('\n');
+    const blockContent = blankCssLineComments(stripCssComments(block.content));
+    const blockLines = blockContent.split('\n');
     findings.push(...runRegexMatchers(blockLines, filePath, block.startLine - 1, true, {
       profile,
       phase: 'style-block',
@@ -1089,8 +1237,8 @@ function detectText(content, filePath, options = {}) {
     // 1-based, so the offset is startLine - 2; startLine - 1 double-counted and
     // reported every selector one line low. runRegexMatchers keeps startLine - 1
     // because it indexes its split lines from zero.
-    findings.push(...scanInsetStripeCss(block.content, filePath, block.startLine - 2));
-    findings.push(...pseudoStripeFindings(block.content, block.startLine - 2));
+    findings.push(...scanInsetStripeCss(blockContent, filePath, block.startLine - 2));
+    findings.push(...pseudoStripeFindings(blockContent, block.startLine - 2));
   }
 
   // Extract and scan CSS-in-JS template literals

.claude/skills/impeccable/scripts/detector/engines/static-html/css-cascade.mjs

@@ -835,10 +835,11 @@ class StaticElement {
     }
   }
   closest(selector) {
+    const matcher = this._doc.matcherFor(selector);
     let cur = this.node;
     while (cur && cur.type === 'tag') {
       try {
-        if (this._doc.is(cur, selector)) return this._doc.wrap(cur);
+        if (matcher(cur)) return this._doc.wrap(cur);
       } catch {
         return null;
       }
@@ -862,9 +863,10 @@ class StaticDocument {
     this.root = root;
     this.selectAll = modules.selectAll;
     this.selectOne = modules.selectOne;
-    this.is = modules.is;
+    this.compile = modules.compile;
     this.domutils = modules.domutils;
     this._wrappers = new WeakMap();
+    this._compiledSelectors = new Map();
     this._styleMap = new WeakMap();
     this._hoverStyleMap = new WeakMap();
     this._accentDashPseudo = new WeakSet();
@@ -882,6 +884,20 @@ class StaticDocument {
     }
     return wrapped;
   }
+  matcherFor(selector) {
+    let matcher = this._compiledSelectors.get(selector);
+    if (!matcher) {
+      try {
+        matcher = this.compile(selector);
+      } catch (err) {
+        // Cache the failure as a rethrower so a bad selector still reaches
+        // closest()'s catch on every call, first and repeat alike.
+        matcher = () => { throw err; };
+      }
+      this._compiledSelectors.set(selector, matcher);
+    }
+    return matcher;
+  }
   querySelectorAll(selector) {
     try {
       return this.selectAll(selector, this.root.children || []).map(node => this.wrap(node));
@@ -948,8 +964,34 @@ function buildStaticWindow(staticDoc) {
   };
 }
 
+function resolveLinkedCssPath(fileDir, href) {
+  const stripped = href.split(/[?#]/)[0];
+  const rootRelative = stripped.startsWith('/') && !stripped.startsWith('//');
+  if (!rootRelative) return path.resolve(fileDir, stripped);
+  // Drop "." and reject ".." so /../outside.css cannot walk out of dir.
+  const segments = stripped.replace(/^\/+/, '').split(/[/\\]/).filter(p => p && p !== '.');
+  if (segments.some(p => p === '..')) return path.join(fileDir, segments.filter(p => p !== '..').join(path.sep));
+  const rel = segments.join(path.sep);
+  let dir = fileDir;
+  for (;;) {
+    const parent = path.dirname(dir);
+    if (parent === dir) break; // never use the filesystem root as document root
+    try {
+      const candidate = path.join(dir, rel);
+      if (fs.statSync(candidate).isFile()) return candidate;
+    } catch { /* missing or unreadable candidate */ }
+    // Stop at the project root so a coincidental ~/static/app.css cannot win.
+    try {
+      if (fs.existsSync(path.join(dir, 'package.json')) || fs.existsSync(path.join(dir, '.git'))) break;
+    } catch { /* unreadable marker */ }
+    dir = parent;
+  }
+  return path.join(fileDir, rel);
+}
+
 function collectStaticCssText(root, fileDir, profile, filePath, modules) {
   const styleTexts = [];
+  const warnedMissingStylesheets = new Set();
   for (const styleEl of modules.selectAll('style', root.children || [])) {
     styleTexts.push(modules.domutils.textContent(styleEl));
   }
@@ -958,10 +1000,10 @@ function collectStaticCssText(root, fileDir, profile, filePath, modules) {
     const rel = link.attribs?.rel || '';
     const href = link.attribs?.href || '';
     if (!/\bstylesheet\b/i.test(rel) || !href || /^(https?:)?\/\//i.test(href)) continue;
-    // Cache-busting hrefs (styles.css?v=3) resolve to the file, not to a
-    // literal path with the query in it; a versioned link otherwise made the
-    // whole stylesheet invisible to every element-level check.
-    const cssPath = path.resolve(fileDir, href.split(/[?#]/)[0]);
+    // Cache-busting (styles.css?v=3) and root-relative (/static/app.css) hrefs
+    // must not resolve as OS-absolute paths; otherwise the whole stylesheet is
+    // invisible to every element-level check.
+    const cssPath = resolveLinkedCssPath(fileDir, href);
     try {
       const css = profileStep(profile, {
         engine: 'static-html',
@@ -971,7 +1013,14 @@ function collectStaticCssText(root, fileDir, profile, filePath, modules) {
         detail: href,
       }, () => fs.readFileSync(cssPath, 'utf-8'));
       styleTexts.push(css);
-    } catch { /* skip unreadable */ }
+    } catch {
+      if (!warnedMissingStylesheets.has(cssPath)) {
+        warnedMissingStylesheets.add(cssPath);
+        process.stderr.write(
+          `impeccable detect: could not read linked stylesheet ${href} (resolved to ${cssPath}); color and custom-property rules will be incomplete\n`
+        );
+      }
+    }
   }
   return styleTexts.join('\n');
 }

.claude/skills/impeccable/scripts/detector/engines/static-html/detect-html.mjs

@@ -134,7 +134,7 @@ async function detectHtml(filePath, options = {}) {
         parseDocument: htmlparser2.parseDocument,
         selectAll: cssSelect.selectAll,
         selectOne: cssSelect.selectOne,
-        is: cssSelect.is,
+        compile: cssSelect.compile,
         csstree,
         domutils,
       };

.claude/skills/impeccable/scripts/detector/rules/checks.mjs

@@ -721,20 +721,19 @@ function scanCssTextForGlow(content) {
   return results;
 }
 
-// Decorative grid or line-field backgrounds drawn with hairline
+// Decorative two-axis grid backgrounds drawn with hairline
 // linear-gradient layers tiled by a fixed pixel cell. Shared by the HTML
 // pattern pass and the regex source engine so standalone CSS, component
 // styles, and inline styles receive the same coverage. Both signals must
 // co-occur in one declaration block; unrelated rules must not add up across
-// the file. Returns [{ index, snippet }], capped at one finding per source to
-// match the page-level HTML check's existing behavior.
+// the file. A single hairline is a line, divider, or rail, not a grid, even
+// when tiled by a 2D px cell. Returns [{ index, snippet }], capped at one
+// finding per source to match the page-level HTML check's existing behavior.
 function scanCssTextForGridBackground(content) {
   const hairlineRe = /\b\d{1,3}px\s*,\s*transparent\s+\d{1,3}px/gi;
   const invertedHairlineRe = /transparent\s+calc\(100%\s*-\s*\d{1,3}px\)/gi;
   const sizeDeclPxRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\b/i;
-  const sizeDeclPxPairRe = /background-size\s*:[^;{}"']*\b\d{1,3}px\s+\d{1,3}px/i;
   const shorthandPxAnyRe = /\/\s*\d{1,3}px\b/;
-  const shorthandPxPairRe = /\/\s*\d{1,3}px\s+\d{1,3}px/;
   const bgDeclRe = /\bbackground(?:-image)?\s*:\s*([^;{}"']*)/gi;
   const blockRe = /\{([^{}]*)\}|style\s*=\s*"([^"]*)"|style\s*=\s*'([^']*)'/gi;
   let blk;
@@ -751,13 +750,10 @@ function scanCssTextForGridBackground(content) {
     }
     if (hairlineCount === 0) continue;
     const hasPxCell = sizeDeclPxRe.test(block) || shorthandPxAnyRe.test(bgJoined);
-    const hasPxPairCell = sizeDeclPxPairRe.test(block) || shorthandPxPairRe.test(bgJoined);
-    if ((hairlineCount >= 2 && hasPxCell) || hasPxPairCell) {
+    if (hairlineCount >= 2 && hasPxCell) {
       return [{
         index: blk.index,
-        snippet: hairlineCount >= 2
-          ? 'two-axis grid-line gradient background'
-          : 'px-tiled hairline line-field background',
+        snippet: 'two-axis grid-line gradient background',
       }];
     }
   }
@@ -2752,7 +2748,7 @@ function checkElementAIPaletteDOM(el) {
   }
 
   // Check for neon text (vivid cyan/purple color on dark background)
-  const textColor = parseRgb(style.color);
+  const textColor = parseRgb(style.color) || parseAnyColor(style.color);
   if (textColor && hasChroma(textColor, 80)) {
     const hue = getHue(textColor);
     const isAIPalette = (hue >= 160 && hue <= 200) || (hue >= 260 && hue <= 310);

.claude/skills/impeccable/scripts/detector/shared/color.mjs

@@ -103,14 +103,22 @@ function extractColorFunctionTokens(value) {
 function parseGradientColors(bgImage) {
   if (!bgImage || !bgImage.includes('gradient')) return [];
   const colors = [];
+  const tokenSpans = [];
+  let from = 0;
   // Stops arrive in whatever syntax the author wrote and the browser kept.
   // A dark ground painted as `linear-gradient(oklch(...), oklch(...))` used
   // to read as a gradient with no stops at all.
   for (const token of extractColorFunctionTokens(bgImage)) {
+    const start = bgImage.indexOf(token, from);
+    if (start < 0) break;
+    tokenSpans.push({ start, end: start + token.length });
+    from = start + token.length;
     const c = parseAnyColor(token);
     if (c) colors.push(c);
   }
   for (const m of bgImage.matchAll(/#([0-9a-f]{6}|[0-9a-f]{3})\b/gi)) {
+    // Nested hex inside color-mix is an ingredient, not a stop (issue #578).
+    if (tokenSpans.some(s => m.index >= s.start && m.index < s.end)) continue;
     const h = m[1];
     if (h.length === 6) {
       colors.push({ r: parseInt(h.slice(0,2),16), g: parseInt(h.slice(2,4),16), b: parseInt(h.slice(4,6),16), a: 1 });

.claude/skills/impeccable/scripts/doctor.mjs

@@ -33,13 +33,8 @@ import {
   stampProductSchema,
 } from './lib/artifact-schema.mjs';
 import {
-  checkBuildPathUnset,
-  checkConfig,
-  checkDesignSidecar,
+  collectBootFindingGroups,
   checkNativePlatformEvidence,
-  checkProduct,
-  checkProjectRoots,
-  checkSurfaceBriefs,
   designSidecarCandidatesFor,
 } from './lib/staleness.mjs';
 import {
@@ -106,34 +101,30 @@ async function collect(cwd, targetOptions) {
     extractPlatform,
     readFile: safeRead,
   });
+  const bootFindings = collectBootFindingGroups(ctx, {
+    absDesignPath,
+    sidecarCandidates,
+    projectRootPatterns: readProjectRootPatterns(ctx.repoRoot),
+    targetCandidates: workspaceCandidates,
+  });
 
   const findings = [
-    ...checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
-    ...(ctx.product
-      ? checkNativePlatformEvidence({
-          projectRoot,
-          platform: ctx.platform,
-          product: ctx.product,
-          productPath: ctx.productPath,
-        })
-      : []),
-    ...checkDesignSidecar({ designPath: absDesignPath, sidecarCandidates, projectRoot }),
+    ...bootFindings.product,
+    ...bootFindings.nativePlatform,
+    ...bootFindings.designSidecar,
     ...checkDesignDrift({ designPath: absDesignPath, projectRoot }),
     ...checkDesignCoverage({ design: ctx.design, designPath: ctx.designPath, parseDesignMd }),
-    ...checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
-    ...checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
+    ...bootFindings.config,
+    ...bootFindings.buildPath,
     ...checkDetectorIgnores({ projectRoot, knownRuleIds }),
-    ...checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
+    ...bootFindings.surfaceBriefs,
     ...checkHookInstallation({
       projectRoot,
       repoRoot: ctx.repoRoot,
       providerId: IMPECCABLE_PROVIDER_ID,
     }),
     ...checkLegacyLiveState({ projectRoot }),
-    ...checkProjectRoots({
-      patterns: readProjectRootPatterns(ctx.repoRoot),
-      candidates: workspaceCandidates,
-    }),
+    ...bootFindings.projectRoots,
     ...workspaceResult.findings,
   ];
 

.claude/skills/impeccable/scripts/embed-prompt.mjs

@@ -5,6 +5,7 @@
 //   node embed-prompt.mjs <image> --prompt "the prompt text"
 //   node embed-prompt.mjs <image> --prompt-file prompt.txt
 //   node embed-prompt.mjs <image> --read
+//   node embed-prompt.mjs --scan <dir...>   # list rasters missing a prompt; exit 3 when any
 //
 // Formats: PNG (tEXt chunk, keyword "impeccable:prompt"), JPEG (COM segment).
 // WebP and anything else fall back to a `<image>.json` sidecar; --read checks
@@ -21,8 +22,49 @@ const KEYWORD = 'impeccable:prompt';
 const args = process.argv.slice(2);
 const file = args.find(a => !a.startsWith('--'));
 const readMode = args.includes('--read');
+const scanMode = args.includes('--scan');
 const argOf = (name) => { const i = args.indexOf(name); return i !== -1 ? args[i + 1] : null; };
 
+function promptOf(imagePath) {
+  const b = fs.readFileSync(imagePath);
+  let prompt = null;
+  if (b.length > 8 && b.readUInt32BE(0) === 0x89504e47) prompt = readPngText(b);
+  else if (b.length > 3 && b[0] === 0xff && b[1] === 0xd8) prompt = readJpegCom(b);
+  if (prompt == null && fs.existsSync(`${imagePath}.json`)) {
+    try { prompt = JSON.parse(fs.readFileSync(`${imagePath}.json`, 'utf8')).prompt ?? null; } catch { /* stays null */ }
+  }
+  return prompt;
+}
+
+if (scanMode) {
+  const targets = args.filter(a => !a.startsWith('--'));
+  if (targets.length === 0) { console.error('embed-prompt: --scan needs at least one directory'); process.exit(1); }
+  const RASTER = /\.(png|jpe?g|webp)$/i;
+  const rasters = [];
+  const walk = (p, isRoot) => {
+    const stat = fs.statSync(p);
+    if (stat.isDirectory()) {
+      const base = p.replace(/\/+$/, '').split('/').pop();
+      // Skip installed deps and hidden dirs found during the walk, but honor a
+      // hidden dir the caller passed explicitly (e.g. .impeccable/mocks).
+      if (!isRoot && (base === 'node_modules' || base.startsWith('.'))) return;
+      for (const entry of fs.readdirSync(p)) walk(`${p.replace(/\/+$/, '')}/${entry}`, false);
+    } else if (RASTER.test(p)) {
+      rasters.push(p);
+    }
+  };
+  for (const target of targets) {
+    if (!fs.existsSync(target)) { console.error(`embed-prompt: no such path ${target}`); process.exit(1); }
+    walk(target, true);
+  }
+  let missing = 0;
+  for (const raster of rasters) {
+    if (promptOf(raster) == null) { console.log(`MISSING: ${raster}`); missing++; }
+  }
+  console.log(`SCAN: ${rasters.length} raster${rasters.length === 1 ? '' : 's'}, ${missing} missing`);
+  process.exit(missing > 0 ? 3 : 0);
+}
+
 if (!file || !fs.existsSync(file)) { console.error('embed-prompt: image file required'); process.exit(1); }
 
 const buf = fs.readFileSync(file);

.claude/skills/impeccable/scripts/hook-admin.mjs

@@ -75,11 +75,11 @@ const HOOK_MANIFEST_TARGETS = [
     destRel: '.claude/settings.local.json',
     sharedDestRel: '.claude/settings.json',
     manifest: () => ({
-      description: 'Impeccable design detector: immediate-tier checks after Edit/Write/MultiEdit on UI files, full-rule deep pass on Stop.',
+      description: 'Impeccable design detector: immediate-tier checks after Edit/Write on UI files, full-rule deep pass on Stop.',
       hooks: {
         PostToolUse: [
           {
-            matcher: 'Edit|Write|MultiEdit',
+            matcher: 'Edit|Write',
             hooks: [
               {
                 type: 'command',

.claude/skills/impeccable/scripts/hook-lib.mjs

@@ -816,9 +816,9 @@ export function splitFindingsByTier(findings) {
 }
 
 // Whether the per-edit pass for this harness should defer non-immediate
-// findings to a Stop deep pass. Only Claude Code and Codex dispatch our Stop
-// hook; Cursor and GitHub Copilot have no deep pass wired, so deferring for
-// them would silently drop the non-immediate rules entirely.
+// findings to a Stop deep pass. Claude Code, Codex, and Grok Build dispatch
+// our Stop hook; Cursor and GitHub Copilot have no deep pass wired, so
+// deferring for them would silently drop the non-immediate rules entirely.
 export function perEditTieringActive(config, harness) {
   if (harness === 'cursor' || harness === 'github') return false;
   return (config?.perEditRules || DEFAULT_CONFIG.perEditRules) !== 'all';
@@ -1251,18 +1251,50 @@ export function resolveHarness(env = {}, event = null) {
   const explicit = env?.IMPECCABLE_HOOK_HARNESS;
   if (explicit === 'cursor') return 'cursor';
   if (explicit === 'github') return 'github';
-  if (explicit === 'claude' || explicit === 'codex') return 'claude';
-  // GitHub Copilot's postToolUse event uses camelCase `toolName`/`toolArgs` and
-  // has no `tool_name`/`tool_input`. That shape is the discriminator.
+  if (explicit === 'grok') return 'grok';
+  if (explicit === 'claude') return 'claude';
+  if (explicit === 'codex') return 'codex';
+  // Grok Build sends camelCase `toolName`/`toolInput`/`hookEventName` and no
+  // snake_case pair. GitHub Copilot sends camelCase `toolName`/`toolArgs`.
+  // Check Grok first: the old GitHub heuristic (`toolName` and no
+  // `tool_input`) also matches Grok, which is how live PostToolUse was
+  // classified as Copilot and then skipped with no-file-path (#646).
+  if (looksLikeGrokEnvelope(event)) return 'grok';
   if (event && typeof event === 'object'
     && (typeof event.toolName === 'string' || event.toolArgs !== undefined)
     && event.tool_name === undefined && event.tool_input === undefined) {
     return 'github';
   }
   if (typeof event?.conversation_id === 'string' && event.conversation_id) return 'cursor';
+  // Codex turn-scoped events carry `turn_id`. Claude Code does not. Detecting
+  // it here means an already-installed Codex hook emits the Codex Stop
+  // contract without rewriting the hook command to set IMPECCABLE_HOOK_HARNESS.
+  // https://developers.openai.com/codex/hooks#stop
+  if (typeof event?.turn_id === 'string' && event.turn_id) return 'codex';
   return 'claude';
 }
 
+function looksLikeGrokEnvelope(event) {
+  if (!event || typeof event !== 'object') return false;
+  if (event.hook_event_name !== undefined
+    || event.tool_name !== undefined
+    || event.tool_input !== undefined) {
+    return false;
+  }
+  if (event.toolArgs !== undefined) return false;
+  if (typeof event.hookEventName === 'string') return true;
+  return typeof event.toolName === 'string' && event.toolInput !== undefined;
+}
+
+// Stop arrives as Claude's `hook_event_name: "Stop"` or Grok Build's
+// `hookEventName: "stop"`. hook.mjs routes on the raw stdin, before any
+// normalize, so both casings must match here.
+export function isStopEvent(event) {
+  if (!event || typeof event !== 'object') return false;
+  const name = event.hook_event_name || event.hookEventName;
+  return typeof name === 'string' && name.toLowerCase() === 'stop';
+}
+
 // GitHub Copilot's postToolUse payload is
 //   { sessionId, timestamp, cwd, toolName, toolArgs, toolResult }
 // mapped onto the internal `{ tool_name, tool_input, cwd, session_id }` shape.
@@ -1354,9 +1386,36 @@ function normalizeGitHubEvent(event, projectCwd) {
   };
 }
 
+// Grok Build 1.0.5 (captured 2026-08-24) sends camelCase `toolName` /
+// `toolInput` / `sessionId` / `stopHookActive`, plus `cwd` alongside a
+// trailing-slashed `workspaceRoot` (every consumer path.resolve()s, so no
+// stripping here). Only the fields the hook reads are copied; the event
+// name stays camelCase because routing already happened on the raw stdin
+// (isStopEvent) and nothing downstream reads `hook_event_name`.
+function normalizeGrokEvent(event, projectCwd) {
+  const cwd = event.cwd || event.workspaceRoot || envProjectDir(projectCwd) || projectCwd;
+  const sessionId = event.sessionId || event.session_id || 'unknown';
+  const rawInput = event.toolInput ?? event.tool_input;
+  const toolInput = rawInput && typeof rawInput === 'object' && !Array.isArray(rawInput)
+    ? { ...rawInput }
+    : {};
+  const out = {
+    ...event,
+    cwd,
+    session_id: sessionId,
+    tool_name: event.toolName || event.tool_name || null,
+    tool_input: toolInput,
+  };
+  if (event.stopHookActive !== undefined && event.stop_hook_active === undefined) {
+    out.stop_hook_active = event.stopHookActive;
+  }
+  return out;
+}
+
 export function normalizeHookEvent(event, projectCwd, harness = 'claude') {
   if (!event || typeof event !== 'object') return event;
   if (harness === 'github') return normalizeGitHubEvent(event, projectCwd);
+  if (harness === 'grok') return normalizeGrokEvent(event, projectCwd);
   if (harness !== 'cursor') return event;
 
   const cwd = event.cwd
@@ -1959,7 +2018,15 @@ export async function runHook({ stdinJson, env = {}, cwd = process.cwd(), now =
       // findings stop being remembered and a reintroduced one reads as fresh.
       // Only the immediate tier is remembered: a deferred finding the per-edit
       // pass never reported must still read as fresh to the Stop deep pass.
-      rememberFindings(cache, sessionId, filePath, immediate);
+      //
+      // Grok ignores PostToolUse stdout, so Stop is the user-visible pass.
+      // Remembering here would dedupe those findings out of Stop. Touch the
+      // file so Stop has it, and leave the finding list empty.
+      if (harness === 'grok') {
+        touchFile(cache, sessionId, filePath);
+      } else {
+        rememberFindings(cache, sessionId, filePath, immediate);
+      }
       cacheDirty = true;
 
       if (fresh.length > 0) {
@@ -2163,8 +2230,11 @@ export const STOP_MAX_FILES = 20;
  *   { exitCode, stdout, audit, emission? }
  *
  * Never throws; exits silent (and fast) when the session touched no UI
- * files. Output uses the Stop hookSpecificOutput channel: additionalContext
- * is delivered to the model and the conversation continues so it can act.
+ * files. Output goes out on the harness's Stop continuation channel: Claude
+ * Code and Grok Build read hookSpecificOutput.additionalContext, Codex takes
+ * a decision: "block" whose reason becomes the continuation prompt. Either
+ * way the findings reach the model and the conversation continues so it
+ * can act.
  */
 export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), now = Date.now, detector } = {}) {
   const audit = { ts: new Date(now()).toISOString(), event: 'Stop' };
@@ -2191,22 +2261,36 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
       return result({ skipped: 'stdin-empty', durationMs: Date.now() - started });
     }
 
-    // Claude Code's Stop-hook contract: `stop_hook_active` is true when this
-    // hook is being re-invoked only because a prior invocation kept the turn
-    // alive (here, via hookSpecificOutput.additionalContext). Re-scanning and
-    // re-blocking now would loop until Claude Code's consecutive-block cap
-    // force-ends the turn (issue #400). The prior fire already surfaced the
-    // findings; whether to act on them is the agent's call. Exit fast with no
-    // output before any scan. Only Claude Code sends this field; other
-    // harnesses omit it, so the strict `=== true` is a no-op for them. This
-    // guard makes the loop impossible regardless of the finding cache key's
-    // line-number sensitivity (out of scope here; see findingCacheKey).
+    const harness = resolveHarness(env, event);
+    audit.harness = harness;
+    event = normalizeHookEvent(event, cwd, harness);
+
+    // Stop-hook re-entry guard: `stop_hook_active` is true when this hook is
+    // being re-invoked only because a prior invocation kept the turn alive
+    // (Claude Code via hookSpecificOutput.additionalContext, Codex via a
+    // decision: "block" continuation). Re-scanning and re-blocking now could
+    // loop (issue #400). The prior fire already surfaced the findings;
+    // whether to act on them is the agent's call. Exit fast with no output
+    // before any scan. Claude Code and Codex both send this field: Codex
+    // mirrors the Claude contract (StopCommandInput in
+    // codex-rs/hooks/src/schema.rs) and latches it true for the rest of the
+    // turn once a block is honored (codex-rs/core/src/session/turn.rs). Grok
+    // sends `stopHookActive`, copied onto the snake_case field above. Cursor
+    // and GitHub Copilot omit the field, so the strict `=== true` is a no-op
+    // for them. The guard makes the loop impossible regardless of the finding
+    // cache key's line-number sensitivity (out of scope here; see
+    // findingCacheKey).
     if (event.stop_hook_active === true) {
       return result({ skipped: 'stop-hook-active', durationMs: Date.now() - started });
     }
 
-    const harness = resolveHarness(env, event);
-    audit.harness = harness;
+    // Grok fires Stop twice: `end_turn` (the gate that can inject
+    // additionalContext) then an observe-only `shutdown`. A second deep
+    // pass would re-emit the same findings. Claude omits `reason`; only
+    // skip when Grok named a reason that is not end_turn.
+    if (harness === 'grok' && typeof event.reason === 'string' && event.reason !== 'end_turn') {
+      return result({ skipped: 'stop-reason', reason: event.reason, durationMs: Date.now() - started });
+    }
 
     // A Stop event carries no file, so the session cwd is the project.
     // Umbrella-dir launches keyed their per-edit cache to the edited file's
@@ -2241,6 +2325,7 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
 
     const freshGroups = [];
     let scanned = 0;
+    let cacheDirty = false;
     for (const filePath of touched) {
       if (scanned >= STOP_MAX_FILES) break;
       if (hasPathTraversal(filePath) || SENSITIVE_PATH.test(filePath)) continue;
@@ -2261,29 +2346,39 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
       try { content = fs.readFileSync(filePath, 'utf-8'); } catch { continue; }
 
       let findings;
+      let detectorThrew = false;
       const useHtmlEngine = configuredExt
         ? configuredExt.engine === 'html'
         : (ext === '.html' || ext === '.htm');
 
       if (useHtmlEngine && typeof det.detectHtml === 'function') {
-        try { findings = await det.detectHtml(filePath, scanOptions); } catch { findings = []; }
+        try { findings = await det.detectHtml(filePath, scanOptions); } catch { findings = []; detectorThrew = true; }
       } else {
-        try { findings = await det.detectText(content, filePath, scanOptions); } catch { findings = []; }
+        try { findings = await det.detectText(content, filePath, scanOptions); } catch { findings = []; detectorThrew = true; }
       }
 
+      // A detector failure tells us nothing about the file. Leave whatever
+      // was remembered alone rather than recording an empty scan as truth.
+      if (detectorThrew) continue;
+
       // Full rule set: no tier split here. Config/inline ignores still apply,
       // and the session dedupe drops everything the per-edit pass (or an
       // earlier Stop pass) already surfaced.
       const filtered = filterFindings(findings || [], content, ext, config);
       const fresh = dedupeAgainstCache(filtered, cache, sessionId, filePath);
+      // Sync to the live scan, including empty. Remembering only `fresh`
+      // (or skipping the write on a clean Stop) left stale keys in place, so
+      // a finding that was fixed and later reintroduced never fired again.
+      rememberFindings(cache, sessionId, filePath, filtered);
+      cacheDirty = true;
       if (fresh.length > 0) {
-        rememberFindings(cache, sessionId, filePath, fresh);
         freshGroups.push({ filePath, findings: fresh });
       }
     }
     audit.scannedFiles = scanned;
 
     if (freshGroups.length === 0) {
+      if (cacheDirty) persistCache(projectCwd, cache);
       return result({ emitted: false, skipped: 'stop-clean', durationMs: Date.now() - started });
     }
 
@@ -2300,8 +2395,8 @@ export async function runStopHook({ stdinJson, env = {}, cwd = process.cwd(), no
     );
     commitFooterShown(cache, sessionId, text);
 
-    // Fresh findings earn the cache write so the next Stop fire is silent
-    // unless new issues appear; the notice flags ride along.
+    // Persist the live finding set so the next Stop fire is silent unless
+    // new issues appear; the notice flags ride along.
     persistCache(projectCwd, cache);
     return {
       exitCode: 0,
@@ -2337,6 +2432,15 @@ export function payload(text, eventName = 'PostToolUse', harness = 'claude') {
   if (harness === 'github') {
     return JSON.stringify({ additionalContext: text });
   }
+  // Codex shares Claude Code's PostToolUse additional-context shape, but its
+  // Stop schema rejects unknown fields. Findings that should continue the
+  // turn must be a top-level blocking decision.
+  // https://developers.openai.com/codex/hooks#stop (schema of record:
+  // codex-rs/hooks/src/schema.rs, StopCommandOutputWire)
+  if (harness === 'codex' && eventName === 'Stop') {
+    if (!String(text ?? '').trim()) return '';
+    return JSON.stringify({ decision: 'block', reason: text });
+  }
   return JSON.stringify({
     hookSpecificOutput: { hookEventName: eventName, additionalContext: text },
   });

.claude/skills/impeccable/scripts/hook.mjs

@@ -2,15 +2,17 @@
 /**
  * Impeccable design hook — PostToolUse + Stop entry point.
  *
- * Reads the Claude Code / Codex / Cursor hook event from stdin and routes by
- * `hook_event_name`:
+ * Reads the Claude Code / Codex / Cursor / Grok Build hook event from stdin
+ * and routes by Stop vs everything else. Claude uses `hook_event_name:
+ * "Stop"`; Grok uses `hookEventName: "stop"`.
  *
  *   - PostToolUse: runs the immediate-tier detector rules against the touched
  *     file and emits a system reminder via
- *     `hookSpecificOutput.additionalContext` when findings exist.
+ *     `hookSpecificOutput.additionalContext` when findings exist. Grok
+ *     discards that stdout; the scan still warms the session cache for Stop.
  *   - Stop: runs the FULL detector rule set over every UI file touched this
  *     session (the deep pass), deduped against what the per-edit pass already
- *     surfaced, and emits once via the Stop additionalContext channel.
+ *     surfaced, and emits once via the harness-specific continuation channel.
  *
  * Contract: never break a turn. Always exit 0. Clean files emit a small ack
  * unless quiet mode is enabled; a clean Stop pass is silent.
@@ -19,7 +21,7 @@
  * subprocess. This file is the thin stdin/stdout adapter.
  */
 
-import { runHook, runStopHook, writeAuditLog } from './hook-lib.mjs';
+import { runHook, runStopHook, writeAuditLog, isStopEvent } from './hook-lib.mjs';
 
 async function readStdin() {
   if (process.stdin.isTTY) return '';
@@ -28,10 +30,9 @@ async function readStdin() {
   return Buffer.concat(chunks).toString('utf-8');
 }
 
-function isStopEvent(stdinJson) {
+function stdinIsStop(stdinJson) {
   try {
-    const event = JSON.parse(stdinJson);
-    return event && typeof event === 'object' && event.hook_event_name === 'Stop';
+    return isStopEvent(JSON.parse(stdinJson));
   } catch {
     // Malformed stdin falls through to runHook, which audits the skip.
     return false;
@@ -48,7 +49,7 @@ async function main() {
   let stdinJson = '';
   try { stdinJson = await readStdin(); } catch { /* fall through */ }
 
-  const run = isStopEvent(stdinJson) ? runStopHook : runHook;
+  const run = stdinIsStop(stdinJson) ? runStopHook : runHook;
   const result = await run({
     stdinJson,
     env: inheritedEnv,

.claude/skills/impeccable/scripts/lib/design-parser.mjs

@@ -196,9 +196,6 @@ function parseScalar(raw) {
 
 const HEX_RE = /#[0-9a-fA-F]{3,8}\b/g;
 const OKLCH_RE = /oklch\([^)]+\)/gi;
-const RGBA_RE = /rgba?\([^)]+\)/gi;
-const BOX_SHADOW_RE = /(?:box-shadow:\s*)?((?:-?\d[\w\d\s\-.,/()#%]*)+)/;
-const NAMED_RULE_RE = /\*\*(The [^*]+?Rule)\.\*\*\s*(.+)/;
 
 // ---------- Section splitting ----------
 
@@ -550,36 +547,6 @@ function detectFormat(v) {
   return 'unknown';
 }
 
-function scanInlineColors(lines) {
-  const out = [];
-  for (const line of lines) {
-    if (!/^\s*[-*]\s/.test(line)) continue;
-    const trimmed = line.replace(/^\s*[-*]\s+/, '');
-    const color = parseColorBullet(trimmed);
-    if (color) out.push(color);
-  }
-  return out;
-}
-
-function parseStitchInlineGroups(lines) {
-  // Stitch writes: `*   **Primary (`#00478d` to `#005eb8`):** Use for "..."`
-  // Each bullet IS its own role. Group them under the spoken role name.
-  const out = [];
-  for (const line of lines) {
-    if (!/^\s*[-*]\s/.test(line)) continue;
-    const trimmed = line.replace(/^\s*[-*]\s+/, '').trim();
-    const m = trimmed.match(
-      /^\*\*([A-Z][a-zA-Z]+)\s*\(([^)]+)\):\*\*\s*(.*)$/
-    );
-    if (m) {
-      const role = m[1];
-      const color = buildColor(role, m[2], m[3]);
-      out.push({ role, colors: [color] });
-    }
-  }
-  return out;
-}
-
 function extractTypography(section) {
   if (!section) return null;
   const text = section.lines.join('\n');

.claude/skills/impeccable/scripts/lib/staleness.mjs

@@ -488,41 +488,46 @@ export function describeWorkspaceContext(candidates = []) {
 // ─── Tier 1 orchestration ──────────────────────────────────────────────────
 
 /**
- * Everything a boot can afford. `ctx` is the loadContext result; `extras`
- * carries values the caller already computed so nothing is recomputed here.
+ * Everything a boot can afford, grouped by artifact so deeper reports can
+ * interleave their own checks without rebuilding this policy. `ctx` is the
+ * loadContext result; `extras` carries values the caller already computed so
+ * nothing is recomputed here.
  */
-export function collectBootFindings(ctx, extras = {}) {
-  if (!ctx) return [];
+export function collectBootFindingGroups(ctx, extras = {}) {
+  if (!ctx) return {};
   const projectRoot = ctx.projectRoot || process.cwd();
-  const absProductPath = extras.absProductPath || null;
   const absDesignPath = extras.absDesignPath || null;
 
-  return [
-    ...checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
+  return {
+    product: checkProduct(ctx.product, ctx.productPath || 'PRODUCT.md'),
     // Only checked once a PRODUCT.md exists. Without one the boot already
     // emits NO_PRODUCT_MD and routes into init, which asks for the platform
     // directly; a second signal saying the same thing is noise.
-    ...(ctx.product
+    nativePlatform: ctx.product
       ? checkNativePlatformEvidence({
           projectRoot,
           platform: ctx.platform,
           product: ctx.product,
           productPath: ctx.productPath,
         })
-      : []),
-    ...checkDesignSidecar({
+      : [],
+    designSidecar: checkDesignSidecar({
       designPath: absDesignPath,
       sidecarCandidates: extras.sidecarCandidates || [],
       projectRoot,
     }),
-    ...checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
-    ...checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
-    ...checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
-    ...(extras.projectRootPatterns
+    config: checkConfig({ projectRoot, repoRoot: ctx.repoRoot }),
+    buildPath: checkBuildPathUnset({ projectRoot, repoRoot: ctx.repoRoot, product: ctx.product }),
+    surfaceBriefs: checkSurfaceBriefs({ candidates: ctx.surfaceBriefCandidates, projectRoot }),
+    projectRoots: extras.projectRootPatterns
       ? checkProjectRoots({
           patterns: extras.projectRootPatterns,
           candidates: extras.targetCandidates || [],
         })
-      : []),
-  ];
+      : [],
+  };
+}
+
+export function collectBootFindings(ctx, extras = {}) {
+  return Object.values(collectBootFindingGroups(ctx, extras)).flat();
 }

.claude/skills/impeccable/scripts/lib/surface-briefs.mjs

@@ -8,6 +8,12 @@ export function getSurfaceBriefDir(projectRoot) {
   return path.join(projectRoot, '.impeccable', 'surfaces');
 }
 
+function normalizeRouteTarget(route) {
+  if (!route.startsWith('/') || route.includes('..')) return null;
+  const normalized = route.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
+  return `route:${normalized}`;
+}
+
 export function normalizeSurfaceTarget(target, { projectRoot = process.cwd() } = {}) {
   if (!target || typeof target !== 'string' || !target.trim()) return null;
   const trimmed = target.trim();
@@ -21,21 +27,13 @@ export function normalizeSurfaceTarget(target, { projectRoot = process.cwd() } =
       return null;
     }
   }
-  if (/^route:/i.test(trimmed)) {
-    const route = trimmed.slice(trimmed.indexOf(':') + 1).trim();
-    if (!route.startsWith('/') || route.includes('..')) return null;
-    const normalizedRoute = route.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
-    return `route:${normalizedRoute}`;
-  }
-  if (trimmed === '/') return 'route:/';
+  if (/^route:/i.test(trimmed)) return normalizeRouteTarget(trimmed.slice(trimmed.indexOf(':') + 1).trim());
+  if (trimmed === '/') return normalizeRouteTarget(trimmed);
   if (trimmed.startsWith('/')) {
     const absolute = path.resolve(trimmed);
     const relativeToProject = path.relative(projectRoot, absolute);
     const isProjectFile = relativeToProject && !relativeToProject.startsWith('..') && !path.isAbsolute(relativeToProject);
-    if (!isProjectFile && !fs.existsSync(absolute) && !trimmed.includes('..')) {
-      const normalizedRoute = trimmed.split(/[?#]/, 1)[0].replace(/\/{2,}/g, '/').replace(/\/$/, '') || '/';
-      return `route:${normalizedRoute}`;
-    }
+    if (!isProjectFile && !fs.existsSync(absolute)) return normalizeRouteTarget(trimmed);
   }
   const abs = path.isAbsolute(trimmed) ? trimmed : path.resolve(projectRoot, trimmed);
   const rel = path.relative(projectRoot, abs);

.claude/skills/impeccable/scripts/live-browser.js

@@ -4902,6 +4902,13 @@
     saveSession();
   }
 
+  function completeParameterGenerationIfReady() {
+    if (expectedVariants <= 0 || arrivedVariants < expectedVariants) return;
+    if (parameterGenerationState === 'pending' || parameterGenerationState === 'loading') {
+      completeParameterPublication();
+    }
+  }
+
   function toggleTunePopover() {
     if (pendingApplyInFlight) { showManualApplyBusyToast(); return; }
     if (tuneOpen) { closeTunePopover(); return; }
@@ -5796,7 +5803,7 @@
         setLiveState('CYCLING');
         showOrUpdateCyclingBar();
         saveSession();
-        if (parameterGenerationState === 'loading') completeParameterPublication();
+        completeParameterGenerationIfReady();
         return;
       }
 
@@ -5884,7 +5891,7 @@
       refreshParamsPanel();
       positionBar();
       saveSession();
-      if (parameterGenerationState === 'loading') completeParameterPublication();
+      completeParameterGenerationIfReady();
       console.log('[impeccable] Mounted ' + arrivedVariants + ' ' + manifest.framework + ' component variants.');
     } catch (err) {
       console.error('[impeccable] Failed to mount component-preview variants:', err);
@@ -6329,7 +6336,7 @@
         refreshParamsPanel();
         positionBar();
         saveSession();
-        if (parameterGenerationState === 'loading') completeParameterPublication();
+        completeParameterGenerationIfReady();
         console.log('[impeccable] Injected ' + arrivedVariants + ' variants from source file.');
       })
       .catch(err => {
@@ -6836,6 +6843,7 @@
 
       const expected = parseInt(wrapper.dataset.impeccableVariantCount || '0');
       if (expected > 0) expectedVariants = expected;
+      completeParameterGenerationIfReady();
 
       if (arrivedVariants > 0) {
         setLiveState('CYCLING');

.claude/skills/impeccable/scripts/live-commit-manual-edits.mjs

@@ -944,8 +944,42 @@ export async function commitManualEdits({
     };
   }
 
+  const repairContext = {
+    batch,
+    cwd,
+    pageUrl,
+    count,
+    provider,
+    env,
+    timeoutMs,
+    applyBatchToSource,
+    chatAvailable,
+    transactionId,
+  };
+
   const baseRollbackScope = collectApplyOwnedFiles(batch, cwd);
   const rollbackSnapshot = snapshotRollbackFiles(cwd, baseRollbackScope);
+  const failWithRollback = ({
+    scope = baseRollbackScope,
+    extraFiles = [],
+    failed,
+    files = [],
+    details = {},
+  }) => {
+    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, extraFiles, scope);
+    return {
+      applied: [],
+      failed,
+      files,
+      cleared: 0,
+      count,
+      pageUrl,
+      ...details,
+      rolledBackFiles: rollback.rolledBackFiles,
+      rollbackFailures: rollback.rollbackFailures,
+      ...countByPage(cwd),
+    };
+  };
   let result;
   try {
     result = repairOnly
@@ -965,42 +999,27 @@ export async function commitManualEdits({
           chatAvailable,
         });
   } catch (err) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, [], baseRollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
       failed: batch.entries.map((entry) => ({
         id: entry.id,
         reason: err.message || String(err),
         candidates: candidatesForEntry(batch, entry.id),
       })),
-      files: [],
-      cleared: 0,
-      count,
-      pageUrl,
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+    });
   }
 
   if (result.status === 'error') {
     const rollbackScope = collectApplyOwnedFiles(batch, cwd, result.files || []);
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
     const failed = normalizeFailedEntries(batch, result, result.message || 'AI copy edit failed');
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: failed.length > 0
         ? failed
         : verificationFailuresForEntries(batch, batch.entries, result.message || 'AI copy edit failed'),
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const reportedAppliedIds = uniqueStrings(result.appliedEntryIds || []);
@@ -1013,72 +1032,44 @@ export async function commitManualEdits({
   const conflictingAppliedIds = reportedAppliedIds.filter((id) => failedIds.has(id));
 
   if (conflictingAppliedIds.length > 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
     const conflictingEntries = batch.entries.filter((entry) => conflictingAppliedIds.includes(entry.id));
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: [
         ...verificationFailuresForEntries(batch, conflictingEntries, 'conflicting_apply_result'),
         ...aiFailed.filter((item) => !conflictingAppliedIds.includes(item.id)),
       ],
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const unreportedFiles = unreportedChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
   if (unreportedFiles.length > 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], [...rollbackScope, ...unreportedFiles]);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: [...rollbackScope, ...unreportedFiles],
+      extraFiles: result.files || [],
       failed: verificationFailuresForEntries(batch, batch.entries, 'unreported_source_changes', { files: unreportedFiles }),
       files: result.files || [],
-      unreportedFiles,
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { unreportedFiles, notes: result.notes || [] },
+    });
   }
 
   if (result.status === 'done' && reportedAppliedIds.length === 0) {
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: verificationFailuresForEntries(batch, batch.entries, 'missing_applied_entry_ids'),
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      notes: result.notes || [],
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   const reportedAppliedEntries = batch.entries.filter((entry) => reportedAppliedIds.includes(entry.id));
   if (reportedAppliedIds.length > 0 && reportedFiles.length === 0) {
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: reportedAppliedIds,
       files: result.files || [],
       failed: aiFailed,
@@ -1089,21 +1080,10 @@ export async function commitManualEdits({
     });
   }
 
-  const verifiedAppliedIds = [];
-  const verificationFailed = [];
-  for (const entry of reportedAppliedEntries) {
-    const failures = verifyAppliedEntry({ batch, entry, reportedFiles, cwd });
-    if (failures.length === 0) {
-      verifiedAppliedIds.push(entry.id);
-    } else {
-      verificationFailed.push({
-        id: entry.id,
-        reason: 'source_verification_failed',
-        failures,
-        candidates: candidatesForEntry(batch, entry.id),
-      });
-    }
-  }
+  const {
+    verifiedIds: verifiedAppliedIds,
+    failed: verificationFailed,
+  } = verifyEntriesAfterRepair({ batch, appliedEntryIds: reportedAppliedIds, files: reportedFiles, cwd });
   const unreportedEntries = result.status === 'done' || result.status === 'partial'
     ? batch.entries.filter((entry) => !reportedAppliedIds.includes(entry.id) && !aiFailed.some((item) => item.id === entry.id))
     : [];
@@ -1133,37 +1113,22 @@ export async function commitManualEdits({
         reason: 'rolled_back_due_to_failed_entry_source_changed',
         candidates: candidatesForEntry(batch, entry.id),
       }));
-    const rollback = rollbackChangedFiles(cwd, rollbackSnapshot, result.files || [], rollbackScope);
-    return {
-      applied: [],
+    return failWithRollback({
+      scope: rollbackScope,
+      extraFiles: result.files || [],
       failed: [
         ...leakedUnapplied,
         ...failed.filter((item) => !leakedIds.has(item.id)),
         ...rolledBackVerified,
       ],
       files: result.files || [],
-      cleared: 0,
-      count,
-      pageUrl,
-      rolledBackFiles: rollback.rolledBackFiles,
-      rollbackFailures: rollback.rollbackFailures,
-      notes: result.notes || [],
-      ...countByPage(cwd),
-    };
+      details: { notes: result.notes || [] },
+    });
   }
 
   if (verificationFailed.length > 0) {
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: reportedAppliedIds,
       files: result.files || [],
       failed: nonRepairFailed,
@@ -1180,16 +1145,7 @@ export async function commitManualEdits({
       ? reportedAppliedEntries.filter((entry) => verifiedAppliedIds.includes(entry.id))
       : batch.entries;
     return repairPostApplyValidation({
-      batch,
-      cwd,
-      pageUrl,
-      count,
-      provider,
-      env,
-      timeoutMs,
-      applyBatchToSource,
-      chatAvailable,
-      transactionId,
+      ...repairContext,
       appliedEntryIds: verifiedAppliedIds.length > 0
         ? verifiedAppliedIds
         : postCheckEntries.map((entry) => entry.id).filter(Boolean),

.claude/skills/impeccable/scripts/live-poll.mjs

@@ -238,10 +238,9 @@ export async function completeAcceptHandling(event, base, token) {
     });
   } catch (err) {
     event._completionAck = { ok: false, error: err.message };
+    return event;
   }
-  if (!event._completionAck) {
-    event._completionAck = completionAckForAcceptResult(event.id, completionType, event._acceptResult);
-  }
+  event._completionAck = completionAckForAcceptResult(event.id, completionType, event._acceptResult);
   return event;
 }
 
@@ -269,9 +268,11 @@ export function printPollEvent(event) {
   // Situational plumbing rides with the event itself: `_instructions` is the
   // authoritative next step, with real ids and paths substituted, so the
   // reference doc can stay lean and can never drift from script behavior.
-  if (event && typeof event === 'object' && !event._instructions) {
+  // A wire-supplied value must never win over the locally generated one.
+  if (event && typeof event === 'object') {
     const instructions = instructionsForEvent(event, { scriptsPath: SELF_DIR });
     if (instructions) event._instructions = instructions;
+    else delete event._instructions;
   }
   console.log(JSON.stringify(event));
 }

.claude/skills/impeccable/scripts/live-server.mjs

@@ -181,8 +181,16 @@ function chatAgentLikelyActive() {
 // cap at 10 MB to guard against runaway writes from a misbehaving client.
 const MAX_ANNOTATION_BYTES = 10 * 1024 * 1024;
 
+const POLLER_OWNED_EVENT_FIELDS = ['_instructions', '_completionAck', '_acceptResult'];
+
+function stripPollerOwnedEventFields(event) {
+  if (!event || typeof event !== 'object') return;
+  for (const key of POLLER_OWNED_EVENT_FIELDS) delete event[key];
+}
+
 function enqueueEvent(event) {
   if (!event) return;
+  stripPollerOwnedEventFields(event);
   // Dedupe by (session, type), except mount failures, which are per-variant:
   // variant 2 failing must not be swallowed because variant 1's failure is
   // still queued.
@@ -936,15 +944,23 @@ function createRequestHandler({ detectScript, liveScriptParts }) {
       const filePath = url.searchParams.get('path');
       if (!filePath || filePath.includes('..')) { res.writeHead(400); res.end('Bad path'); return; }
       const absPath = path.resolve(process.cwd(), filePath);
-      // Confine to the project root. A bare `startsWith(cwd)` string check lets a
-      // sibling dir whose name extends the root name (projeto -> projeto-backup)
-      // slip through; compare on the relative path instead (same pattern as
-      // sessionFileMetadataFromPollReply below). An empty rel means the request
-      // resolved to the root directory itself, which this file route never serves.
-      const rel = path.relative(process.cwd(), absPath);
+      let realRoot, realTarget;
+      try {
+        realRoot = fs.realpathSync(process.cwd());
+        realTarget = fs.realpathSync(absPath);
+      } catch {
+        res.writeHead(404); res.end('File not found'); return;
+      }
+      // Confine to the project root after symlink resolution. A bare
+      // `startsWith(cwd)` string check lets a sibling dir whose name extends the
+      // root name (projeto -> projeto-backup) slip through; compare on the
+      // relative path instead (same pattern as sessionFileMetadataFromPollReply
+      // below). An empty rel means the request resolved to the root directory
+      // itself, which this file route never serves.
+      const rel = path.relative(realRoot, realTarget);
       if (!rel || rel.startsWith('..') || path.isAbsolute(rel)) { res.writeHead(403); res.end('Forbidden'); return; }
       let content;
-      try { content = fs.readFileSync(absPath, 'utf-8'); }
+      try { content = fs.readFileSync(realTarget, 'utf-8'); }
       catch { res.writeHead(404); res.end('File not found'); return; }
       res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8' });
       res.end(content);
@@ -1026,6 +1042,7 @@ function createRequestHandler({ detectScript, liveScriptParts }) {
           res.end(JSON.stringify({ error }));
           return;
         }
+        stripPollerOwnedEventFields(msg);
         if (msg.type === 'agent_phase') {
           recordAgentPhase(msg.id, msg.phase, {
             ...(Number.isFinite(msg.durationMs) ? { durationMs: msg.durationMs } : {}),