* Give the verify-and-review pipeline a native leg
The build-verify-review loop assumed a browser end to end while the
comp side of the system was already platform-aware: new-work.md,
visualize.md, and the asset producer all comp a native app portrait at
its device viewport, and then the verification steps asked for desktop
and mobile browser screenshots of it. Concretely:
- new-work.md step 7 ordered detect.mjs on every hookless build with no
platform guard. routing.md declares the detector web-only and the
design hook skips native projects, so a native build was always
hookless and always ordered to run an HTML rule engine over
Swift/Kotlin/RN code. The playbook now guards it: web-only, and on
native the reviewer's floor check is the named slop gate.
- The inspection round and the SKILL.src.md batched-round principle
named desktop and mobile as the only viewports. Both now map per
platform: web keeps desktop and mobile; native inspects the shipped
device classes per OS, captured from the simulator or emulator.
- ios.md and android.md carried no verification guidance at all, so
nothing told a native run how to produce the screenshots the evidence
chain depends on. Each gains a Verifying the build section: simctl /
adb capture commands, dark-appearance and type-scale checks, and the
simulator-vs-hardware honesty line.
- The finish reviewer judged native builds blind: it never runs
context.mjs and its packet carried no platform guidance. On native
the packet now includes the platform reference path(s) and a
no-detector-ran line, and the reviewer's Input Contract says to judge
in the platform's conventions.
Assisted-by: Claude Code
* fix: address PR review bot findings
- greptile: carry the capture's device selector through the
state-changing verification commands (simctl appearance, adb uimode
and font_scale); unqualified forms fail with several targets attached
- Copilot: align new-work.md's cross-reference with the actual heading
(Verifying the build)
- Copilot: give the finish reviewer's Input Contract the native
filename example new-work.md establishes (phone.png / tablet.png,
suffixed per OS on adaptive)
Assisted-by: Claude Code
* fix: identify simulators by UDID, not display name
greptile follow-up: display names can collide across booted simulators,
so the capture and appearance commands now both key on the UDID from
simctl list devices booted.
Assisted-by: Claude Code
* Resolve {{scripts_path}} in the agent bodies Codex ships
Three code paths emit an agent body: the degraded fallback reference, the
.toml nested inside the skill for Codex, and the native agent file. Only the
nested .toml skipped placeholder substitution and rule-marker stripping, so
the codex and .agents dists shipped `node {{scripts_path}}/embed-prompt.mjs`
verbatim in the asset producer, and every caller had to substitute the token
itself at load time.
All three now render through renderAgentBody(), and the new regression test
asserts a runnable embed-prompt command on each emitted surface plus a
synthetic agent proving markers and placeholders resolve in the nested .toml.
Prepared by an AI agent (Claude Code) under pbakaus's instruction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Give the finish reviewer's screenshots one fixed address
The Input Contract asked for "desktop and mobile screenshot paths captured by
the parent" and named none, so each session invented a filename and the
verdict pass went looking for a recapture that was never written there. Two
reviewer passes burned on that in the eval runs.
The parent now captures and recaptures to .impeccable/review/desktop.png and
.impeccable/review/mobile.png, and the reviewer reads those two first,
treating a brief-named path as the fallback for a parent that wrote elsewhere.
Prepared by an AI agent (Claude Code) under pbakaus's instruction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Lead Setup with the base directory the runtime reports
The rendered claude and codex skills opened with
`node .claude/skills/impeccable/scripts/context.mjs`, a project-relative path
that resolves in this repo and in nothing a user installs: a personal or
plugin install puts the scripts outside the project entirely. The working form
was already in the text, parenthesized, after the one that fails.
Setup now leads with `node <skill-base-dir>/scripts/context.mjs` and says once
that the base directory resolves every scripts-path command in the skill and
its references, leaving the project-relative path as the fallback for runtimes
that report no base directory.
Prepared by an AI agent (Claude Code) under pbakaus's instruction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Answer the Copilot review: brittle model assertion, missing review dir
Assert that {{model}} resolved rather than that it resolved to "GPT", which
belongs to PROVIDER_PLACEHOLDERS and can change without touching what the test
guards. And have the parent create .impeccable/review/ when the harness does
not, so a fresh project's first capture has somewhere to land.
Prepared by an AI agent (Claude Code) under pbakaus's instruction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Make the review-screenshot contract directory-based, not web-viewport-named
Two amendments to the recapture contract from review feedback:
1. The canonical location is the directory .impeccable/review/, one file
per captured viewport; desktop.png and mobile.png are the web case,
not the contract. Baking web-viewport names into the reviewer's spec
would have hardened a web assumption into paths that a native
(ios/android/adaptive) build cannot honestly write.
2. Precedence restored to explicit-beats-convention: paths the calling
brief names are authoritative when the files exist; the canonical
directory is where the reviewer looks when the brief names none or a
named path is missing. This avoids stale canonical files from an
earlier run silently winning over fresh explicit paths. The observed
failure (the verdict round inventing a round-stamped filename) stays
fixed: recapture happens over the same files, and invented filenames
are still called out as pointing at nothing.
Assisted-by: Claude Code
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Probe attribution on Opus 5 showed the screenshot bound working (42
to 16) while the real burner ran free: five rounds of node -e
micro-edits, eight rebuilds, and inline defect hunts absorbed the
reviewer's and documenter's jobs until the turn cap killed the run
mid-hunt. The two-round ceiling now names scans, micro-edits, and
rebuilds; after the second round the build thread stops polishing and
ships the rest through the reviewer (one batched fix pass, one
rebuild, stop) and the documenter.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Opus 5 turned the iterate-with-screenshots-until-it-meets-the-bar
instruction into 42 screenshot trips and 150 tool calls per build,
about forty dollars of cache churn a page, before ever reaching the
reviewer. Verification now batches: one desktop-and-mobile round after
the full build, fixes applied together, one confirming round, ceiling
two. Craft-floor's checks share those renders instead of earning
separate trips; per-tweak iteration is live mode's channel.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
v4 changed PRODUCT.md's shape and retired the register axis, so an
upgraded project can carry answers nothing reads. Nothing measured that.
Two tiers, and the split is a performance contract:
- Boot (context.mjs, emitting CONTEXT_STALE) spends only what a boot
already spends: markdown already in memory, a bounded set of stats,
the small JSON files the boot reads anyway. No new directory walks.
One directive for the whole set, throttled to once a week per project
so a finding the user declined does not reappear tomorrow.
- doctor.mjs runs the deep pass on demand: git drift, ignore lists
validated against the live rule registry, hook script paths that stop
resolving, and the monorepo workspace sweep. --fix applies only the
migrations that carry no decision.
Findings are data, not prose, so the boot directive, the text report and
--json all render one set. Severity says what should happen: auto (fix on
the next write anyway), mention (state once), route (name the command
that owns the repair).
PRODUCT.md now carries a schema stamp so the checks stop reconstructing a
file's vintage from which sections it happens to have. Schema version,
not release version: a record written by 4.0.0 is not stale under 4.0.1.
DESIGN.md gets no stamp, because it follows the external design.md spec
that Stitch lints and every DESIGN.md signal is measurable without one.
The highest-value catch is a project that resolves to web while carrying
native build files, including a monorepo app inheriting a root record
that says web. That one costs output quality silently; nothing failed
before.
doctor follows the hooks/pin pattern rather than the Commands table, so
it stays out of the design menu and the count stays at 23.
Also corrects CLAUDE.md, which still documented the register axis,
reference/brand.md, reference/product.md, eleven deleted domain reference
files, and an extractRegister() whose only occurrence in the repo was
that sentence.
Prepared with AI assistance (Claude Code).
Co-Authored-By: Claude <noreply@anthropic.com>
Setup step 1 tells the agent to follow context.mjs's directives, and the
no-PRODUCT.md-with-existing-code directive explicitly permits a narrow
refinement to proceed on the incumbent implementation and offer init
afterward. Routing's "Otherwise" branch said missing PRODUCT.md routes
through init, with no carve-out, so the two instructions disagreed on the
same request and the agent could block work context.mjs had cleared.
Rule 3 now splits the way the directive does: a new surface or
replacement world goes through init then new-work, a narrow refinement
proceeds and offers init afterward. Explicit and implied commands were
never affected; they route one rule earlier, which is what skill-behavior
scenario 10 already covers.
Co-Authored-By: Claude <noreply@anthropic.com>
Bugbot flagged the "resume without rerunning context.mjs" instruction
after init. It is right, and the gap is wider than the platform half it
named: context.mjs has two output branches, and the no-PRODUCT.md branch
omits DESIGN.md, the native platform references, and the unrecognized
`## Platform` warning. Because the skill never reruns the script once
init writes PRODUCT.md, whatever that first run withheld is gone for the
whole session. A greenfield iOS project would be designed without
reference/ios.md ever loading, and a project carrying DESIGN.md without
PRODUCT.md never saw its own design system.
The two halves need different fixes. DESIGN.md is authority in its own
right and does not depend on PRODUCT.md existing, so context.mjs now
emits it on both branches. Platform is unknowable before PRODUCT.md
exists, so no change to the script can recover it; init.md, the one step
that learns the answer, now loads ios.md / android.md / both right after
recording a native platform, and SKILL.src.md says so where it tells the
agent not to rerun.
Verified end to end against a temp project on both branches.
Co-Authored-By: Claude <noreply@anthropic.com>
The detector-blind slop review existed because the AI-tell rules had been
stripped out of SKILL.md and nothing carried them. The floor is a better
home: it loads after concept ideation and immediately before editing UI,
which is the placement that made stripping them necessary in the first
place. Models tread lightly when a ban list is present during ideation;
by the time the floor loads, the direction is already committed.
- Rename build-floor.md to craft-floor.md and restore the absolute bans
(side-stripes, gradient text, glassmorphism, hero-metric, identical card
grids, eyebrow-on-every-section, numbered markers, text overflow), the
codex and gemini defect lists, and the reflexes no scanner catches.
Rule ids match the ones the ablation catalog already knows.
- Delete lib/slop-review.mjs and both injections. The Stop hook is now
purely a mechanical pass and stays silent with nothing to report.
- context.mjs replaces AI_SLOP_REVIEW_REQUIRED with the narrower
MANUAL_DETECTOR_REQUIRED, emitted only when a session has no hook at
all. A per-edit hook already covers the mechanical gap, and the floor
covers the judgment one either way.
Co-Authored-By: Claude <noreply@anthropic.com>
Snapshot the current app-server implementation, shared Live optimizations, generated harness output, and in-progress site work before restoring polling as the primary runtime path.
Prepared with Codex assistance under maintainer direction.
Four lines from the r10 dual consultation (codex gpt-5.6-sol + gemini
3.5-pro on the actual HTMLs) and the hero-probe micro-eval: the probe
isolated a first-viewport monoculture (same split template in every
sample, control and skill alike) and showed these lines break it while
codex's raw 15-liner alone does not. The incumbent sentence swap fixes
the r10 root cause both consultants independently identified.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Replaces the a14/a15 attempts (both deleted). Diagnosis: incentive
stacking; the placeholder-completion MUST plus the image tool turned
'bolder' into full-bleed photo insertion. Scope preservation is the
missing rule, not imagery policy.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
x02 a14 rerun: 3/3 samples still imported photos — the unscoped MUST in
the Persuade mode block overrode the existing-worlds principle. Scoping
keeps the greenfield ablation win, frees iteration asks.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
x02-tidewater-bolder eval: 3/3 skill-on samples imported photography into
a photo-free seed system (0% arena vs competitor, which amplified the
seed's own vocabulary instead). One sentence, shape-level, no examples.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Paul's a11 review: heroes are safe SaaS viewports, everything
predictable; mobile Operate ships dark despite a brief that specifies
outdoors-in-motion use. Decide-then-build now opens with three
one-line directions differing in concept (the instinctive pick that
any studio would reach for is the default wearing your name); the
Operate mode adds: the usage scene is part of the spec, the theme
follows the scene, not the category's habit.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Per Paul: rather than gating a second file, fold what made the craft
path superior into the file both models already read 21/21 through the
gate. new-work.md gains 'Decide, then build' (direction as one
confirmable paragraph; attended pauses, unattended records-and-goes;
codex.md mock flow when image generation exists) and 'Finish like a
studio' (inspect, honest critique, patch, detector). craft becomes a
deprecated alias like teach: invoking it forces attended checkpoints,
nothing else differs; the reference is a redirect stub. codex.md
retargeted. Existing-world feature builds remain governed by the core
floor (unmeasured path, noted in the plan doc).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Invocation A/B on Fable (a9 craft-path vs a9-direct plain): the plain
path scored 38% vs the competitor against the craft path's 50%, and
brief fidelity collapsed to 14% vs bare — the direct path drops asked-
for features that craft's direction step and engineering bar preserve.
Routing now sends any build request through the craft orchestration
unprompted (its gates pause only when a user can respond), and the
craft floor gains a brief-coverage recheck: every requirement the brief
names must exist on the page.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
a7 transcript evidence: 01-observability samples drew orange-honey and
green seeds, recited the color-strategy menu, and shipped dark
category-reflex palettes anyway; the model applied the subject's
workmanlike grammar to its own landing page. Two generic lines: the
mode belongs to the surface, not the subject (a landing page for a
dense tool is still Persuade; deciding a page can be plain because its
subject is workmanlike is the category error in reverse), and the
palette has exactly two legitimate sources (seed or the subject's
world; the category's habitual palette is neither).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The when-to-choose guidance sat inside the file that only loads after
the choice is made. SKILL.md's routing now says it: bare build requests
build directly through the gate and floor; craft is routed only when
named or when the user asks for a guided, checkpointed build. The
Commands row describes craft by its checkpoints. craft.md's intro just
describes the supervised flow it orchestrates.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Stubs removed per Paul (register: values remain harmless family hints;
nothing points at the files anymore). craft.md now opens by defining
itself against plain invocation: a bare build request goes straight
through the gate and the craft floor; craft is the supervised path with
guaranteed checkpoints and the mock pipeline. One shipping-discipline
line joins the core floor (real content, interaction states, respect
the build pipeline) so one-shots inherit the bar that previously lived
only in craft's Step 4.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Answering the obvious question the family-depth framing dodged: with
modes derived per task, files named for the old two-register taxonomy
had no architectural reason to exist. brand.md's surviving depth (lane
test + inverse test, reflex-reject lanes, color discipline, layout
moves, permissions) folds into new-work.md, where all of it belonged:
it is new-identity Persuade/Experience guidance. product.md's content
moves unchanged to operate.md, its true name. Both old files remain as
one-line redirect stubs because register: brand|product in existing
PRODUCT.md files and older links point there. All cross-references
retargeted (SKILL.md modes intro, context.mjs REGISTER hint, live.md,
typeset.md); 85 tests green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Architecture per Paul: impeccable is primarily a daily driver on
existing codebases; the always-loaded core should serve that 90% path,
not carry the full generative arsenal on every invocation. SKILL.md now
holds brief-wins, existing-worlds (the headline path), the four visitor
modes, the full craft floor, and a hard gate: new identity work
(greenfield, or a redesign discarding the current look) MUST read
reference/new-work.md before any design decision. That file carries the
generative playbook (seed, subject grounding, plan/self-check/signature,
hero-thesis, everything-bold, prove-don't-claim, color commitment,
calibration, persuade type/imagery). context.mjs enforces the gate
mechanically: NEW_WORK directive when no PRODUCT.md/DESIGN.md exists,
and the old mandatory register-file read is replaced by a REGISTER
family hint. No surfaces: map anywhere; mode is derived per task.
Gate compliance is measurable via skillEvidence.directSkillFileReads.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Field report: impeccable SaaS-ified a developer docs page; the Opus
galleries showed the same on an album page. Root cause: two registers
force every surface into persuade-or-operate grammar. The register
section now names the visitor's mode first (Persuade / Operate / Read /
Experience) with mode-borrowing called out as the canonical failure,
and PRODUCT.md's register field maps as family (brand = Persuade +
Experience, product = Operate + Read) for compatibility. Read mode:
comprehension deliverable, navigable structure, chrome out of the way.
Experience mode: the artifact leads at every screen size.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Paul's Opus gallery observation: every impeccable 05-experimental-album
generation reads decidedly SaaS while frontend-design's open with the
art itself, especially at narrow viewports. Cause: the brand register
prescribed stop-the-scroll/earn-the-click/convert for ALL brand
surfaces. Split the register's deliverable by surface: product/service
pages convert; cultural surfaces (album, portfolio, publication, body
of work) lead with the artifact, recede the interface, and treat
conversion grammar as a category error — the visitor meets the work in
the first viewport at every screen size.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Paul: everything should be bold, nothing bland; bold is neither
decoration nor clutter but commitment to the concept, whose form the
concept chooses (maximal or severely clean, drenched or monochrome,
piercing copy, the product demonstrating itself). Replaces the
'spend your boldness in one place' rule imported from frontend-design,
whose one-bold-element-on-a-quiet-page framing pulled pages toward the
tasteful softness the galleries showed losing. The signature becomes
where the concept peaks rather than the only place it lives.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Paul's spot-check of the Fable validation galleries: frontend-design's
lektor generations read vastly more distinctive and subject-faithful
despite losing the overall pairwise verdict on craft. The arena agrees
on the axis (distinctiveness 8-31 at n=5). Two additions to the core:
the opening viewport is a thesis (open with the most characteristic
thing in the subject's world, with a concrete memory test), and an
explicit polish-is-the-floor counterweight so the craft floor stops
reading as a mandate for quiet.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
gpt-5.6-sol evals: skill-on lost craft 0-25 to bare gpt-5.6; removing
the enumerated codex ban block recovered it to 4-16, confirming the
block's literal CSS patterns self-prime the defects they ban (the same
mechanism the v2.1 ablation sweep documented). Replaced with three
shape-level calibration lines: tracking floor (kept, it's a numeric
ceiling), elevation-declared-once + modest container radius, and
material honesty (real assets, surfaces not decoration, specific
claims). Detector rules continue to enforce the mechanical patterns.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pairwise evals on Fable one-shot (6-task regression set, opus-4-8 judge,
position-bias-cancelled): the hand-distilled ~55-line lean core beat the
heavy v4 core 66% overall / 67% craft head-to-head, and moved the
decisive win-rate vs frontend-design from 13% to 27% (40% with the
completion-time QA scan; craft went positive 6-5 for the first time).
18/18 lean samples ran context.mjs + palette.mjs vs a minority under the
heavy core: shorter instructions get followed. Context weight itself was
suppressing both compliance and boldness.
Structure: persona + brief-wins + existing-worlds + subject-grounding +
plan/self-check + boldness + prove-don't-claim + commit + calibration +
compressed craft floor + two-paragraph registers. Commands table kept;
the no-arg context-aware menu logic moved to reference/routing.md (read
on demand in the only case that is inherently interactive). Provider
blocks and rule anchors preserved.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Judge rationales across cand-v4a2 arenas: competitor wins by showing
the product working (mix panels, comparison tables, live demos) and by
signatures big enough to organize the page; our samples claim, decorate,
and sometimes stop at the hero.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Eval evidence (cand-v4a1-prose): palette.mjs handed a random violet seed
to the Polish-TV lektor brief and the model anchored on it, overriding
subject-grounding; craft/shape user gates can't fire in one-shot runs
and each model improvises around them. Seed is now a reflex-check that
yields to a subject-dictated palette; craft/shape gain an explicit
unattended mode (same bar, no waiting); init interview is skipped when
no user can respond.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Per Paul's guidance: (1) existing committed design systems are the
bread-and-butter case and get a first-class core rule (work inside the
world, no parallel colors/fonts/styles, no perf regressions); (2) a
redesign that discards the current look is new identity work and runs
the full concept/tokens/signature process instead of anchoring to the
incumbent skeleton (the lektor failure); (3) the reflex-reject font list
and physical-object font procedure are brand-register rules, moved out
of the universal Commit section — system stacks and workhorse UI faces
are legitimate, often correct, for product UI, stated positively in the
product register.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
One-shot evals on Fable 5 (impeccable-evals notes/fable-oneshot-craft-plan.md)
showed the reference-file architecture failing: models skip the register
reads, so most design guidance never reaches them, and skill-on collapses
toward bare-model output (0/9 pairwise wins vs frontend-design on r10).
SKILL.md is now self-contained for one-shot work: persona, the-brief-wins
rule, ground-it-in-the-subject, a plan/tokens/signature/self-check process
gate, commitment guidance, a compact inline craft floor, and distilled
brand/product registers. Reference files remain as sub-command flows and
optional depth. The enumerated absolute-bans list is retired from prose;
mechanical slop enforcement moves to the detector/hook.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Route native projects to native command variants for audit and adapt
Follow-up to #269. The web audit.md and adapt.md carried "translate this
yourself" Platform notes, so a native invocation paid for the full web
file (~1.8k / ~2.6k tokens, mostly inapplicable) and did error-prone
run-time translation. Authored with AI assistance (Claude Code) under
maintainer direction.
- New reference/audit.native.md and reference/adapt.native.md: authored
native content (VoiceOver/TalkBack, platform conformance, adaptivity
dimensions; phone-to-tablet, platform-to-platform, web-to-native
strategies). One variant per command covers ios, android, and
adaptive; per-OS specifics stay in the platform refs Setup loads
regardless.
- SKILL.src.md: Commands table lists the variants; Setup step 2 reads
the variant instead of the web file when the platform is native.
- audit.md / adapt.md: Platform sections replaced with a one-line
web-only guard pointing at the variant.
- animate.md / layout.md: Platform sections deleted; the Motion and
Layout sections of the already-loaded platform refs carry that
content. Web users now pay zero tokens for the platform axis in
these files.
- Skill-behavior scenario 15 pins the route-instead behavior (passes
live on claude-sonnet-4-6); CLAUDE.md documents the variant
convention.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Phrase command-reference routing as one rule, not rule-plus-exception
Copilot review catch: step 2 said "MUST read reference/<command>.md"
and then carved out the native variant, which invites loading both
files. Now a single rule: read the web reference or the table's native
variant, one file, not both. Scenario 15 re-verified live. Applied with
AI assistance (Claude Code) under maintainer direction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Anchor native runs in animate/layout, drop loaded-refs assumption
Review-thread fixes, applied with AI assistance (Claude Code) under
maintainer direction:
- Greptile: deleting the animate/layout Platform sections left native
runs alone with web tooling instructions (CSS keyframes, GSAP, Grid,
clamp()). Restore a one-line anchor in each pointing at the loaded
platform reference's Motion / Layout section (~20 tokens, not the old
restatements).
- Bugbot: audit.native.md and adapt.native.md asserted the platform
refs were "already loaded in Setup", but the command reference loads
at step 2, before step 5. Now they instruct: read the platform
reference first if Setup hasn't already.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Carry the native-variant rule into routing rules 2 and 3
Bugbot catch: Setup step 2 routed native projects to the variant, but
routing rules 2 and 3 (the operative text at command time) still said
to load the generic reference file. Both now reference the same
one-file variant rule. Applied with AI assistance (Claude Code) under
maintainer direction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Point animate/layout native anchors at the files, not "loaded" refs
Bugbot catch, same class as the variant wording fix: the anchor lines
said "the loaded platform reference" but command files load at step 2,
before the platform refs at step 5. Both anchors now name the files and
instruct reading them first if Setup hasn't already. Applied with AI
assistance (Claude Code) under maintainer direction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Add a platform axis (web / ios / android / adaptive) to the skill
Orthogonal to register: register decides whether design IS or SERVES the
product; platform decides the delivery target and which native conventions
apply. Set `## Platform` in PRODUCT.md; a missing field defaults to `web`,
so legacy projects are unaffected.
- extractPlatform() in skill/scripts/context.mjs (mirrors extractRegister);
the CLI appends a NEXT STEP directive to read the native reference(s).
`adaptive` (Flutter / RN / KMP shipping both iOS and Android) loads both
ios.md and android.md.
- New reference/ios.md (Apple HIG distilled) and reference/android.md
(Material 3 distilled); reference/web.md is a thin pointer. The native
refs frame register's role as narrow: platform conformance is the bar,
brand lives in the expressive layer the platform gives you, never by
breaking the rails.
- Setup step 5 loads the native reference(s) when platform is native. Live
mode and the detect CLI stay web-only, gated off ios/android/adaptive.
- init asks platform right after register; adapt/audit/animate/layout carry
short platform divergence notes; all secondary spots thread `adaptive`.
- a11y stays in audit.md (loading it at design time makes output timid), so
the native refs carry no Accessibility section; audit.md's Platform
section owns native a11y.
- Tests: extractPlatform unit coverage + skill-behavior scenario 10
(PRODUCT.md platform ios -> agent loads ios.md).
Source-first: only skill/, scripts/, tests/, CLAUDE.md, NOTICE.md, the
changelog and version are committed; the sync workflow regenerates the
provider trees and ./plugin on merge.
ios.md / android.md are distilled from the MIT-licensed
ehmo/platform-design-skills; attribution in NOTICE.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Address review: gate web tools on native platforms, drop version churn
Maintainer-review fixes applied with AI assistance (Claude Code), on top
of the rebased platform-axis commit:
- Design hook (post-edit and Cursor pre-edit) now resolves the project
platform via loadContext + extractPlatform and skips its web rule scan
for ios / android / adaptive projects, so React Native / Flutter code
never draws web-shaped findings (new hook-lib resolveProjectPlatform /
isNativePlatform helpers, covered by unit and subprocess tests).
- context.mjs CLI warns on an unrecognized ## Platform value (e.g. a
toolchain name like `flutter`) instead of silently defaulting to web;
extractRegister / extractPlatform now share extractSectionValue.
- Removed reference/web.md: nothing loaded it; CLAUDE.md carries the
"web has no extra rulebook" explanation.
- init.md: skip live-mode config (Step 6) for native platforms; note the
per-app PRODUCT.md pattern for repos shipping web + native.
- android.md: Material-everywhere apps that also ship on iPhone still
owe iOS OS guarantees (safe areas, Reduce Motion, edge-swipe back).
- ios.md: reworded a design-time line that framed Dynamic Type as an
accessibility check (a11y stays owned by audit.md).
- Renumbered the new skill-behavior scenario to 14 after main's 10-13;
updated CLAUDE.md scenario list; added android + unrecognized-value
CLI test cases.
- No version or changelog changes: versioning happens at release time.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tighten platform reference prose
Editorial pass on the platform-axis text, applied with AI assistance
(Claude Code) under maintainer direction:
- ios.md / android.md rewritten to house style: single-line paragraphs
(no hard wraps), one-sentence scope intro, deduplicated intro/slop-test,
register-compression down to two sentences. In-file attribution
paragraphs removed (NOTICE.md owns attribution); "read on top of the
register reference" cruft removed (SKILL step 5 and the context.mjs
directive already say it). Bans sections dropped: they restated the
rules above them; the two additive items (tab-bar overload,
hover-dependent affordances) folded into rules. ~40% smaller each.
- Sub-command Platform sections (adapt, audit, animate, layout), SKILL
step 5, init.md platform prose, and the context.mjs directive trimmed
the same way.
Build (prose validators, counts) and both test runners green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Treat an empty PRODUCT.md section as absent, not the next heading
Copilot review catch: extractSectionValue read the next `## ...` heading
as the section value when a field was left empty, which made the CLI
warn "value `## Product Purpose` is not recognized". Stop at the next
heading and return null instead. Regression tests for extractPlatform,
extractRegister, and the CLI warning path. Applied with AI assistance
(Claude Code) under maintainer direction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Only read a token list of both native targets as adaptive
Bugbot catch: after the exact platform tokens failed, any Platform line
containing the words ios and android was classified adaptive, so
negated or explanatory prose ("web only, not ios or android") silently
loaded both native refs and skipped the hook, with no warning. The
combo parse now accepts only list separators and the two platform
words; anything else falls through to the CLI's unrecognized-value
WARNING. Regression tests added. Applied with AI assistance (Claude
Code) under maintainer direction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Paul Bakaus <paul.bakaus@gmail.com>
* Don't force init on scoped commands when PRODUCT.md is missing
Setup step 1 told the agent: "If it reports NO_PRODUCT_MD, stop and
follow reference/init.md before doing anything else." For a project with
no PRODUCT.md, that turned every scoped request (polish, critique, audit,
layout, ...) into a full from-scratch init detour. The user asks to
polish one button and the skill instead starts writing PRODUCT.md from
the beginning. Faced with that gate, agents also frequently abandon the
command and do an ad-hoc pass without loading the command reference.
Make the gate command-aware. A missing PRODUCT.md still routes into init
for the from-scratch build flows where captured product context is the
point (init, craft, shape). For any other command, a scoped request
against existing code, the code is the context: proceed with the
requested command, infer the register from the surface in focus, and
offer /impeccable init once as a suggestion rather than a blocker.
- skill/SKILL.src.md: rewrite the step 1 NO_PRODUCT_MD rule; reconcile
the no-argument routing rule so it leads the menu with init instead of
silently jumping into it; extend the craft init-then-resume footnote to
cover shape, now also a from-scratch flow.
- skill/scripts/context.mjs: soften the NO_PRODUCT_MD message to defer to
the step 1 rule instead of "Stop the current task"; refresh the stale
file-level JSDoc that still described the old empty-stdout signal.
- tests/skill-behavior/scenarios.test.mjs: add scenario 10 (scoped
command + no PRODUCT.md proceeds without forcing init) and scenario 11
(shape + no PRODUCT.md still diverts into init). Scenario 1 (craft
diverts) stays green and pins the build path.
Source-only per repo convention; provider and plugin copies are
regenerated by the maintainer's build:skills sync.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Fix missing-context routing for build intent
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Paul Bakaus <paul.bakaus@gmail.com>
The skill declared only `Bash(npx impeccable *)` in allowed-tools, but Setup and the no-arg menu shell out to `node {{scripts_path}}/*.mjs`. Under a default-deny Claude Code allowlist those calls are blocked, so Setup fails on context.mjs.
Add a provider-aware `Bash(node {{scripts_path}}/*)` entry and resolve {{scripts_path}} in the frontmatter (the build previously substituted it only in the body). Provider-aware rather than the hardcoded `.claude/...` path the issue suggested, since five providers honor allowed-tools with different script dirs.
Context files (PRODUCT.md / DESIGN.md) resolve child-first then fall back to the repo root, and /impeccable live lets the user pick a child app in a monorepo. Single-app behavior is unchanged. Closes#202. Co-Authored-By: abdulwahabone
* docs: add PRD for design detector hook integration
Plans a PostToolUse hook for Claude Code and Codex that runs the
existing design detector after every relevant file write and feeds
findings back to the agent as advisory system-reminder context. No
implementation in this commit; covers UX, technical design, build
pipeline changes, distribution, coverage tradeoffs, and rollout.
Co-authored-by: Cursor <cursoragent@cursor.com>
* docs: revise hook PRD with best-practices review
Folds in the P0/P1/P2 findings from an online best-practices critique
against the official Claude Code and Codex hook references plus 10+
2026 community guides and similar prior-art tools (claw-hooks,
claude-code-hooks-mastery).
Key changes:
- Exec form everywhere (Codex snippet was shell form), with Windows
rationale.
- Default timeout dropped from 10s to 5s.
- Re-entrancy guard (CLAUDE_HOOK_DEPTH) and per-file edit counter.
- Session-scoped finding dedup promoted from open question to v1.
- Per-language inline-ignore syntax map (HTML/JSX/CSS/JS).
- Hard-skip rules for sensitive paths and generated/lock files.
- Honest framing about Claude Code lacking per-plugin hook disable.
- Honest framing about Bash-written files being invisible in v1.
- Codex Windows-not-supported call-out, feature flag note, trust ceremony detail.
- Optional NDJSON audit log via IMPECCABLE_HOOK_LOG.
- Findings cap lowered 8 → 5 with attention-budget rationale.
- Versioned envelope ([impeccable@1]) on rendered template.
- Expanded test plan, decision log, and stdin payload appendix.
Co-authored-by: Cursor <cursoragent@cursor.com>
* feat(hooks): ship the design detector hook for Claude Code and Codex
Implements docs/hooks-prd.md: a PostToolUse hook that runs the
impeccable design detector after every Edit/Write/MultiEdit on a UI
file and pushes findings into the agent's next-turn context as a
short system reminder. Silent on clean files. Never blocks an edit.
Why this matters: today, design slop (side-tab borders, gradient
text, purple/cyan palettes, bounce easing, etc.) only gets caught
when a human notices or someone explicitly runs /impeccable audit.
The hook closes the loop at the moment slop is written.
What ships in v1
- skill/scripts/hook.mjs: PostToolUse entry. Reads stdin, runs the
detector in-process (no `npx impeccable` cold start), emits
hookSpecificOutput.additionalContext when fresh findings exist.
- skill/scripts/hook-lib.mjs: extracted helpers (config, cache,
filter, render, audit log, runHook orchestrator). 100% unit-testable.
- skill/scripts/hook-session-start.mjs: SessionStart greeting,
gated by a project-scannable probe + 30-day throttle.
- skill/scripts/hook-admin.mjs: backs /impeccable hooks
on/off/status/ignore-rule/ignore-file/reset.
Hardening built in
- Re-entrancy guard (IMPECCABLE_HOOK_DEPTH) so the hook can never
recursively spawn itself.
- Hard-skip regexes for sensitive paths (.env, .pem, id_rsa,
secrets, credentials, .git) and generated/lock/build output. These
fire before the file is even read; cannot be turned off via config.
- Path-traversal check on the inbound file_path.
- Session-scoped dedup keyed by (session, file, rule, line) so the
same finding never lands in context twice. Prevents the ~12.5K
wasted tokens per chatty session called out in the PRD.
- Per-(session, file) edit counter with a one-shot suppression
notice on the 7th edit, silent after.
- Fail-open contract: every error path returns exit 0 with no
stdout. Optional NDJSON audit log via IMPECCABLE_HOOK_LOG.
Three kill switches (precedence high to low):
1. IMPECCABLE_HOOK_DISABLED env var (1/true/yes/on, case-insensitive)
2. .impeccable/hook.json `enabled: false`
3. /impeccable hooks off slash command (writes the JSON)
Inline ignores are language-aware. `// impeccable: ignore <rule>` for
JS/TS, `<!-- impeccable: ignore <rule> -->` for HTML/Vue/Svelte/Astro,
`{/* impeccable: ignore <rule> */}` for JSX/TSX, `/* impeccable:
ignore <rule> */` for CSS. `*` matches any rule. Directive applies
to the next non-blank line. Same shape as ESLint, Stylelint, Biome.
Build pipeline
- scripts/lib/transformers/hooks.js: per-provider hooks.json
builders, plus the slim .codex-plugin/plugin.json manifest.
- providers.js: emitHooks: 'claude' for claude-code, emitHooks:
'codex' for codex and agents. Codex also emits emitCodexPlugin.
- factory.js: emits hooks/hooks.json next to the skills tree.
- build.js: syncs hooks/ into harness roots and into the slim
plugin/ subtree; writes .codex-plugin/plugin.json. Build is
idempotent (verified: 98 staged files unchanged across two runs).
Claude Code wiring uses exec form (command + args) and the
${CLAUDE_PLUGIN_ROOT} placeholder. Matcher: Edit|Write|MultiEdit.
`if:` glob filters to UI extensions before spawning Node. PostToolUse
timeout 5s, SessionStart timeout 3s.
Codex wiring uses ${PLUGIN_ROOT} (Codex's native placeholder),
matcher Edit|Write|apply_patch, no `if:` analog (the script does the
extension filter). macOS and Linux only; hooks are disabled on
Windows in current Codex builds. The trust ceremony and feature flag
are documented in README.md.
Routing
- /impeccable hooks lives outside the 23-command router table on
purpose: it is plumbing, not a design skill. The hidden
routing slot is added to SKILL.md alongside pin/unpin so the LLM
knows to dispatch it. The 23-command count and all stale-count
validators remain happy.
Tests
- tests/hook.test.mjs: 38 unit tests covering env parsing, config
load + defaults + malformed, cache round-trip + GC,
ignoreRules/minSeverity/inline ignores (all four languages),
globbing with **/*/{a,b}, render template with cap + clamp + 0-line
prefix drop, audit log NDJSON, payload event-name parameterization,
re-entrancy, kill switches, sensitive-path + generated-path +
traversal skips, allowlist filter, config ignoreFiles, edit
counter cycle including the 7th-edit notice, MultiEdit and
apply_patch payload shapes, detector throw swallow, malformed
stdin, missing file race.
- tests/hook-build.test.mjs: 18 integration tests covering hook
manifest shape (matcher, timeouts, exec form, if: glob, placeholders),
Codex differences (${PLUGIN_ROOT}, no if:, no SessionStart),
Codex plugin manifest (no inline hooks field to avoid the
duplicate-file error), routing across the hooksJsonFor table, and
presence of all three committed artifacts plus the bundled detector
the runtime relative-import path depends on.
Full suite: 175 bun tests + 186 node tests, all green.
Docs
- README.md: new "Design hook" section explaining default behavior,
per-project / global / inline disable paths, the JSON schema knobs,
the audit log debug flag, and the slop / a11y coverage split.
- HARNESSES.md: flips the `hooks` row for Codex from No -> Yes
(Claude was already Yes), adds a per-harness hook-surface table
with the manifest location and matcher each provider uses.
Open questions from the PRD intentionally deferred to v2: Bash-write
blind spot, effort-aware suppression, Stop-hook session summary,
per-rule severity, async hook mode. None block v1.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Fix Codex hook scanning: apply_patch paths and co-located stylesheets
Parse file targets from Codex apply_patch command bodies, co-scan imported
and sibling CSS when UI components are edited, drop the git-sweep PostToolUse
group, and align Codex SessionStart manifest and trust docs with the official
hooks spec.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Gitignore hook session cache and drop local test HTML
Hook dedup/throttle state in .impeccable/hook.cache.json is per-project
runtime data like other .impeccable/ sidecars. Remove an untracked
bad-nested-flexbox scratch page from site/public/.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Fix Claude Code hook: drop Edit-only if filter so Write/MultiEdit fire
Claude's if permission rule binds to one tool name, so Edit(*.{…}) never
spawned the hook on Write or MultiEdit despite the matcher listing them.
Extension filtering now lives in hook-lib on both Claude and Codex.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Surface Cursor design findings via stop-hook followup
Replace dropped postToolUse additional_context with afterFileEdit recording
and a one-shot stop followup_message so anti-pattern nudges reach the agent.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Fix design hook packaging and scans
* Fix Cursor hook pending bucket fallback
* Fix Sass hook scan coverage
* Fix Cursor hook review findings
* Fix session start dead hook normalization
* Fix hook config and relative scan paths
* Remove SessionStart design hook
* Remove redundant afterFileEdit normalization
* Fix Cursor suppression and module style scans
* Fix sensitive path hook filter
* Fix disabled Cursor stop hook emission
* Refresh hook harness artifacts
* Fix Cursor hook manifest install
* Add hook ignore-value support
* Ignore hook runtime files locally
* Fix Codex plugin hook packaging
* fix: address PR review bot findings
Block numeric hook depth counters from re-entering.
Avoid following stylesheet imports from traversal-looking hook targets.
* fix: gate ignore-value suggestions by supported rules
Only render exact ignore-value commands when the same finding can be suppressed by ignoreValues.
* Package Codex plugin as hook-only
* Remove Codex plugin packaging
* Recover hook install probe plumbing
* Remove Codex hook packaging follow-up doc
* Remove extra hook docs and skill wording changes
* Install real design hooks via skills CLI
* Add provider hook smoke runner
* Fix Cursor hook delivery with preToolUse gate
* Simplify Cursor hook install to preToolUse
* Clarify confirmed hook exceptions
* Persist hook ignores in shared config
* Guard font hook exceptions
* Fix hook install after main rebase
* Fix hook scan target handling
* fix: address hook review findings
* Address hook review feedback
* Stabilize DeepSeek insert live fixture
* Fix Cursor hook Python shell write bypass
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Copy guidance (em-dash bans, buzzword bans, button-label / link-text
phrasing, aphoristic-cadence) doesn't belong in the main design skill.
It's not design-specific — the skill is trying to do too much. The six
rules being dropped (every-word-earns, no-em-dashes, no-aphoristic-cadence,
no-buzzwords, button-verb-object, link-standalone) are now better served
by:
- The impeccable engine's antipattern detectors (em-dash-overuse,
marketing-buzzword, aphoristic-cadence, copy-slop) for linting at scan
time.
- The /clarify subcommand for surfacing the same checks when reviewing
copy specifically.
The em-dash ban for the SKILL prose itself still lives in STYLE.md and the
build-time prose validator — that's separate from the skill's guidance to
agents.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The v2.1 ablation sweep (n=10 × 4 brand niches × 3 providers, anchored to
commit 54c3a502, ~544 cells) confirmed these four rules carry no weight in
the skill:
- skill-typo-no-all-caps-body — duplicate of brand-ban-all-caps-body; brand
version is more specific (reserves caps for labels + headings)
- skill-typo-codex-hero-ceiling-repeat — the codex-block restatement of
skill-typo-hero-ceiling didn't add reinforcement on top of the universal
rule
- skill-typo-scale-ratio — duplicate of brand-typo-modular-scale; same
signal, brand version carries the clamp() / fluid implementation detail
- skill-typo-font-count — models don't reach for ≥4 font families in any
niche we test, so the rule has no measurable effect
Each deletion is the Agent A / B / C / D Phase-2 audit recommendation;
none of the four ever validated under either prose state.
Adds EMPIRICAL_VALIDATION.md naming the seven cross-provider winners as the
trustworthy core, and documents the systemic findings (self-priming, detector
saturation, vocabulary anchoring) so future skill edits can avoid the same
traps.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mirrors the 5 prose changes in skill/SKILL.src.md + skill/reference/brand.md
out to every harness directory (`.claude`, `.gemini`, `.cursor`, `.codex`,
`.agents`, etc.) so the staged skill that workers / agents read matches the
source. Auto-generated by `bun run build:skills`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase-2 ablation audit caught these rules causing the exact behavior they
ban via the literal examples in their own prose. Verified: OpenAI samples
under skill-on produced "fake theater", "vendor theater", "heatmap theater"
as verbatim copies of the 'X theater' example. Same pattern for the
restrained-on-cream example, the aphoristic-cadence template, and the
"reserve uppercase for…" enumeration.
- skill-ban-codex-x-theater: drop the 3 syntactic templates + 3 example
phrases ("Productivity theater" etc.)
- brand-imagery-required: drop the niche enumeration that cued
"imagery not required elsewhere"
- skill-typo-no-all-caps-body: drop the "Reserve uppercase for labels /
eyebrows / badges" enumeration that primed uppercase usage
- brand-color-no-converge: drop the "restrained-on-cream" example that
was priming cream-heavy palettes
- skill-copy-no-aphoristic-cadence: drop the literal cadence template
("serious statement, then punchy short negation") that named the
rhythm it bans
Ablation re-run pending in impeccable-evals to measure impact.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Rework context-signals' detect target after review: a URL meant a costly
Puppeteer render (and a probed port might not even be this project), and the
index.html-or-bail fallback failed most real apps (no root index.html).
New priority: (1) the scannable markup/style files in the dirty git tree
(what the user is working on, small and local); (2) a local source dir
(src / app / components / pages / public — the detector walks these and skips
node_modules / dist / build); (3) a root index.html, else the project root as
a last resort when there's code. Emits `scan.targets` (a list) + `scan.via`.
Never a URL.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Paul BakausandGitHub
Abdul WahabandGitHub
