Trust the OS certificate store for engine HTTPS requests (#757)

The engine verified TLS against the Mozilla roots bundled through
webpki-roots only, so behind a TLS-inspecting proxy (Aikido, Zscaler,
Netskope) whose root lives in the OS trust store, `impeccable update`
and `install` failed with `invalid peer certificate: UnknownIssuer`
while curl and npm on the same machine succeeded.

crates/context/src/http.rs builds one rustls ClientConfig per process:
the OS trust store (rustls-native-certs: Keychain, Windows store, the
OpenSSL paths on Linux) merged with the bundled roots. A union, not a
replacement, so a container without ca-certificates or a store that
fails to load still verifies exactly as before. SSL_CERT_FILE and
SSL_CERT_DIR replace the OS store the way they do for OpenSSL and curl.
Every HTTPS call site (bundle and signature downloads, /api/version,
/api/commands, the roll API, image generation) builds its agent from
this module; the plain-HTTP live-server calls on localhost are
untouched.

Verified against a local HTTPS server signed by a throwaway CA: trusted
through SSL_CERT_FILE the update check reaches it; without it the same
server is rejected as UnknownIssuer; with SSL_CERT_FILE pointing at that
CA or at a missing file, impeccable.style still verifies through the
bundled roots. cargo test --workspace and the oracle corpus (832) pass.

Written with AI assistance (Claude Code).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

.agent/skills/impeccable/SKILL.md

@@ -17,9 +17,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.agent/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.agent/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.agent/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.agent/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.agent/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.agent/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.agent/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.agent/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.agent/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.agents/skills/impeccable/SKILL.md

@@ -14,9 +14,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.agents/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.agents/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.agents/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.agents/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.agents/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.agents/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.agents/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.agents/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.agents/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 Codex: run live helper commands, the app dev server, and any dependency-installing setup with `sandbox_permissions: "require_escalated"` from the start; live mode depends on localhost and package-manager network access that the sandbox blocks.
 
 ## The contract (read once)

.claude/skills/impeccable/SKILL.md

@@ -16,9 +16,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.claude/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.claude/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.claude/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.claude/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.claude/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.claude/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.claude/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.claude/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.claude/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.cursor/skills/impeccable/SKILL.md

@@ -14,9 +14,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.cursor/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.cursor/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.cursor/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.cursor/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.cursor/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.cursor/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.cursor/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.cursor/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.cursor/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.dsh/skills/impeccable/SKILL.md

@@ -15,9 +15,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.dsh/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.dsh/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.dsh/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.dsh/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.dsh/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.dsh/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.dsh/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.dsh/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.dsh/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.gemini/skills/impeccable/SKILL.md

@@ -13,9 +13,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.gemini/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.gemini/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.gemini/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.gemini/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.gemini/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.gemini/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.gemini/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.gemini/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.gemini/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.github/skills/impeccable/SKILL.md

@@ -16,9 +16,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.github/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.github/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.github/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.github/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.github/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.github/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.github/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.github/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.github/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.github/workflows/ci.yml

@@ -11,6 +11,9 @@ on:
   schedule:
     - cron: '0 7 * * *'
 
+permissions:
+  contents: read
+
 concurrency:
   # Scheduled runs get their own group: the 07:00 UTC nightly and a push to
   # main share github.ref, and cancel-in-progress would let them kill each
@@ -34,12 +37,12 @@ jobs:
       live_svelte_adapter_deepseek: ${{ steps.plan.outputs.live_svelte_adapter_deepseek }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 0
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
@@ -59,15 +62,15 @@ jobs:
         node-version: [22.18.0, 24]
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: ${{ matrix.node-version }}
 
       - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
@@ -102,7 +105,7 @@ jobs:
         if: needs.changes.outputs.detector == 'true'
         run: rustup show && rustup target add wasm32-unknown-unknown
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
         if: needs.changes.outputs.detector == 'true'
 
       - name: Install wasm-pack
@@ -140,7 +143,7 @@ jobs:
         run: git diff --exit-code -- .agents .claude .cursor .dsh .gemini .github/skills plugin extension/detector
 
       - name: Upload build artifacts
-        uses: actions/upload-artifact@v7
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: impeccable-build-node-${{ matrix.node-version }}
           # Ship the packaged zips, not the unpacked Firefox staging tree.
@@ -157,14 +160,14 @@ jobs:
     if: needs.changes.outputs.rust == 'true'
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       # rust-toolchain.toml names the channel; `rustup show` installs it.
       # Never override the toolchain here.
       - name: Install the pinned toolchain
         run: rustup show
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
 
       - name: Build
         run: cargo build --workspace --all-targets
@@ -181,10 +184,10 @@ jobs:
     if: needs.changes.outputs.rust == 'true'
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
       - name: Install the pinned toolchain
         run: rustup show
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
       - run: cargo build --workspace --all-targets
       - run: cargo test --workspace --no-fail-fast
 
@@ -194,9 +197,9 @@ jobs:
     if: needs.changes.outputs.core == 'true'
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
       - name: Exercise Windows launcher downloads and verification
@@ -213,15 +216,15 @@ jobs:
     if: needs.changes.outputs.oracle == 'true' || needs.changes.outputs.rust == 'true'
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
@@ -231,7 +234,7 @@ jobs:
       - name: Install the pinned toolchain
         run: rustup show
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
 
       - name: Build the engine from source
         run: cargo build --release -p impeccable
@@ -255,10 +258,10 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
@@ -291,15 +294,15 @@ jobs:
     if: needs.changes.outputs.cli_remote_e2e == 'true'
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
@@ -311,7 +314,7 @@ jobs:
       - name: Install the pinned toolchain
         run: rustup show
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
 
       - name: Build the engine
         run: cargo build --release -p impeccable
@@ -337,20 +340,20 @@ jobs:
             fixtures: vite8-react-css-modules,vite8-react-insert,vite8-react-plain
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
       - name: Cache fixture npm downloads
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.npm
           key: ${{ runner.os }}-fixture-npm-${{ hashFiles('tests/framework-fixtures/**/files/package.json') }}
@@ -358,7 +361,7 @@ jobs:
             ${{ runner.os }}-fixture-npm-
 
       - name: Cache Playwright Chromium
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.cache/ms-playwright
           key: ${{ runner.os }}-playwright-chromium-${{ hashFiles('package.json', 'bun.lock') }}
@@ -376,7 +379,7 @@ jobs:
       - name: Install the pinned toolchain
         run: rustup show
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
 
       - name: Build the engine
         run: cargo build --release -p impeccable
@@ -393,7 +396,7 @@ jobs:
 
       - name: Upload live E2E failure artifacts
         if: failure()
-        uses: actions/upload-artifact@v7
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: live-e2e-smoke-${{ matrix.group }}-artifacts
           path: test-results/live-e2e
@@ -423,20 +426,20 @@ jobs:
             fixtures: vite8-react-tailwindv3,vite8-react-tailwindv4,vite8-react-ts,vite8-react-tsx-repeated-aside,vite8-react-unocss,vite8-react-vanilla-extract
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
       - name: Cache fixture npm downloads
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.npm
           key: ${{ runner.os }}-fixture-npm-${{ hashFiles('tests/framework-fixtures/**/files/package.json') }}
@@ -444,7 +447,7 @@ jobs:
             ${{ runner.os }}-fixture-npm-
 
       - name: Cache Playwright Chromium
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.cache/ms-playwright
           key: ${{ runner.os }}-playwright-chromium-${{ hashFiles('package.json', 'bun.lock') }}
@@ -462,7 +465,7 @@ jobs:
       - name: Install the pinned toolchain
         run: rustup show
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
 
       - name: Build the engine
         run: cargo build --release -p impeccable
@@ -478,7 +481,7 @@ jobs:
 
       - name: Upload live E2E failure artifacts
         if: failure()
-        uses: actions/upload-artifact@v7
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: live-e2e-full-${{ matrix.group }}-artifacts
           path: test-results/live-e2e
@@ -499,23 +502,23 @@ jobs:
 
       - name: Checkout repository
         if: ${{ env.ANTHROPIC_API_KEY != '' || env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
         if: ${{ env.ANTHROPIC_API_KEY != '' || env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
         if: ${{ env.ANTHROPIC_API_KEY != '' || env.DEEPSEEK_API_KEY != '' }}
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
       - name: Cache fixture npm downloads
         if: ${{ env.ANTHROPIC_API_KEY != '' || env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.npm
           key: ${{ runner.os }}-fixture-npm-${{ hashFiles('tests/framework-fixtures/**/files/package.json') }}
@@ -524,7 +527,7 @@ jobs:
 
       - name: Cache Playwright Chromium
         if: ${{ env.ANTHROPIC_API_KEY != '' || env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.cache/ms-playwright
           key: ${{ runner.os }}-playwright-chromium-${{ hashFiles('package.json', 'bun.lock') }}
@@ -545,7 +548,7 @@ jobs:
         if: ${{ env.ANTHROPIC_API_KEY != '' || env.DEEPSEEK_API_KEY != '' }}
         run: rustup show
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
         if: ${{ env.ANTHROPIC_API_KEY != '' || env.DEEPSEEK_API_KEY != '' }}
 
       - name: Build the engine
@@ -576,23 +579,23 @@ jobs:
 
       - name: Checkout repository
         if: ${{ env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
         if: ${{ env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
         if: ${{ env.DEEPSEEK_API_KEY != '' }}
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
       - name: Cache fixture npm downloads
         if: ${{ env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.npm
           key: ${{ runner.os }}-fixture-npm-${{ hashFiles('tests/framework-fixtures/**/files/package.json') }}
@@ -601,7 +604,7 @@ jobs:
 
       - name: Cache Playwright Chromium
         if: ${{ env.DEEPSEEK_API_KEY != '' }}
-        uses: actions/cache@v6
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: ~/.cache/ms-playwright
           key: ${{ runner.os }}-playwright-chromium-${{ hashFiles('package.json', 'bun.lock') }}
@@ -622,7 +625,7 @@ jobs:
         if: ${{ env.DEEPSEEK_API_KEY != '' }}
         run: rustup show
 
-      - uses: Swatinem/rust-cache@v2
+      - uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
         if: ${{ env.DEEPSEEK_API_KEY != '' }}
 
       - name: Build the engine
@@ -643,15 +646,15 @@ jobs:
       GOOGLE_CLOUD_API_KEY: ${{ secrets.GOOGLE_CLOUD_API_KEY }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 

.github/workflows/sheriff.yml

@@ -48,10 +48,10 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 

.github/workflows/sync-generated-output.yml

@@ -49,7 +49,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 0
           # Optional PAT or GitHub App token. With the default GITHUB_TOKEN,
@@ -57,12 +57,12 @@ jobs:
           token: ${{ secrets.SYNC_GENERATED_OUTPUT_TOKEN || github.token }}
 
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: 24
 
       - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 

.grok/skills/impeccable/SKILL.md

@@ -19,9 +19,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.grok/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.grok/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.grok/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.grok/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.grok/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.grok/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.grok/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.grok/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.grok/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.hermes/skills/impeccable/SKILL.md

@@ -14,9 +14,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.hermes/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.hermes/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.hermes/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.hermes/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.hermes/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.hermes/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.hermes/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.hermes/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.hermes/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.kiro/skills/impeccable/SKILL.md

@@ -14,9 +14,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.kiro/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.kiro/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.kiro/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.kiro/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.kiro/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.kiro/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.kiro/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.kiro/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.kiro/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.opencode/skills/impeccable/SKILL.md

@@ -19,9 +19,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.opencode/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.opencode/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.opencode/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.opencode/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.opencode/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.opencode/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.opencode/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.opencode/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.opencode/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.pi/skills/impeccable/SKILL.md

@@ -17,9 +17,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.pi/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.pi/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.pi/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.pi/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.pi/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.pi/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.pi/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.pi/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.pi/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.qoder/skills/impeccable/SKILL.md

@@ -19,9 +19,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.qoder/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.qoder/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.qoder/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.qoder/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.qoder/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.qoder/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.qoder/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.qoder/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.qoder/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.rovodev/skills/impeccable/SKILL.md

@@ -19,9 +19,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.rovodev/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.rovodev/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.rovodev/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.rovodev/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.rovodev/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.rovodev/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.rovodev/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.rovodev/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.rovodev/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.trae-cn/skills/impeccable/SKILL.md

@@ -16,9 +16,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.trae-cn/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.trae-cn/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.trae-cn/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.trae-cn/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.trae-cn/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.trae-cn/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.trae-cn/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.trae-cn/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.trae-cn/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.trae/skills/impeccable/SKILL.md

@@ -16,9 +16,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.trae/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.trae/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.trae/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.trae/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.trae/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.trae/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.trae/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.trae/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.trae/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.veto/skills/impeccable/SKILL.md

@@ -14,9 +14,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.veto/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.veto/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.veto/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.veto/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.veto/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.veto/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.veto/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.veto/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.veto/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

.vibe/skills/impeccable/SKILL.md

@@ -18,9 +18,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.vibe/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.vibe/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.vibe/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `.vibe/skills/impeccable/scripts/impeccable <verb>` command in this skill and its references, and `.vibe/skills/impeccable/scripts` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `.vibe/skills/impeccable/scripts/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

.vibe/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

.vibe/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `.vibe/skills/impeccable/scripts/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

Cargo.lock

@@ -111,6 +111,22 @@ version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3d7b894f5411737b7867f4827955924d7c254fc9f4d91a6aad6b097804b1018b"
 
+[[package]]
+name = "core-foundation"
+version = "0.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b2a6cd9ae233e7f62ba4e9353e81a88df7fc8a5987b8d445b4d90c879bd156f6"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
+[[package]]
+name = "core-foundation-sys"
+version = "0.8.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "773648b94d0e5d620f64f280777445740e61fe701025087ec8b57f45c791888b"
+
 [[package]]
 name = "cpufeatures"
 version = "0.2.17"
@@ -590,12 +606,14 @@ dependencies = [
  "impeccable-core",
  "once_cell",
  "regex",
+ "rustls-native-certs",
  "serde",
  "serde_json",
  "sha2",
  "tiny_http",
  "unicode-normalization",
  "ureq",
+ "webpki-roots 1.0.9",
 ]
 
 [[package]]
@@ -830,6 +848,12 @@ version = "1.21.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9f7c3e4beb33f85d45ae3e3a1792185706c8e16d043238c593331cc7cd313b50"
 
+[[package]]
+name = "openssl-probe"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7c87def4c32ab89d880effc9e097653c8da5d6ef28e6b539d313baaacfbafcbe"
+
 [[package]]
 name = "parking_lot"
 version = "0.12.5"
@@ -1063,7 +1087,7 @@ dependencies = [
  "getrandom 0.2.17",
  "libc",
  "untrusted",
- "windows-sys",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -1096,6 +1120,18 @@ dependencies = [
  "zeroize",
 ]
 
+[[package]]
+name = "rustls-native-certs"
+version = "0.8.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dab5152771c58876a2146916e53e35057e1a4dfa2b9df0f0305b07f611fdea4d"
+dependencies = [
+ "openssl-probe",
+ "rustls-pki-types",
+ "schannel",
+ "security-framework",
+]
+
 [[package]]
 name = "rustls-pki-types"
 version = "1.15.1"
@@ -1122,6 +1158,15 @@ version = "1.0.23"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cf54715a573b99ac80df0bc206da022bcd442c974952c7b9720069370852e21f"
 
+[[package]]
+name = "schannel"
+version = "0.1.29"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91c1b7e4904c873ef0710c1f407dde2e6287de2bebc1bbbf7d430bb7cbffd939"
+dependencies = [
+ "windows-sys 0.61.2",
+]
+
 [[package]]
 name = "scopeguard"
 version = "1.2.0"
@@ -1142,6 +1187,29 @@ dependencies = [
  "tendril",
 ]
 
+[[package]]
+name = "security-framework"
+version = "3.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b7f4bc775c73d9a02cde8bf7b2ec4c9d12743edf609006c7facc23998404cd1d"
+dependencies = [
+ "bitflags",
+ "core-foundation",
+ "core-foundation-sys",
+ "libc",
+ "security-framework-sys",
+]
+
+[[package]]
+name = "security-framework-sys"
+version = "2.17.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6ce2691df843ecc5d231c0b14ece2acc3efb62c0a398c7e1d875f3983ce020e3"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
 [[package]]
 name = "selectors"
 version = "0.38.0"
@@ -1603,6 +1671,15 @@ dependencies = [
  "windows-targets",
 ]
 
+[[package]]
+name = "windows-sys"
+version = "0.61.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ae137229bcbd6cdf0f7b80a31df61766145077ddf49416a728b02cb3921ff3fc"
+dependencies = [
+ "windows-link",
+]
+
 [[package]]
 name = "windows-targets"
 version = "0.52.6"

README.md

@@ -397,6 +397,8 @@ Installed hook surfaces:
 
 Every command goes through the launcher shipped in the skill's `scripts/` directory (`impeccable`, or `impeccable.cmd` on Windows), guarded so a missing launcher is a silent no-op. The launcher runs the engine binary that ships next to it, or downloads the pinned version once into `~/.impeccable/bin/`. No Node or other runtime is required for the hook or the skill.
 
+In Claude Code, installed command hooks run independently of model-tool approval. The first edit or Stop event can therefore download and cache the engine even if the session denies the model's launcher command. Review installed hooks before unattended runs; to disable all Claude Code hooks for a run, pass `--settings '{"disableAllHooks": true}'`. See [Claude Code's hook security guidance](https://code.claude.com/docs/en/hooks#security-considerations).
+
 The installer preserves unrelated hook entries and settings. If a hook manifest is malformed, install/update aborts by default; rerun with `--force` to back up the malformed file as `.bak` and replace it.
 
 On an interactive `install`/`update`, Impeccable explains the hook and offers to install it (default yes). Your choice is remembered per-developer in the gitignored `.impeccable/config.local.json`, so you are not asked again; `--no-hooks` skips it for that run without recording anything. Hook lifecycle settings live under the `hook` key of `.impeccable/config.json`; detector ignores live under `detector`, shared by `/impeccable hooks` and `npx impeccable detect`.
@@ -419,6 +421,8 @@ Codex requires one platform step that Impeccable cannot safely skip: open `/hook
 
 Full hook docs: [impeccable.style/docs/hooks](https://impeccable.style/docs/hooks).
 
+The Stop pass suppresses confirmed pre-existing findings when a verified before-edit baseline is available (currently Claude Edit/Write results for text scans). Other findings are marked new or attribution unknown; unknown is not evidence that your session caused the problem. Explicit `detect` scans remain unchanged.
+
 Manual copy commands are fallback/debug instructions. The normal path is:
 
 ```bash
@@ -426,6 +430,14 @@ npx impeccable install
 npx impeccable update
 ```
 
+## Live mode and production sites
+
+Live mode edits a local checkout through a development server or local static HTML. Injecting its localhost HTTP helper into a deployed production site, including an HTTPS site, is not supported. Do not disable browser security or weaken production CSP to make it work.
+
+Use live mode only in projects you trust to run locally. Applying copy edits automatically runs `package.json`'s optional `scripts["impeccable:manual-edit-validate"]` command in a shell, with your user permissions; review that script before using live mode in an unfamiliar checkout.
+
+For production inspection, use `npx impeccable detect https://example.com` or the browser extension. These inspect the rendered page; they do not provide live variant editing or write changes back to your source.
+
 ## CLI
 
 Impeccable includes a standalone CLI for detecting anti-patterns without an AI harness. `npx impeccable` is a small shim that runs the same engine binary the skill uses (installed as a platform-specific optional dependency, or fetched once into `~/.impeccable/bin/`); Node is needed only for `npx` itself, and you can also download the binary directly and put it on your PATH.

README.npm.md

@@ -81,6 +81,7 @@ impeccable detect [options] [file-or-dir-or-url...]
 
 - Node.js 22.18+ to run `npx impeccable`. The engine itself is a self-contained binary and needs no runtime; the skill installed into your harness calls it directly.
 - For URL scans, an installed Chrome, Chromium, or Edge (set `IMPECCABLE_BROWSER` to point at one).
+- Behind a TLS-inspecting proxy, downloads trust your OS certificate store as well as the bundled Mozilla roots. Set `SSL_CERT_FILE` or `SSL_CERT_DIR` to use a specific CA bundle instead.
 
 Binary lookup order: `IMPECCABLE_BIN`, the platform package, `~/.impeccable/bin/<version>/`, then a download of the pinned version into that cache. Set `IMPECCABLE_BIN` to a local build to skip all of that.
 

crates/context/Cargo.toml

@@ -21,5 +21,7 @@ once_cell = { workspace = true }
 sha2 = "0.10"
 flate2 = { version = "1", default-features = false, features = ["zlib-rs"] }
 ureq = { version = "2", default-features = false, features = ["tls", "json"] }
+rustls-native-certs = "0.8"
+webpki-roots = "1"
 tiny_http = "0.12"
 unicode-normalization = "0.1.25"

crates/context/src/concept_seed.rs

@@ -47,7 +47,7 @@ fn card_base(env: &Env) -> String {
 }
 
 fn agent(timeout: Duration) -> ureq::Agent {
-    ureq::AgentBuilder::new().timeout_connect(timeout).timeout(timeout).build()
+    crate::http::agent_builder().timeout_connect(timeout).timeout(timeout).build()
 }
 
 /// URLSearchParams serialization (application/x-www-form-urlencoded).

crates/context/src/context_cli.rs

@@ -468,7 +468,7 @@ fn fetch_latest_skill_version(env: &Env) -> Option<String> {
         .cloned()
         .unwrap_or_else(|| "https://impeccable.style".to_string());
     let host = host.strip_suffix('/').unwrap_or(&host).to_string();
-    let agent = ureq::AgentBuilder::new()
+    let agent = crate::http::agent_builder()
         .timeout(std::time::Duration::from_millis(FETCH_TIMEOUT_MS))
         .build();
     let res = agent.get(&format!("{}/api/version", host)).call().ok()?;

crates/context/src/generate_image.rs

@@ -294,7 +294,7 @@ pub fn run(args: &[String], io: &mut Io) -> i32 {
             }
         }
     }
-    let agent = ureq::AgentBuilder::new().build();
+    let agent = crate::http::agent_builder().build();
     let response = if !refs.is_empty() {
         let boundary = format!("----impeccable{:x}", crate::util::now_ms() as u64);
         let mut body: Vec<u8> = Vec::new();

crates/context/src/http.rs

@@ -0,0 +1,104 @@
+//! One TLS trust configuration for every HTTPS request the engine makes.
+//!
+//! `ureq`'s default rustls config trusts only the Mozilla roots compiled in
+//! through `webpki-roots`. On a machine where an endpoint security agent
+//! inspects TLS (Aikido, Zscaler, Netskope: routine in managed corporate
+//! setups), every connection terminates at a proxy whose root lives in the
+//! OS trust store and nowhere else, so `update` and `install` failed with
+//! `invalid peer certificate: UnknownIssuer` while curl and npm on the same
+//! machine succeeded (#757).
+//!
+//! The store built here is the union of the OS trust store
+//! (`rustls-native-certs`: the macOS Keychain, the Windows store, the
+//! OpenSSL paths on Linux) and the bundled Mozilla roots. A union, not a
+//! replacement: a container without `ca-certificates`, or a store that
+//! fails to load, verifies against the bundled roots exactly as before.
+//! `SSL_CERT_FILE` / `SSL_CERT_DIR` stand in for the OS store, as they do
+//! for OpenSSL and curl; the bundled roots stay either way.
+
+use std::sync::Arc;
+
+use once_cell::sync::Lazy;
+use ureq::rustls::pki_types::CertificateDer;
+use ureq::rustls::{self, ClientConfig, RootCertStore};
+
+/// `ureq::AgentBuilder::new()` with the engine's trust store installed.
+/// Every HTTPS call site builds its agent from this; the plain-HTTP calls
+/// to the live server on localhost do not need it.
+pub fn agent_builder() -> ureq::AgentBuilder {
+    ureq::AgentBuilder::new().tls_config(tls_config())
+}
+
+fn tls_config() -> Arc<ClientConfig> {
+    static CONFIG: Lazy<Arc<ClientConfig>> = Lazy::new(|| {
+        // Mirrors ureq's own default config (provider and protocol versions);
+        // only the root store differs.
+        let config =
+            ClientConfig::builder_with_provider(rustls::crypto::ring::default_provider().into())
+                .with_protocol_versions(&[&rustls::version::TLS12, &rustls::version::TLS13])
+                .expect("the ring provider supports TLS 1.2 and 1.3")
+                .with_root_certificates(root_store(rustls_native_certs::load_native_certs().certs))
+                .with_no_client_auth();
+        Arc::new(config)
+    });
+    CONFIG.clone()
+}
+
+/// The bundled Mozilla roots plus every parsable certificate in `native`.
+/// Unparsable entries are dropped, so one broken certificate in the OS
+/// store cannot take the bundled roots down with it.
+fn root_store(native: Vec<CertificateDer<'static>>) -> RootCertStore {
+    let mut store = RootCertStore {
+        roots: webpki_roots::TLS_SERVER_ROOTS.to_vec(),
+    };
+    store.add_parsable_certificates(native);
+    store
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use ureq::rustls::pki_types::pem::PemObject;
+
+    /// Self-signed CA minted for this test (P-256, v3, CA:TRUE): the shape
+    /// of the root a TLS-inspecting proxy installs into the OS store.
+    const PROXY_ROOT_PEM: &str = "-----BEGIN CERTIFICATE-----
+MIIBdTCCARugAwIBAgIJANhTZvQvv7HJMAoGCCqGSM49BAMCMB0xGzAZBgNVBAMM
+EmltcGVjY2FibGUgdGVzdCBDQTAgFw0yNjA5MDcwNjQxMjdaGA8yMTI2MDgxNDA2
+NDEyN1owHTEbMBkGA1UEAwwSaW1wZWNjYWJsZSB0ZXN0IENBMFkwEwYHKoZIzj0C
+AQYIKoZIzj0DAQcDQgAEYVZtCOXaZsY71/0Roy62iBVcyx8UfMDkPbEbf/IEw5Bm
+yNBfKTFS/8FbRBMWHXOwNE0Ns1BLVOB1oQ1XFC5Bz6NCMEAwDwYDVR0TAQH/BAUw
+AwEB/zAOBgNVHQ8BAf8EBAMCAQYwHQYDVR0OBBYEFFONzBxi7ewOfuP6cBIIqsxu
+3pEiMAoGCCqGSM49BAMCA0gAMEUCIQD98Q0ZRe8ceuopnUwQKYleZd5IzfWhhpmO
+tB0WGTOG3QIgdJa8gBPU9Y6WsrursItsnUeGTYHKDCZZ6MjlekLFuoc=
+-----END CERTIFICATE-----
+";
+
+    fn bundled() -> usize {
+        webpki_roots::TLS_SERVER_ROOTS.len()
+    }
+
+    #[test]
+    fn bundled_roots_alone_when_the_os_store_is_empty() {
+        assert_eq!(root_store(Vec::new()).len(), bundled());
+    }
+
+    #[test]
+    fn os_store_root_joins_the_bundled_roots() {
+        let proxy = CertificateDer::from_pem_slice(PROXY_ROOT_PEM.as_bytes()).unwrap();
+        assert_eq!(root_store(vec![proxy]).len(), bundled() + 1);
+    }
+
+    #[test]
+    fn unparsable_os_store_entry_is_dropped() {
+        let junk = CertificateDer::from(b"not a certificate".to_vec());
+        assert_eq!(root_store(vec![junk]).len(), bundled());
+    }
+
+    #[test]
+    fn agent_builds_from_this_hosts_store() {
+        // Runs the real rustls-native-certs load: it must not panic, and the
+        // shared config must be accepted by a ureq agent.
+        let _agent = agent_builder().build();
+    }
+}

crates/context/src/lib.rs

@@ -5,6 +5,7 @@
 pub mod jsp;
 pub mod util;
 pub mod url;
+pub mod http;
 pub mod provider;
 pub mod hook_markers;
 pub mod target_args;

crates/hook/src/hook.rs

@@ -7,6 +7,7 @@ use impeccable_core::js;
 use serde_json::{Map, Value};
 
 use crate::hook_lib::*;
+use crate::stop_baseline;
 use crate::util::{
     exists, iso_now, jsp, node_read_error, now_ms, str_field, truthy_value, utf16_len,
 };
@@ -237,11 +238,19 @@ pub fn run_hook(rt: &Runtime, stdin: &str) -> RunResult {
             }
         }
 
+        let use_html_engine = match configured {
+            Some(c) => c.engine == "html",
+            None => ext == ".html" || ext == ".htm",
+        };
         if primary_files.contains(file_path) {
+            if harness == "claude" {
+                stop_baseline::capture(rt, &event, &mut cache, &session_id, file_path, use_html_engine);
+            }
             let edit_count = bump_edit_count(&mut cache, &session_id, file_path);
             cache_dirty = true;
             audit.insert("editCount".into(), Value::from(edit_count as u64));
             if edit_count > EDIT_COUNT_THRESHOLD as f64 {
+                stop_baseline::invalidate(&mut cache, &session_id, file_path);
                 let just_crossed = edit_count == (EDIT_COUNT_THRESHOLD + 1) as f64;
                 if just_crossed && suppression_winner.is_none() {
                     suppression_winner = Some(file_path.clone());
@@ -266,10 +275,6 @@ pub fn run_hook(rt: &Runtime, stdin: &str) -> RunResult {
                 };
             }
         };
-        let use_html_engine = match configured {
-            Some(c) => c.engine == "html",
-            None => ext == ".html" || ext == ".htm",
-        };
         let mut detector_threw = false;
         let findings: Vec<Finding> = if use_html_engine {
             match detector_detect_html(rt, file_path, &scan) {
@@ -282,6 +287,9 @@ pub fn run_hook(rt: &Runtime, stdin: &str) -> RunResult {
         } else {
             detector_detect_text(&content, file_path, &scan)
         };
+        if !detector_threw && !use_html_engine {
+            stop_baseline::reconcile(&mut cache, &session_id, file_path, &findings);
+        }
         let raw_count = findings.len();
         let filtered = filter_findings(findings, &config);
         let (immediate, deferred) = if tiered {
@@ -660,6 +668,9 @@ pub fn run_stop_hook(rt: &Runtime, stdin: &str) -> RunResult {
     let mut fresh_groups: Vec<Group> = Vec::new();
     let mut scanned = 0usize;
     let mut cache_dirty = false;
+    let mut pre_existing = 0usize;
+    let mut new_findings = 0usize;
+    let mut unknown = 0usize;
     for file_path in &touched {
         if scanned >= STOP_MAX_FILES {
             break;
@@ -705,8 +716,15 @@ pub fn run_stop_hook(rt: &Runtime, stdin: &str) -> RunResult {
         } else {
             detector_detect_text(&content, file_path, &scan)
         };
+        if !use_html_engine {
+            stop_baseline::reconcile(&mut cache, &session_id, file_path, &findings);
+        }
         let filtered = filter_findings(findings, &config);
-        let fresh = dedupe_against_cache(&filtered, &mut cache, &session_id, file_path);
+        let classified = stop_baseline::classify(&cache, &session_id, file_path, use_html_engine, filtered.clone());
+        pre_existing += classified.pre_existing;
+        new_findings += classified.new;
+        unknown += classified.unknown;
+        let fresh = dedupe_against_cache(&classified.findings, &mut cache, &session_id, file_path);
         // JS: 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
@@ -721,6 +739,9 @@ pub fn run_stop_hook(rt: &Runtime, stdin: &str) -> RunResult {
         }
     }
     audit.insert("scannedFiles".into(), Value::from(scanned));
+    audit.insert("preExistingFindings".into(), Value::from(pre_existing));
+    audit.insert("newFindings".into(), Value::from(new_findings));
+    audit.insert("unknownFindings".into(), Value::from(unknown));
     if fresh_groups.is_empty() {
         if cache_dirty {
             persist_cache(rt, &project_cwd, &cache);
@@ -735,19 +756,52 @@ pub fn run_stop_hook(rt: &Runtime, stdin: &str) -> RunResult {
         );
     }
     let short = footer_mode_short(&mut cache, &session_id);
-    let reserve = design_note_reserve(rt, &scan, &mut cache, &session_id);
-    let rendered = render_grouped_template(
+    let first_unknown = fresh_groups.iter().flat_map(|group| &group.findings)
+        .position(|f| f.name.starts_with("[attribution unknown]"));
+    let mut attribution_note = if first_unknown.is_some() {
+        format!("{ENVELOPE_PREFIX} {}", stop_baseline::UNKNOWN_NOTE)
+    } else {
+        String::new()
+    };
+    // Findings and attribution take priority. Append the lower-priority stale
+    // DESIGN.md notice only if it fits, without consuming its session flag.
+    let render = |note: &str, render_config: &HookConfig| render_grouped_template(
         rt,
         &fresh_groups,
-        &config,
+        render_config,
         &RenderOpts {
             cwd: Some(project_cwd.clone()),
             short_footer: short,
-            reserve_chars: reserve,
+            reserve_chars: if note.is_empty() { 0.0 } else { (utf16_len(note) + 2) as f64 },
         },
     );
+    let mut rendered = render(&attribution_note, &config);
+    if !attribution_note.is_empty() && !rendered.lines().any(|line| {
+        line.starts_with("- ") && (line.contains("[attribution unknown]") || line.contains("[new]"))
+    }) {
+        // At the minimum budget, a grouped header and policy footer may crowd
+        // out even the first finding. Shorten the notice before losing it.
+        attribution_note = format!("{ENVELOPE_PREFIX} {}", stop_baseline::COMPACT_UNKNOWN_NOTE);
+        rendered = render(&attribution_note, &config);
+    }
+    // maxFindings / maxChars may also remove all unknown findings. Do not
+    // attach their guidance to an output that only shows confirmed new debt.
+    let shows_unknown = rendered.lines().any(|line| {
+        line.starts_with("- ") && line.contains("[attribution unknown]")
+    });
+    if !shows_unknown {
+        if let Some(prefix @ 1..) = first_unknown {
+            // Reclaim the unused notice budget for the known-new prefix.
+            // Keep the unknown suffix omitted: simply expanding the budget
+            // could reveal an unknown finding without its required guidance.
+            let mut visible_config = config.clone();
+            visible_config.limits.max_findings = cap_of(&config).min(prefix) as f64;
+            rendered = render("", &visible_config);
+        }
+    }
+    let text = if shows_unknown { format!("{attribution_note}\n\n{rendered}") } else { rendered };
     let text =
-        append_design_system_note_once(rt, &rendered, &scan, &mut cache, &session_id, &config);
+        append_design_system_note_once(rt, &text, &scan, &mut cache, &session_id, &config);
     commit_footer_shown(rt, &mut cache, &session_id, &text);
     persist_cache(rt, &project_cwd, &cache);
     let all: usize = fresh_groups.iter().map(|g| g.findings.len()).sum();

crates/hook/src/hook_lib.rs

@@ -1314,7 +1314,7 @@ pub struct RenderOpts {
     pub reserve_chars: f64,
 }
 
-fn cap_of(config: &HookConfig) -> usize {
+pub(crate) fn cap_of(config: &HookConfig) -> usize {
     let mf = config.limits.max_findings;
     let mf = if mf == 0.0 || mf.is_nan() {
         DEFAULT_MAX_FINDINGS

crates/hook/src/lib.rs

@@ -14,6 +14,7 @@ pub mod admin;
 pub mod before_edit;
 pub mod hook;
 pub mod hook_lib;
+mod stop_baseline;
 pub mod util;
 
 use impeccable_common::Io;

crates/hook/src/stop_baseline.rs

@@ -0,0 +1,279 @@
+//! Conservative Stop attribution. Never use HEAD as a session baseline: the
+//! working tree may already be dirty. Only a verified first Edit/Write preimage
+//! from the tool result establishes a baseline, and only for the pure text
+//! detector. DOM and design-system findings can depend on other files.
+
+use impeccable_core::findings::Finding;
+use serde_json::{json, Map, Value};
+use sha2::{Digest, Sha256};
+
+use crate::hook_lib::{
+    detector_detect_text, ensure_file, sessions, Cache, HookScanOptions, Runtime,
+};
+
+const FIELD: &str = "stopBaseline";
+const MAX_BYTES: usize = 512 * 1024;
+const MAX_FINDINGS: usize = 256;
+pub const UNKNOWN_NOTE: &str = "Findings marked attribution unknown may predate this session; do not treat them as regressions or broaden the task without asking.";
+pub const COMPACT_UNKNOWN_NOTE: &str = "Unknown findings may predate this session; ask before expanding scope.";
+
+fn independent(finding: &Finding) -> bool {
+    !finding.antipattern.starts_with("design-system-")
+}
+
+// Exclude line numbers so an unrelated insertion/deletion does not make old
+// debt new. Preserve multiplicity so adding an identical occurrence is new.
+// Hash the detector identity rather than storing source or snippets in cache.
+fn key(finding: &Finding) -> String {
+    let identity = json!([finding.antipattern, finding.snippet, finding.extras]);
+    format!("{:x}", Sha256::digest(identity.to_string().as_bytes()))
+}
+
+fn counts(findings: &[Finding]) -> Map<String, Value> {
+    let mut counts = Map::new();
+    for finding in findings.iter().filter(|f| independent(f)) {
+        let key = key(finding);
+        let count = counts.get(&key).and_then(Value::as_u64).unwrap_or(0);
+        counts.insert(key, Value::from(count + 1));
+    }
+    counts
+}
+
+fn entry<'a>(cache: &'a Cache, session: &str, file: &str) -> Option<&'a Map<String, Value>> {
+    sessions(cache)?
+        .get(session)?
+        .get("files")?
+        .get(file)?
+        .as_object()
+}
+
+fn baseline(cache: &Cache, session: &str, file: &str) -> Option<Map<String, Value>> {
+    let value = entry(cache, session, file)?.get(FIELD)?;
+    if value.get("version")?.as_u64()? != 1
+        || value.get("engine")?.as_str()? != env!("CARGO_PKG_VERSION")
+    {
+        return None;
+    }
+    let counts = value.get("counts")?.as_object()?;
+    if counts.len() > MAX_FINDINGS
+        || counts.iter().any(|(k, v)| {
+            k.len() != 64
+                || !k.bytes().all(|b| b.is_ascii_hexdigit())
+                || !matches!(v.as_u64(), Some(1..=256))
+        })
+    {
+        return None;
+    }
+    Some(counts.clone())
+}
+
+/// Called before the first primary edit is recorded. An entry without a
+/// baseline (old cache, co-scan, incomplete payload) must stay unknown rather
+/// than adopting a later, already-edited file as its starting point.
+pub fn capture(
+    rt: &Runtime,
+    event: &Map<String, Value>,
+    cache: &mut Cache,
+    session: &str,
+    file: &str,
+    html: bool,
+) {
+    if html || session.is_empty() || session == "unknown" || entry(cache, session, file).is_some() {
+        return;
+    }
+    let Some(response) = event.get("tool_response").and_then(Value::as_object) else {
+        return;
+    };
+    let Some(path) = response.get("filePath").and_then(Value::as_str) else {
+        return;
+    };
+    let cwd = event
+        .get("cwd")
+        .and_then(Value::as_str)
+        .unwrap_or(&rt.proc_cwd);
+    if rt.resolve(&[cwd, path]) != file || response.get("userModified") == Some(&Value::Bool(true))
+    {
+        return;
+    }
+    let Some(tool) = event.get("tool_name").and_then(Value::as_str) else {
+        return;
+    };
+    let original = match response.get("originalFile") {
+        Some(Value::String(text)) if text.len() <= MAX_BYTES => text.as_str(),
+        // null on an update may mean "too large", not an empty original.
+        Some(Value::Null)
+            if tool == "Write"
+                && response.get("type").and_then(Value::as_str) == Some("create") =>
+        {
+            ""
+        }
+        _ => return,
+    };
+    let expected = match tool {
+        "Edit" => {
+            let Some(old) = response.get("oldString").and_then(Value::as_str) else {
+                return;
+            };
+            let Some(new) = response.get("newString").and_then(Value::as_str) else {
+                return;
+            };
+            if old.is_empty() || new.len() > MAX_BYTES || !original.contains(old) {
+                return;
+            }
+            match response.get("replaceAll").and_then(Value::as_bool) {
+                Some(true) => {
+                    let occurrences = original.matches(old).count();
+                    let size = original.len() - occurrences * old.len()
+                        + occurrences.saturating_mul(new.len());
+                    if size > MAX_BYTES {
+                        return;
+                    }
+                    original.replace(old, new)
+                }
+                Some(false) if original.matches(old).count() == 1 => original.replacen(old, new, 1),
+                _ => return,
+            }
+        }
+        "Write" => {
+            if !matches!(
+                response.get("type").and_then(Value::as_str),
+                Some("create" | "update")
+            ) {
+                return;
+            }
+            let Some(content) = response.get("content").and_then(Value::as_str) else {
+                return;
+            };
+            if content.len() > MAX_BYTES {
+                return;
+            }
+            content.to_string()
+        }
+        _ => return,
+    };
+    if expected.len() > MAX_BYTES
+        || std::fs::metadata(file)
+            .map(|m| m.len() > MAX_BYTES as u64)
+            .unwrap_or(true)
+    {
+        return;
+    }
+    // A formatter, stale event, or concurrent write invalidates attribution.
+    if std::fs::read_to_string(file).ok().as_deref() != Some(expected.as_str()) {
+        return;
+    }
+    let findings = detector_detect_text(original, file, &HookScanOptions::default());
+    if findings.len() > MAX_FINDINGS {
+        return;
+    }
+    ensure_file(cache, session, file).insert(
+        FIELD.into(),
+        json!({
+            "version": 1, "engine": env!("CARGO_PKG_VERSION"), "counts": counts(&findings),
+        }),
+    );
+}
+
+/// Once existing debt disappears, it cannot exempt a later reintroduction.
+pub fn reconcile(cache: &mut Cache, session: &str, file: &str, findings: &[Finding]) {
+    let Some(mut old) = baseline(cache, session, file) else {
+        return;
+    };
+    let current = counts(findings);
+    old.retain(|key, value| {
+        let count = current
+            .get(key)
+            .and_then(Value::as_u64)
+            .unwrap_or(0)
+            .min(value.as_u64().unwrap_or(0));
+        *value = Value::from(count);
+        count > 0
+    });
+    ensure_file(cache, session, file).get_mut(FIELD).unwrap()["counts"] = Value::Object(old);
+}
+
+/// Do not retain an exemption through edits we deliberately stop scanning.
+pub fn invalidate(cache: &mut Cache, session: &str, file: &str) {
+    ensure_file(cache, session, file).remove(FIELD);
+}
+
+#[derive(Default)]
+pub struct Classified {
+    pub findings: Vec<Finding>,
+    pub pre_existing: usize,
+    pub new: usize,
+    pub unknown: usize,
+}
+
+pub fn classify(
+    cache: &Cache,
+    session: &str,
+    file: &str,
+    html: bool,
+    findings: Vec<Finding>,
+) -> Classified {
+    let mut baseline = if html {
+        None
+    } else {
+        baseline(cache, session, file)
+    };
+    let mut result = Classified::default();
+    for mut finding in findings {
+        let known = baseline.as_mut().filter(|_| independent(&finding));
+        if let Some(counts) = known {
+            let key = key(&finding);
+            let count = counts.get(&key).and_then(Value::as_u64).unwrap_or(0);
+            if count > 0 {
+                counts.insert(key, Value::from(count - 1));
+                result.pre_existing += 1;
+                continue;
+            }
+            result.new += 1;
+            finding.name = format!("[new] {}", finding.name);
+        } else {
+            result.unknown += 1;
+            finding.name = format!("[attribution unknown] {}", finding.name);
+        }
+        result.findings.push(finding);
+    }
+    result
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use impeccable_core::findings::finding;
+
+    #[test]
+    fn stop_baseline_dependency_sensitive_findings_stay_unknown() {
+        let mut cache = Cache::new();
+        let font = finding("design-system-font", "a.css", "font-family: serif", 1.0);
+        let css = finding("side-tab", "a.html", "border-left: 4px solid red", 1.0);
+        ensure_file(&mut cache, "s", "a.css").insert(
+            FIELD.into(),
+            json!({
+                "version": 1, "engine": env!("CARGO_PKG_VERSION"), "counts": {},
+            }),
+        );
+        // Even a known text preimage cannot establish the state of DESIGN.md
+        // before the session, or of the stylesheets a DOM scan reads.
+        assert_eq!(classify(&cache, "s", "a.css", false, vec![font]).unknown, 1);
+        assert_eq!(classify(&cache, "s", "a.css", true, vec![css]).unknown, 1);
+    }
+
+    #[test]
+    fn stop_baseline_old_engine_and_malformed_cache_stay_unknown() {
+        let mut cache = Cache::new();
+        let f = finding("side-tab", "a.css", "border-left: 4px solid red", 1.0);
+        for record in [
+            json!({"version": 1, "engine": "0.0.0", "counts": {}}),
+            json!({"version": 1, "engine": env!("CARGO_PKG_VERSION"), "counts": {"bad": -1}}),
+        ] {
+            ensure_file(&mut cache, "s", "a.css").insert(FIELD.into(), record);
+            assert_eq!(
+                classify(&cache, "s", "a.css", false, vec![f.clone()]).unknown,
+                1
+            );
+        }
+    }
+}

crates/hook/tests/hook_tests.rs

@@ -124,6 +124,350 @@ fn stop_event(cwd: &str, session: &str) -> String {
 const GRADIENT_CSS: &str = ".title { background: linear-gradient(90deg, #f472b6, #a78bfa); -webkit-background-clip: text; color: transparent; }\n";
 const SIDE_TAB_CSS: &str = ".card { border-left: 4px solid #6366f1; border-radius: 8px; }\n";
 
+fn edit_with_original(cwd: &str, file: &str, session: &str, before: &str, old: &str, new: &str) -> String {
+    json!({
+        "session_id": session, "cwd": cwd, "hook_event_name": "PostToolUse",
+        "tool_name": "Edit", "tool_input": {"file_path": file, "old_string": old, "new_string": new},
+        "tool_response": {"filePath": file, "originalFile": before, "oldString": old,
+            "newString": new, "replaceAll": false, "userModified": false},
+    }).to_string()
+}
+
+#[test]
+fn stop_baseline_import_only_edit_does_not_blame_existing_font() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    // This is the actual working-tree preimage, not HEAD (which might differ).
+    let before = "import dead from 'dead';\nconst report = `<style>body { font-family: Fraunces; }</style>`;\n";
+    let after = before.replacen("import dead from 'dead';\n", "", 1);
+    let file = t.write("query.ts", &after);
+    let r = rt(&cwd);
+    assert!(detector_detect_text(before, &file, &HookScanOptions::default()).iter().any(|f| f.antipattern == "overused-font"));
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", before, "import dead from 'dead';\n", ""));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert!(stop.stdout.is_empty(), "{}", stop.stdout);
+    assert_eq!(stop.audit["preExistingFindings"], json!(1));
+    assert!(!t.read(".impeccable/hook.cache.json").contains("const report"), "do not persist source contents");
+    assert!(!t.exists(".impeccable/config.local.json"), "baseline is not an ignore");
+    assert!(detector_detect_text(&after, &file, &HookScanOptions::default()).iter().any(|f| f.antipattern == "overused-font"), "explicit scans stay unchanged");
+}
+
+#[test]
+fn stop_baseline_reports_new_findings_and_keeps_first_preimage() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    let file = t.write("card.css", SIDE_TAB_CSS);
+    let r = rt(&cwd);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", ".card {}\n", ".card {}\n", SIDE_TAB_CSS));
+    let second = format!("/* later */\n{SIDE_TAB_CSS}");
+    t.write("card.css", &second);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", SIDE_TAB_CSS, SIDE_TAB_CSS, &second));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert!(stop.stdout.contains("[side-tab]"));
+    assert!(stop.stdout.contains("[new]"), "{}", stop.stdout);
+    assert_eq!(stop.audit["newFindings"], json!(1));
+}
+
+#[test]
+fn stop_baseline_missing_or_mismatched_preimage_stays_unknown() {
+    for original in [None, Some("not the actual preimage")] {
+        let t = Tmp::new();
+        let cwd = t.path();
+        t.write("package.json", "{}");
+        let file = t.write("card.css", SIDE_TAB_CSS);
+        let r = rt(&cwd);
+        let event = original.map(|before| edit_with_original(&cwd, &file, "s1", before, ".card {}", SIDE_TAB_CSS))
+            .unwrap_or_else(|| edit_event(&cwd, &file, "s1"));
+        hook::run_hook(&r, &event);
+        let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+        assert!(stop.stdout.contains("[attribution unknown]"), "{}", stop.stdout);
+        assert!(stop.stdout.contains("may predate this session"));
+        assert_eq!(stop.audit["unknownFindings"], json!(1));
+    }
+}
+
+#[test]
+fn stop_baseline_existing_debt_fixed_then_reintroduced_is_new() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    let clean = ".card {}\n";
+    let file = t.write("card.css", clean);
+    let r = rt(&cwd);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", SIDE_TAB_CSS, SIDE_TAB_CSS, clean));
+    t.write("card.css", SIDE_TAB_CSS);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", clean, clean, SIDE_TAB_CSS));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert!(stop.stdout.contains("[new]"), "{}", stop.stdout);
+    assert_eq!(stop.audit["preExistingFindings"], json!(0));
+}
+
+#[test]
+fn stop_baseline_late_preimage_does_not_relabel_unknown_debt() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    let file = t.write("card.css", SIDE_TAB_CSS);
+    let r = rt(&cwd);
+    hook::run_hook(&r, &edit_event(&cwd, &file, "s1"));
+    let second = format!("/* later */\n{SIDE_TAB_CSS}");
+    t.write("card.css", &second);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", SIDE_TAB_CSS, SIDE_TAB_CSS, &second));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert_eq!(stop.audit["unknownFindings"], json!(1));
+    assert_eq!(stop.audit["preExistingFindings"], json!(0));
+}
+
+#[test]
+fn stop_baseline_keeps_indirect_stylesheet_findings_unknown() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    t.write("src/styles.css", SIDE_TAB_CSS);
+    let before = "import './styles.css';\nexport const Card = () => <div>Before</div>;\n";
+    let after = before.replace("Before", "After");
+    let file = t.write("src/Card.tsx", &after);
+    let r = rt(&cwd);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", before, "Before", "After"));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert!(stop.stdout.contains("[side-tab]"), "{}", stop.stdout);
+    assert_eq!(stop.audit["unknownFindings"], json!(1));
+    assert_eq!(stop.audit["preExistingFindings"], json!(0));
+}
+
+#[test]
+fn stop_baseline_extra_identical_occurrence_is_not_suppressed() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    // The text detector deduplicates identical snippets within two lines.
+    let after = format!("{SIDE_TAB_CSS}\n\n\n{SIDE_TAB_CSS}");
+    let file = t.write("card.css", &after);
+    let r = rt(&cwd);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", SIDE_TAB_CSS, SIDE_TAB_CSS, &after));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert_eq!(stop.audit["preExistingFindings"], json!(1));
+    assert_eq!(stop.audit["newFindings"], json!(1));
+}
+
+#[test]
+fn stop_baseline_write_create_is_new_but_missing_update_preimage_is_unknown() {
+    for (kind, expected) in [("create", "newFindings"), ("update", "unknownFindings")] {
+        let t = Tmp::new();
+        let cwd = t.path();
+        t.write("package.json", "{}");
+        let file = t.write("card.css", SIDE_TAB_CSS);
+        let r = rt(&cwd);
+        let event = json!({"cwd": cwd, "session_id": "s1", "tool_name": "Write",
+            "tool_input": {"file_path": file, "content": SIDE_TAB_CSS},
+            "tool_response": {"type": kind, "filePath": file, "content": SIDE_TAB_CSS, "originalFile": null}}).to_string();
+        hook::run_hook(&r, &event);
+        let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+        assert_eq!(stop.audit[expected], json!(1));
+    }
+}
+
+#[test]
+fn stop_baseline_unknown_notice_respects_small_output_budget() {
+    for (budget, stale) in [(500, false), (500, true), (8000, true)] {
+        let t = Tmp::new();
+        let cwd = t.path();
+        t.write("package.json", "{}");
+        t.write(".impeccable/config.json", &json!({"hook":{"limits":{"maxChars":budget}}}).to_string());
+        let file = t.write("card.css", SIDE_TAB_CSS);
+        let r = rt(&cwd);
+        hook::run_hook(&r, &edit_event(&cwd, &file, "s1"));
+        if stale {
+            // Make the notice eligible only at Stop; no sleeps or clock races.
+            t.write("DESIGN.md", "---\nname: Test\n---\n");
+            let sidecar = t.write(".impeccable/design.json", "{}");
+            std::fs::File::options().write(true).open(sidecar).unwrap()
+                .set_modified(std::time::UNIX_EPOCH + std::time::Duration::from_secs(1_600_000_000)).unwrap();
+            assert!(design_system_options(&read_config(&cwd), &cwd).md_newer_than_json());
+        }
+        let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+        let output: Value = serde_json::from_str(&stop.stdout).unwrap();
+        let text = output["hookSpecificOutput"]["additionalContext"].as_str().unwrap();
+        assert!(text.encode_utf16().count() <= budget, "{text}");
+        assert!(text.contains("may predate this session"));
+        assert!(text.contains("[side-tab]"), "{text}");
+        assert!(text.contains("[attribution unknown]"), "{text}");
+        assert!(text.contains("card.css"), "{text}");
+        if stale {
+            assert_eq!(text.contains("DESIGN.md is newer"), budget > 500, "{text}");
+            let cache: Value = serde_json::from_str(&t.read(".impeccable/hook.cache.json")).unwrap();
+            assert_eq!(cache["sessions"]["s1"]["designNoteShown"] == json!(true), budget > 500);
+        }
+    }
+}
+
+#[test]
+fn stop_baseline_deduplicated_unknown_does_not_add_notice_to_new_finding() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    let r = rt(&cwd);
+    let old = t.write("old/card.css", SIDE_TAB_CSS);
+    hook::run_hook(&r, &edit_event(&cwd, &old, "s1"));
+    assert!(hook::run_stop_hook(&r, &stop_event(&cwd, "s1")).stdout.contains("[attribution unknown]"));
+    let new = t.write("new/card.css", SIDE_TAB_CSS);
+    // Use a verified create event (an empty Edit preimage is not trusted).
+    let create = json!({"cwd": cwd, "session_id": "s1", "tool_name": "Write",
+        "tool_input": {"file_path": new, "content": SIDE_TAB_CSS},
+        "tool_response": {"type": "create", "filePath": new, "content": SIDE_TAB_CSS, "originalFile": null}}).to_string();
+    hook::run_hook(&r, &create);
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert_eq!(stop.audit["unknownFindings"], json!(1), "audit retains the full scan");
+    assert!(stop.stdout.contains("[new]"), "{}", stop.stdout);
+    assert!(!stop.stdout.contains("may predate this session"), "{}", stop.stdout);
+}
+
+#[test]
+fn stop_baseline_capped_unknown_does_not_add_notice_to_new_finding() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    t.write(".impeccable/config.json", r#"{"hook":{"limits":{"maxFindings":1}}}"#);
+    let r = rt(&cwd);
+    let new = t.write("new/card.css", SIDE_TAB_CSS);
+    hook::run_hook(&r, &edit_with_original(&cwd, &new, "s1", ".card {}", ".card {}", SIDE_TAB_CSS));
+    let old = t.write("old/card.css", SIDE_TAB_CSS);
+    hook::run_hook(&r, &edit_event(&cwd, &old, "s1"));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert_eq!(stop.audit["unknownFindings"], json!(1));
+    assert!(stop.stdout.contains("[new]"), "{}", stop.stdout);
+    assert!(!stop.stdout.contains("[attribution unknown]"), "{}", stop.stdout);
+    assert!(!stop.stdout.contains("may predate this session"), "{}", stop.stdout);
+}
+
+#[test]
+fn stop_baseline_small_grouped_output_keeps_finding_and_attribution() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    t.write(".impeccable/config.json", r#"{"hook":{"limits":{"maxChars":500}}}"#);
+    let r = rt(&cwd);
+    for path in ["one/card.css", "two/card.css"] {
+        let file = t.write(path, SIDE_TAB_CSS);
+        hook::run_hook(&r, &edit_event(&cwd, &file, "s1"));
+    }
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    let output: Value = serde_json::from_str(&stop.stdout).unwrap();
+    let text = output["hookSpecificOutput"]["additionalContext"].as_str().unwrap();
+    assert!(text.encode_utf16().count() <= 500, "{text}");
+    assert!(text.contains("[side-tab]"), "{text}");
+    assert!(text.contains("[attribution unknown]"), "{text}");
+    assert!(text.contains("may predate this session"), "{text}");
+}
+
+#[test]
+fn stop_baseline_dropped_notice_reclaims_its_rendering_budget() {
+    for max_findings in [1, 5] {
+        let t = Tmp::new();
+        let cwd = t.path();
+        t.write("package.json", "{}");
+        t.write(".impeccable/config.json", &json!({"hook":{"limits":{"maxChars":500,"maxFindings":max_findings}}}).to_string());
+        let r = rt(&cwd);
+        let new = t.write("new/card.css", SIDE_TAB_CSS);
+        hook::run_hook(&r, &edit_with_original(&cwd, &new, "s1", ".card {}", ".card {}", SIDE_TAB_CSS));
+        let old = t.write("old/card.css", SIDE_TAB_CSS);
+        hook::run_hook(&r, &edit_event(&cwd, &old, "s1"));
+        let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+        let output: Value = serde_json::from_str(&stop.stdout).unwrap();
+        let text = output["hookSpecificOutput"]["additionalContext"].as_str().unwrap();
+        let groups: Vec<Group> = [(new, "[new]"), (old, "[attribution unknown]")].into_iter().map(|(file_path, label)| {
+            let mut findings = detector_detect_text(SIDE_TAB_CSS, &file_path, &HookScanOptions::default());
+            for f in &mut findings { f.name = format!("{label} {}", f.name); }
+            Group { file_path, findings }
+        }).collect();
+        let mut config = read_config(&cwd);
+        // Unknown is not displayed at this budget. All available space goes
+        // to the known-new prefix, rather than a discarded notice.
+        config.limits.max_findings = 1.0;
+        let expected = render_grouped_template(&r, &groups, &config, &RenderOpts {
+            cwd: Some(cwd), short_footer: false, reserve_chars: 0.0,
+        });
+        assert_eq!(text, expected, "maxFindings={max_findings}");
+        assert!(text.contains("[new] Side-tab accent border"), "{text}");
+        assert!(!text.contains("may predate this session"), "{text}");
+    }
+}
+
+#[test]
+fn stop_baseline_uses_dirty_worktree_not_git_head() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    t.write("card.css", ".card {}\n");
+    std::fs::create_dir(t.0.join("empty-hooks")).unwrap();
+    let hooks = format!("core.hooksPath={}/empty-hooks", cwd);
+    let git = |args: &[&str]| {
+        let result = std::process::Command::new("git").current_dir(&t.0)
+            .args(["-c", "user.name=Test", "-c", "user.email=test@example.invalid",
+                "-c", "commit.gpgsign=false", "-c", &hooks])
+            .args(args).output().unwrap();
+        assert!(result.status.success(), "{}", String::from_utf8_lossy(&result.stderr));
+    };
+    git(&["init", "--quiet"]);
+    git(&["add", "card.css", "package.json"]);
+    git(&["commit", "--quiet", "-m", "clean baseline"]);
+    // The user introduced this debt before the agent session; HEAD is clean.
+    let before = format!("/* unrelated */\n{SIDE_TAB_CSS}");
+    let file = t.write("card.css", SIDE_TAB_CSS);
+    let r = rt(&cwd);
+    hook::run_hook(&r, &edit_with_original(&cwd, &file, "s1", &before, "/* unrelated */\n", ""));
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert_eq!(stop.audit["preExistingFindings"], json!(1));
+    assert!(stop.stdout.is_empty());
+}
+
+#[test]
+fn stop_baseline_untrusted_shapes_do_not_suppress_findings() {
+    for variant in ["modified", "wrong-path", "no-session", "oversized", "ambiguous", "other-provider"] {
+        let t = Tmp::new();
+        let cwd = t.path();
+        t.write("package.json", "{}");
+        let file = t.write("card.css", SIDE_TAB_CSS);
+        let mut event: Value = serde_json::from_str(&edit_with_original(&cwd, &file, "s1", SIDE_TAB_CSS, SIDE_TAB_CSS, SIDE_TAB_CSS)).unwrap();
+        match variant {
+            "modified" => event["tool_response"]["userModified"] = json!(true),
+            "wrong-path" => event["tool_response"]["filePath"] = json!("another.css"),
+            "no-session" => event["session_id"] = Value::Null,
+            "oversized" => event["tool_response"]["originalFile"] = json!("x".repeat(512 * 1024 + 1)),
+            "ambiguous" => {
+                let repeated = SIDE_TAB_CSS.repeat(2);
+                t.write("card.css", &repeated);
+                event["tool_response"]["originalFile"] = json!(repeated);
+            }
+            _ => {},
+        }
+        let r = if variant == "other-provider" { rt_with(&cwd, env(&[("IMPECCABLE_HOOK_HARNESS", "codex")])) } else { rt(&cwd) };
+        hook::run_hook(&r, &event.to_string());
+        let session = if variant == "no-session" { "unknown" } else { "s1" };
+        let stop = hook::run_stop_hook(&r, &stop_event(&cwd, session));
+        assert!(stop.stdout.contains("[attribution unknown]"), "{variant}: {}", stop.stdout);
+        assert_eq!(stop.audit["preExistingFindings"], json!(0), "{variant}");
+    }
+}
+
+#[test]
+fn stop_baseline_scan_suppression_discards_exemptions() {
+    let t = Tmp::new();
+    let cwd = t.path();
+    t.write("package.json", "{}");
+    let file = t.write("card.css", SIDE_TAB_CSS);
+    let r = rt(&cwd);
+    let event = edit_with_original(&cwd, &file, "s1", SIDE_TAB_CSS, SIDE_TAB_CSS, SIDE_TAB_CSS);
+    for _ in 0..=EDIT_COUNT_THRESHOLD {
+        hook::run_hook(&r, &event);
+    }
+    let stop = hook::run_stop_hook(&r, &stop_event(&cwd, "s1"));
+    assert_eq!(stop.audit["unknownFindings"], json!(1));
+    assert_eq!(stop.audit["preExistingFindings"], json!(0));
+}
+
 fn audit_str<'a>(a: &'a Map<String, Value>, k: &str) -> Option<&'a str> {
     a.get(k).and_then(Value::as_str)
 }

crates/skills/src/bundle.rs

@@ -49,7 +49,7 @@ const RATIO_GUARD_FLOOR: u64 = 1024 * 1024;
 /// through `download_file`, which streams to disk. Reads are capped at
 /// [`MAX_DOWNLOAD_BYTES`]; a longer response is an error, not a truncation.
 pub fn download(url: &str) -> Result<Vec<u8>, String> {
-    let agent = ureq::AgentBuilder::new()
+    let agent = impeccable_context::http::agent_builder()
         .timeout_connect(std::time::Duration::from_secs(30))
         .build();
     match agent.get(url).call() {
@@ -79,7 +79,7 @@ pub struct FetchResponse {
 }
 
 fn ureq_fetch(url: &str) -> Result<FetchResponse, String> {
-    let agent = ureq::AgentBuilder::new()
+    let agent = impeccable_context::http::agent_builder()
         .timeout_connect(std::time::Duration::from_secs(30))
         .timeout(std::time::Duration::from_secs(120))
         .redirects(0)

docs/CLI-CONTRACT.md

@@ -352,6 +352,8 @@ content, and failures fetching either asset exit nonzero, including when
 installed skill or hook files. Explicit `IMPECCABLE_BUNDLE_PATH` and `link`
 retain their local-development trust behavior. See [bundle signing](BUNDLE-SIGNING.md).
 
+**TLS trust (#757):** every HTTPS request the engine makes (bundle and signature downloads, `/api/version`, `/api/commands`, the roll API, image generation) verifies against the OS trust store plus the bundled Mozilla roots, built once in `crates/context/src/http.rs`. `SSL_CERT_FILE` / `SSL_CERT_DIR` replace the OS store, never the bundled roots. The live server calls on localhost are plain HTTP and unaffected.
+
 - **Invoked from**: README.md ("npx impeccable install / update"), README.npm.md Quick Start (`npx impeccable skills install`, `... install -y --providers=claude,codex --scope=project`, `... update`, `... install --no-hooks`, `... link --source=.impeccable --providers=claude,cursor`, `... skills help`), `README.md:360` (hook consent explanation).
 - `run(args)`: `args[0]` ∈ `undefined|help|--help|-h` → `showHelp()`; `install` → `install(rest)`; `link`; `update`; `check` (ignores flags); else `stderr> Unknown skills command: ${sub}` + `Run 'impeccable --help' for available commands.`, `exit 1`.
 - Constants: `API_BASE = 'https://impeccable.style'`; `PROVIDER_DIRS = ['.claude','.cursor','.dsh','.gemini','.agents','.agent','.github','.grok','.hermes','.kiro','.opencode','.pi','.qoder','.trae','.trae-cn','.rovodev','.vibe']`; aliases (`agent`→`.agent`, `agents`/`codex`→`.agents`, `antigravity`→`.agent`, `claude`/`claude-code`→`.claude`, `copilot`/`github`→`.github`, `cursor`, `deepseek`/`deepseek-harness`/`dsh`→`.dsh`, `gemini`, `grok`/`grok-build`/`xai`→`.grok`, `hermes`, `kiro`, `opencode`, `pi`, `qoder`, `rovo-dev`/`rovodev`→`.rovodev`, `trae`, `trae-cn`, `vibe`); leading `.` stripped and lowercased before alias lookup; a literal PROVIDER_DIR value is accepted as-is. `DEFAULT_TARGETS = ['.claude','.agents']`. User-scope skill dir overrides: `.agent`→`~/.gemini/config/skills`, `.dsh`→`$DSH_HOME/skills` (only when DSH_HOME under home) else `~/.dsh/skills`, `.hermes`→`$HERMES_HOME/skills` (only when HERMES_HOME under home) else `~/.hermes/skills`, `.pi`→`~/.pi/agent/skills`, `.opencode`→`$OPENCODE_CONFIG_DIR|$XDG_CONFIG_HOME/opencode|~/.config/opencode` + `/skills`; others `~/<provider>/skills`. Project scope: `<root>/<provider>/skills`.
@@ -1125,6 +1127,16 @@ Candidates in order: `<scripts>/detector/detect-antipatterns.mjs` (built skill l
 
 ---
 
+#### Stop finding attribution (#522)
+
+The Rust Stop pass still scans whole touched files; it does not infer causation from changed line ranges. It suppresses confirmed pre-existing text findings and marks remaining findings `[new]` or `[attribution unknown]`. Unknown findings remain visible, with a reminder not to treat them as regressions or broaden the task without asking. Explicit `detect` scans and per-edit/pre-edit output are unchanged.
+
+A baseline currently requires the first observed edit of that file in a named Claude session to carry a complete `Edit`/`Write` result (`tool_response.originalFile`, `filePath`, and the replacement/content fields). Replaying that result must exactly match the current file. `Write` with `type: "create"` and a null original is an empty baseline; a null original on an update is unknown. No Git/HEAD comparison is used, so existing uncommitted work is part of the baseline. Missing, ambiguous, oversized, mismatched, or user-modified results remain unknown; a later edit cannot establish a missing initial baseline.
+
+The baseline compares pure text-detector findings, independent of line numbers, with multiplicity preserved. It stores only hashed finding identities and counts under the existing session cache's `stopBaseline` field, versioned by schema and engine. A finding observed to disappear loses its exemption, so reintroducing it is new. Capture is capped at 512 KiB per file and 256 findings. Cache eviction or incompatible metadata falls back to unknown, never suppression.
+
+DOM scans, design-system findings, co-scanned stylesheets without their own baseline, and providers/events without a verified preimage remain unknown: other files or earlier edits may affect the result. `[new]` means absent from the verified first-observed-edit baseline, not proof of who caused it. Stop audit output records `preExistingFindings` (suppressed), `newFindings`, and `unknownFindings` before notification deduplication. No permanent ignores or new hook permissions are created.
+
 #### `hook-before-edit.mjs` -> `impeccable hook-before-edit` (Cursor preToolUse write gate)
 
 - **Invoked from**: Cursor project manifest `.cursor/hooks.json`:

plugin/skills/impeccable/SKILL.md

@@ -16,9 +16,11 @@ Core principles:
 
 ## Setup
 
-1. Run `"${CLAUDE_SKILL_DIR}/scripts/impeccable" context` once per session, where `${CLAUDE_SKILL_DIR}` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. Every `"${CLAUDE_SKILL_DIR}/scripts/impeccable" <verb>` command in this skill and its references resolves against that base directory. In reference files, replace the skill-base-dir placeholder with this directory before running commands; it is not a shell variable. On a Windows shell without `sh`, call `"${CLAUDE_SKILL_DIR}/scripts/impeccable.cmd"` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3.
+1. Run `"${CLAUDE_SKILL_DIR}/scripts/impeccable" context` once per session, where `${CLAUDE_SKILL_DIR}` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. Every `"${CLAUDE_SKILL_DIR}/scripts/impeccable" <verb>` command in this skill and its references resolves against that base directory. In reference files, replace the skill-base-dir placeholder with this directory before running commands; it is not a shell variable. On a Windows shell without `sh`, call `"${CLAUDE_SKILL_DIR}/scripts/impeccable.cmd"` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it.
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures.
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work.
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

plugin/skills/impeccable/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

plugin/skills/impeccable/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `"<skill-base-dir>/scripts/impeccable" detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 ## The contract (read once)
 
 Execute in order. No step skipped, no step reordered. Every tool output in live mode may carry an `_instructions` field: it is the authoritative next step for that exact situation, with real ids and paths substituted; when it conflicts with your recollection of this document, `_instructions` wins.

scripts/test-suites.mjs

@@ -52,6 +52,7 @@ export const SUITES = {
           'tests/validate-plugin-manifest.test.js',
           'tests/plugin-paths.test.js',
           'tests/release-engine-workflow.test.js',
+          'tests/workflow-security.test.js',
         ],
       },
       {

skill/SKILL.src.md

@@ -18,9 +18,11 @@ Core principles:
 
 ## Setup
 
-1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `{{scripts_path}}/impeccable <verb>` command in this skill and its references, and `{{scripts_path}}` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `{{scripts_path}}/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. If the launcher is refused, missing, or fails, tell the user before editing that context loading did not run. Read existing **PRODUCT.md** and **DESIGN.md** without inventing missing context, then continue with steps 2–3. <!-- rule:skill-setup-context -->
+1. Run `<skill-base-dir>/scripts/impeccable context` once per session, where `<skill-base-dir>` is the directory that contains this SKILL.md (the skill folder, not a plugin root two levels above it); keep cwd at the user's project. That base directory resolves every `{{scripts_path}}/impeccable <verb>` command in this skill and its references, and `{{scripts_path}}` is the fallback only when the runtime reports no base directory. On a Windows shell without `sh`, call `{{scripts_path}}/impeccable.cmd` instead. The launcher runs a self-contained binary that ships next to it or is downloaded once on first run; no Node or other runtime is required. Pass a named source file or route as `--target <path>`. It loads PRODUCT.md, DESIGN.md, the matching surface brief, and native-platform guidance when applicable; follow its directives and do not rerun it. <!-- rule:skill-setup-context -->
 2. Load the request's playbook: its Commands-table reference for an explicit/implied sub-command, or [reference/new-work.md](reference/new-work.md) for a new surface or replacement visual world. Inspect target and incumbent visual truth before editing. When the app cannot run, start with committed visual-regression goldens or screenshot fixtures; verify target and freshness against current tokens, CSS, components, or assets, resolve conflicts, and compare theme/variant captures. <!-- rule:skill-setup-command-ref --> <!-- rule:skill-setup-read-project -->
-3. After analysis and direction are resolved, load [reference/craft-floor.md](reference/craft-floor.md) immediately before editing UI. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work. <!-- rule:skill-craft-floor-load -->
+3. After resolving analysis and direction, read [reference/craft-floor.md](reference/craft-floor.md) immediately before any UI edit, including small refinements. It carries the quality floor, the absolute bans, and the reflexes no detector catches. Do not load it for planning-only work. <!-- rule:skill-craft-floor-load -->
+
+**Launcher unavailable:** If refused, missing, or failed, **first send the user a message** that context loading did not run. Then read existing PRODUCT.md and DESIGN.md without inventing missing context, follow the applicable steps 2–3, and perform the requested work through permitted tools. Launcher failure alone does not block otherwise-permitted edits.
 
 ## How to design
 

skill/reference/live-setup.md

@@ -42,6 +42,8 @@ On every boot the project is scanned for HTML files under common page roots (`pu
 
 ## CSP detection (first-time only)
 
+Keep all allowances below development-only, including manual middleware and meta-tag edits. Do not change a deployed production site's CSP to load the localhost helper; see [live.md](live.md) for production inspection alternatives.
+
 If `config.cspChecked === true`, skip this whole section; the user was already asked once.
 
 ```bash

skill/reference/live.md

@@ -4,6 +4,8 @@ Interactive live variant mode: select elements in the browser, pick a design act
 
 A running dev server with HMR (Vite, Next.js, Bun, etc.), OR a static HTML file open in the browser. If the dev server's default port is busy, the app is very likely ALREADY running; probe the default URL before spawning a second server.
 
+Live editing requires a local checkout; injection into deployed production sites (including HTTPS) is unsupported. For production inspection, use `{{scripts_path}}/impeccable detect <url>` or the browser extension, not the live helper. Do not disable browser security or weaken production CSP to enable live mode.
+
 <codex>
 Codex: run live helper commands, the app dev server, and any dependency-installing setup with `sandbox_permissions: "require_escalated"` from the start; live mode depends on localhost and package-manager network access that the sandbox blocks.
 </codex>

tests/oracle/cases/hooks.mjs

@@ -17,6 +17,40 @@ const claudeEdit = (file, extra = {}) => ({
 const stop = (extra = {}) => ({ session_id: 's1', cwd: WS, hook_event_name: 'Stop', stop_hook_active: false, ...extra });
 
 export default [
+  {
+    id: 'hook-stop-baseline-new-finding', workspace: 'hook-project', files: CACHE_FILES,
+    normalize: [['("stopBaseline":\\{"version":1,"engine":")[^"]+', 'g', '$1<ENGINE_VERSION>']],
+    setup(ws) {
+      fs.writeFileSync(`${ws}/src/new.css`, '.card { border-left: 4px solid #6366f1; border-radius: 8px; }\n');
+    },
+    steps: [
+      { verb: 'hook', stdin: claudeEdit('src/new.css', {
+        tool_name: 'Write',
+        tool_response: {
+          type: 'create', filePath: `${WS}/src/new.css`, originalFile: null,
+          content: '.card { border-left: 4px solid #6366f1; border-radius: 8px; }\n',
+        },
+      }) },
+      { verb: 'hook', stdin: stop() },
+    ],
+  },
+  {
+    id: 'hook-stop-baseline-import-only', workspace: 'hook-project', files: CACHE_FILES,
+    normalize: [['("stopBaseline":\\{"version":1,"engine":")[^"]+', 'g', '$1<ENGINE_VERSION>']],
+    setup(ws) {
+      fs.writeFileSync(`${ws}/src/report.ts`, "const report = `<style>body { font-family: Fraunces; }</style>`;\n");
+    },
+    steps: [
+      { verb: 'hook', stdin: claudeEdit('src/report.ts', {
+        tool_response: {
+          filePath: `${WS}/src/report.ts`,
+          originalFile: "import dead from 'dead';\nconst report = `<style>body { font-family: Fraunces; }</style>`;\n",
+          oldString: "import dead from 'dead';\n", newString: '', replaceAll: false, userModified: false,
+        },
+      }) },
+      { verb: 'hook', stdin: stop() },
+    ],
+  },
   // --- hook.mjs: per-edit ---
   { id: 'hook-edit-tsx-fresh', verb: 'hook', workspace: 'hook-project', stdin: claudeEdit('src/components/Card.tsx'), files: CACHE_FILES },
   { id: 'hook-edit-css-fresh', verb: 'hook', workspace: 'hook-project', stdin: claudeEdit('src/components/Card.module.css'), files: CACHE_FILES },

tests/oracle/golden/hook-session-codex-stop-decision.json

@@ -7,7 +7,7 @@
       "signal": null
     },
     {
-      "stdout": "{\"decision\":\"block\",\"reason\":\"[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (1 issue(s)):\\n- L1 [side-tab] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}",
+      "stdout": "{\"decision\":\"block\",\"reason\":\"[impeccable@1] Findings marked attribution unknown may predate this session; do not treat them as regressions or broaden the task without asking.\\n\\n[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (1 issue(s)):\\n- L1 [side-tab] [attribution unknown] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}",
       "stderr": "",
       "exit": 0,
       "signal": null

tests/oracle/golden/hook-session-fresh-then-pending-then-stop.json

@@ -25,7 +25,7 @@
       "signal": null
     },
     {
-      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"Stop\",\"additionalContext\":\"[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (1 issue(s)):\\n- L1 [side-tab] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}}",
+      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"Stop\",\"additionalContext\":\"[impeccable@1] Findings marked attribution unknown may predate this session; do not treat them as regressions or broaden the task without asking.\\n\\n[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (1 issue(s)):\\n- L1 [side-tab] [attribution unknown] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}}",
       "stderr": "",
       "exit": 0,
       "signal": null

tests/oracle/golden/hook-session-grok-edit-then-stop.json

@@ -7,7 +7,7 @@
       "signal": null
     },
     {
-      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"Stop\",\"additionalContext\":\"[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (2 issue(s)):\\n- L1 [side-tab] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n- L2 [gradient-text] Gradient text. Gradient text is decorative rather than meaningful — a common AI tell, especially on headings and metrics. Use solid colors for text.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}}",
+      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"Stop\",\"additionalContext\":\"[impeccable@1] Findings marked attribution unknown may predate this session; do not treat them as regressions or broaden the task without asking.\\n\\n[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (2 issue(s)):\\n- L1 [side-tab] [attribution unknown] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n- L2 [gradient-text] [attribution unknown] Gradient text. Gradient text is decorative rather than meaningful — a common AI tell, especially on headings and metrics. Use solid colors for text.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}}",
       "stderr": "",
       "exit": 0,
       "signal": null

tests/oracle/golden/hook-session-two-sessions.json

@@ -13,7 +13,7 @@
       "signal": null
     },
     {
-      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"Stop\",\"additionalContext\":\"[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (1 issue(s)):\\n- L1 [side-tab] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}}",
+      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"Stop\",\"additionalContext\":\"[impeccable@1] Findings marked attribution unknown may predate this session; do not treat them as regressions or broaden the task without asking.\\n\\n[impeccable@1] Design hook findings requiring review in src/components/Card.module.css (1 issue(s)):\\n- L1 [side-tab] [attribution unknown] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n\\nTriage per the session policy: fix real problems; persist confident false-positive or sanctioned-exception ignores via `impeccable hooks ignore-value` and disclose them in your reply; unsure, ask in one line.\"}}",
       "stderr": "",
       "exit": 0,
       "signal": null

tests/oracle/golden/hook-stop-baseline-import-only.json

@@ -0,0 +1,19 @@
+{
+  "steps": [
+    {
+      "stdout": "",
+      "stderr": "",
+      "exit": 0,
+      "signal": null
+    },
+    {
+      "stdout": "",
+      "stderr": "",
+      "exit": 0,
+      "signal": null
+    }
+  ],
+  "files": {
+    ".impeccable/hook.cache.json": "{\"version\":1,\"sessions\":{\"s1\":{\"updatedAt\": <EPOCH>,\"files\":{\"<WS>/src/report.ts\":{\"editCount\":1,\"findings\":[\"overused-font:1:fraunces\"],\"stopBaseline\":{\"version\":1,\"engine\":\"<ENGINE_VERSION>\",\"counts\":{\"57802aad69167e30e1dc001298a4ad9552f238c951be5b990b3071d6e2bdf827\":1}}}}}}}"
+  }
+}

tests/oracle/golden/hook-stop-baseline-new-finding.json

@@ -0,0 +1,19 @@
+{
+  "steps": [
+    {
+      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"PostToolUse\",\"additionalContext\":\"[impeccable@1] Design hook scanned src/new.css. No deterministic design-quality issues found. That does not mean the design is good: keep following the project design system and the impeccable skill guidance.\"}}",
+      "stderr": "",
+      "exit": 0,
+      "signal": null
+    },
+    {
+      "stdout": "{\"hookSpecificOutput\":{\"hookEventName\":\"Stop\",\"additionalContext\":\"[impeccable@1] Design hook findings requiring review in src/new.css (1 issue(s)):\\n- L1 [side-tab] [new] Side-tab accent border. Thick colored border on one side of a card — the most recognizable tell of AI-generated UIs. Use a subtler accent or remove it entirely.\\n\\nTriage each finding, then state in your reply what you fixed, what you suppressed, and what you left standing:\\n- Real design problem: fix it. Keep intentional design as designed.\\n- Confident false positive or sanctioned exception (an intentional demo or fixture, documentation of bad design, literal or domain-appropriate motion, a choice the user confirmed): persist the narrowest ignore yourself and disclose it. Run `<HOOK_ADMIN_CMD> ignore-value <rule> \\\"<value>\\\" --reason \\\"<who decided: evidence>\\\"` with the pair shown on the finding line, or value \\\"*\\\" plus `--file <path>` when the line shows none. Write \\\"user confirmed\\\" in a reason only when the user did.\\n- Unsure: leave it as is and ask the user in one line.\\nSelf-serve ends at ignore-value: `ignore-file` and `ignore-rule` need the user's explicit approval, and never add an ignore to push a blocked write through. Full suppression ladder: /impeccable hooks.\"}}",
+      "stderr": "",
+      "exit": 0,
+      "signal": null
+    }
+  ],
+  "files": {
+    ".impeccable/hook.cache.json": "{\"version\":1,\"sessions\":{\"s1\":{\"updatedAt\": <EPOCH>,\"files\":{\"<WS>/src/new.css\":{\"editCount\":1,\"findings\":[\"side-tab:1\"],\"stopBaseline\":{\"version\":1,\"engine\":\"<ENGINE_VERSION>\",\"counts\":{}},\"cleanAcked\":true}},\"footerShown\":true}}}"
+  }
+}

tests/skill-behavior-harness.test.mjs

@@ -2,7 +2,66 @@ import { it } from 'node:test';
 import assert from 'node:assert/strict';
 import fs from 'node:fs';
 import path from 'node:path';
-import { prepareWorkspace, cleanupWorkspace, makeTools } from './skill-behavior/harness.mjs';
+import { MockLanguageModelV3 } from 'ai/test';
+import { prepareWorkspace, cleanupWorkspace, makeTools, runTurn, SKILL_BODY } from './skill-behavior/harness.mjs';
+import { assertPlanningFallbackWarning } from './skill-behavior/assertions.mjs';
+
+it('planning fallback requires an assistant warning between the denial and context reads', () => {
+  const call = { role: 'assistant', content: [{ type: 'tool-call', toolCallId: 'context', toolName: 'bash', input: { command: '.claude/skills/impeccable/scripts/impeccable context' } }] };
+  const denial = { role: 'tool', content: [{ type: 'tool-result', toolCallId: 'context', toolName: 'bash', output: { type: 'text', value: 'Error: Bash permission denied by the host. This command was not executed.' } }] };
+  const warning = { role: 'assistant', content: 'Context loading did not run because the launcher was denied.' };
+  const read = { role: 'assistant', content: [{ type: 'tool-call', toolCallId: 'read', toolName: 'read', input: { path: 'PRODUCT.md' } }] };
+  assert.doesNotThrow(() => assertPlanningFallbackWarning([call, denial, warning, read]));
+  assert.doesNotThrow(() => assertPlanningFallbackWarning([call, denial, { role: 'assistant', content: [{ type: 'text', text: warning.content }, ...read.content] }]));
+  for (const messages of [
+    [call, denial, read], // Silent continuation.
+    [call, denial, read, warning], // Final-only disclosure.
+    [warning, call, denial, read], // Not a response to the actual denial.
+    [call, denial, { ...warning, role: 'user' }, read],
+    [call, { ...denial, content: [{ ...denial.content[0], toolCallId: 'unrelated' }] }, warning, read],
+  ]) {
+    assert.throws(() => assertPlanningFallbackWarning(messages), assert.AssertionError);
+  }
+});
+
+it('DeepSeek gets an explicit output ceiling instead of the compatibility SDK default', async () => {
+  const workspace = prepareWorkspace();
+  try {
+    for (const modelId of ['deepseek-v4-flash', 'claude-sonnet-5']) {
+      const model = new MockLanguageModelV3({
+        modelId,
+        doGenerate: {
+          content: [{ type: 'text', text: 'done' }],
+          finishReason: { unified: 'stop', raw: 'stop' },
+          usage: { inputTokens: { total: 1 }, outputTokens: { total: 1 } },
+          warnings: [],
+        },
+      });
+      await runTurn({ workspace, model, userPrompt: 'Test the harness.', maxSteps: 1 });
+      const request = model.doGenerateCalls[0];
+      assert.equal(request.maxOutputTokens, modelId.startsWith('deepseek-') ? 16_384 : undefined);
+      assert.ok(request.prompt.some((message) => message.role === 'system' && message.content === SKILL_BODY));
+    }
+  } finally {
+    cleanupWorkspace(workspace);
+  }
+});
+
+it('loaded-skill metadata resolves to the staged launcher and readable references', async () => {
+  const workspace = prepareWorkspace();
+  try {
+    const baseDir = SKILL_BODY.match(/^Base directory for this skill \(workspace-relative\): (.+)$/m)?.[1];
+    assert.ok(baseDir, 'the host must supply the skill directory separately from its instructions');
+    assert.ok(fs.statSync(path.join(workspace, baseDir, 'scripts/impeccable')).isFile());
+    const { tools, trace } = makeTools(workspace, {}, {}, { denyBash: true });
+    await tools.read.execute({ path: `${baseDir}/reference/polish.md` });
+    await tools.read.execute({ path: `${baseDir}/reference/craft-floor.md` });
+    assert.ok(trace.toolCalls.every((call) => call.succeeded));
+    assert.ok(SKILL_BODY.includes('<skill-base-dir>/scripts/impeccable context'), 'metadata must not rewrite away the path-resolution behavior under test');
+  } finally {
+    cleanupWorkspace(workspace);
+  }
+});
 
 it('denied-launcher tools reject every shell attempt without executing or modifying the skill', async () => {
   const workspace = prepareWorkspace({ files: { 'index.html': 'before' } });

tests/skill-behavior/README.md

@@ -75,7 +75,7 @@ The trace is the source of truth, not the model's free-form reply.
 | 16 | existing surface, with and without PRODUCT.md; asks where to start | loads `routing.md`, delivers advice, and does not edit project files, start an interview, archive a critique, or run menu scans |
 | 17 | existing surface; asks whether critique is required before polish | loads `routing.md` and both command references, then delivers advice without executing the playbooks |
 | 18 | existing surface; explicitly requests polish followed by a next-command recommendation | loads `polish.md` rather than substituting workflow advice for the requested work |
-| 19 | tiny spacing edit with PRODUCT.md + DESIGN.md; Bash denied, plus a real-loader success control | actually reads playbook and craft floor before editing; denial also requires direct context-file reads and a user-visible warning before the edit |
+| 19 | tiny spacing edit with PRODUCT.md + DESIGN.md; Bash denied, a real-loader success control, and a denied-launcher planning-only case | edits require successful playbook/craft-floor reads and a pre-edit denial warning; planning stays read-only and skips craft-floor |
 
 ## Setup launcher-failure branch (2026-09-06, PR #750)
 
@@ -98,6 +98,67 @@ summary. This does not reproduce the reporter's complete reference-loading
 failure or establish a multi-provider pass. An earlier scenario 6 result used
 attempt-based reference assertions and is not counted as a success control.
 
+### Refusal follow-up (2026-09-06, #744)
+
+The provider-neutral harness models a loaded skill with a known base directory
+using synthetic workspace-relative host metadata, not each provider's exact
+generated prompt. The source instructions and `<skill-base-dir>` resolution
+remain under test; reference reads still have to succeed. Provider transforms
+and plugin loading have separate path/loader tests. The harness also sets an
+explicit 16,384-token response ceiling for DeepSeek: the Anthropic-compatible
+SDK otherwise treats that model as unknown and caps it at 4,096. Truncation
+still fails the scenario; this changes the test runner, not the shipped skill.
+
+The unchanged-source baseline with directory metadata passed 5/8 focused
+cases: Sonnet skipped craft-floor in its successful-launcher control, OpenAI
+stopped without editing after denial, and Gemini warned only after editing.
+DeepSeek passed both cases. An initial candidate got OpenAI to edit but still
+warned late on Sonnet, OpenAI, and Gemini; DeepSeek's denial response truncated.
+All four successful-launcher controls passed that candidate.
+
+The pre-review candidate separates the fallback from the long first step, says to
+send the warning first and continue through permitted tools, and clarifies
+that craft-floor also applies to small refinements. Setup grows by 18
+whitespace-separated words; the description is unchanged. Sonnet and OpenAI
+passed both final cases, as did DeepSeek with the explicit output ceiling.
+Gemini still warned after the edit; its control passed. The final result is
+7/8; the warning-order assertion remains unchanged.
+
+Review follow-up: the fallback now says to follow the **applicable** steps
+2–3, preserving step 3's planning-only exclusion (20 added Setup words overall).
+A new denied-launcher planning case requires a real plan, no mutations or
+craft-floor read, and successful context/target/playbook reads. On Sonnet,
+the editing denial and successful-loader cases both passed again. The planning
+run stayed read-only and skipped craft-floor, but failed because it did not
+read `polish.md`. That assertion remains intact: this is another reference-loading
+gap under #744, not a green planning result. Other providers were not rerun for
+this wording-only review clarification.
+
+The planning case also checks the response-message sequence: the launcher must
+actually be denied, then an assistant warning must precede the first fallback
+PRODUCT.md or DESIGN.md read. Deterministic tests reject silent continuation,
+final-only warnings, warnings before denial, and user-authored warnings. This
+checks disclosure even when no editing occurs; the editing cases retain their
+existing pre-edit warning assertion.
+
+One focused Sonnet rerun with this guard read the playbook and produced a
+read-only plan without craft-floor, but omitted the launcher warning entirely.
+The strengthened assertion correctly failed that run; #744 remains open for
+the behavior failure rather than treating this coverage fix as a skill fix.
+
+These are single samples per case and candidate, not reliability estimates.
+This API harness starts with the skill loaded and readable references. It
+does not measure activation, reproduce Windows command parsing, or establish
+fallback behavior when the host also denies required file reads or writes.
+Keep #744 open; evaluate activation separately with #375.
+
+To repeat only these cases (provider keys and an engine binary required):
+
+```sh
+IMPECCABLE_SKILL_BEHAVIOR_MODELS=claude-sonnet-5,gpt-5.6-terra,gemini-3.7-flash,deepseek-v4-flash \
+  node --test --test-name-pattern='scenario 19:' tests/skill-behavior/scenarios.test.mjs
+```
+
 ## Workflow-advice baseline (2026-09-05, PR #737)
 
 The four cases in scenarios 16-18 are new; prior scenario results do not

tests/skill-behavior/assertions.mjs

@@ -0,0 +1,24 @@
+import assert from 'node:assert/strict';
+
+export const LAUNCHER_FAILURE_WARNING = /(?:context|launcher|bash)[^.!?\n]{0,160}(?:denied|refused|unavailable|blocked|could(?:n't| not)|cannot|can't|did(?:n't| not)|fail|unable)|(?:denied|refused|unavailable|blocked|could(?:n't| not)|cannot|can't|unable)[^.!?\n]{0,160}(?:context|launcher|bash)/i;
+
+export function assertPlanningFallbackWarning(responseMessages) {
+  const blocks = responseMessages.flatMap((message) =>
+    (typeof message.content === 'string' ? [{ type: 'text', text: message.content }] : message.content)
+      .map((block) => ({ ...block, role: message.role })),
+  );
+  const contextCalls = new Set(blocks.filter((block) => block.role === 'assistant'
+    && block.type === 'tool-call' && block.toolName === 'bash'
+    && /impeccable\s+context\b/.test(block.input?.command ?? '')).map((block) => block.toolCallId));
+  const denialIndex = blocks.findIndex((block) => block.role === 'tool'
+    && block.type === 'tool-result' && contextCalls.has(block.toolCallId)
+    && block.output?.type === 'text' && /Bash permission denied by the host/.test(block.output.value));
+  assert.ok(denialIndex >= 0, 'must observe the context launcher denial in the response sequence');
+  const warningIndex = blocks.findIndex((block, index) => index > denialIndex
+    && block.role === 'assistant' && block.type === 'text' && LAUNCHER_FAILURE_WARNING.test(block.text));
+  const contextReadIndex = blocks.findIndex((block, index) => index > denialIndex
+    && block.role === 'assistant' && block.type === 'tool-call' && block.toolName === 'read'
+    && /(?:^|\/)(?:PRODUCT|DESIGN)\.md$/.test(block.input?.path ?? ''));
+  assert.ok(warningIndex > denialIndex && contextReadIndex > warningIndex,
+    'planning fallback must warn after denial and before reading project context, not only in the final response');
+}

tests/skill-behavior/harness.mjs

@@ -86,7 +86,12 @@ function loadSkillBody() {
   return md.trim();
 }
 
-export const SKILL_BODY = loadSkillBody();
+// This provider-neutral fixture assumes a loaded skill with a known base
+// directory, not an exact copy of each host's transformed prompt. Claude's
+// loader supplies a base-directory prefix; here it is workspace-relative
+// because the file tools reject absolute paths. Provider rewrite/loader
+// contracts are tested separately, not established by these behavior cases.
+export const SKILL_BODY = `Base directory for this skill (workspace-relative): .claude/skills/impeccable\n\n${loadSkillBody()}`;
 
 /**
  * Create a temp workspace and prepopulate it.
@@ -404,6 +409,10 @@ export async function runTurn({ workspace, model, userPrompt, priorMessages = []
       // Real client-side deadline on the provider call: without it a stalled
       // stream wedges the whole sweep with no tally.
       abortSignal: controller.signal,
+      // The Anthropic-compatible adapter does not recognize DeepSeek and
+      // otherwise caps each response at 4096 tokens, truncating valid tool
+      // continuations. Keep an explicit ceiling; length remains a test failure.
+      maxOutputTokens: model?.modelId?.startsWith('deepseek-') ? 16_384 : undefined,
       // Resolved from the model object so the 21 runTurn call sites stay
       // unchanged. Reasoning models run at the provider default otherwise,
       // which is not the tier this suite is meant to measure.

tests/skill-behavior/scenarios.test.mjs

@@ -27,6 +27,7 @@ import {
   ENGINE_MISSING_MESSAGE,
 } from './harness.mjs';
 import { detectProvider, getModel, hasKey, resolveModelList, PROVIDERS } from './providers.mjs';
+import { assertPlanningFallbackWarning, LAUNCHER_FAILURE_WARNING } from './assertions.mjs';
 import {
   PRODUCT_MD_SAMPLE,
   PRODUCT_MD_SAMPLE_NO_REGISTER,
@@ -682,10 +683,9 @@ for (const modelId of resolveModelList()) {
             const readIndex = trace.toolCalls.findIndex((call) => call.name === 'read' && call.succeeded && (call.input.path === filename || call.input.path.endsWith(`/${filename}`)));
             assert.ok(readIndex >= 0 && readIndex < writeIndex, `${filename} must actually be read before editing`);
           }
-          const warning = /(?:context|launcher|bash)[^.!?\n]{0,160}(?:denied|refused|unavailable|blocked|could(?:n't| not)|cannot|can't|did(?:n't| not)|fail|unable)|(?:denied|refused|unavailable|blocked|could(?:n't| not)|cannot|can't|unable)[^.!?\n]{0,160}(?:context|launcher|bash)/i;
           const assistantBlocks = responseMessages.filter((message) => message.role === 'assistant')
             .flatMap((message) => typeof message.content === 'string' ? [{ type: 'text', text: message.content }] : message.content);
-          const warningIndex = assistantBlocks.findIndex((block) => block.type === 'text' && warning.test(block.text));
+          const warningIndex = assistantBlocks.findIndex((block) => block.type === 'text' && LAUNCHER_FAILURE_WARNING.test(block.text));
           const writeBlockIndex = assistantBlocks.findIndex((block) => block.type === 'tool-call' && block.toolName === 'write');
           if (denyBash) assert.ok(warningIndex >= 0 && writeBlockIndex > warningIndex, 'must disclose the failed context launcher before editing, not only in the final summary');
           assert.ok(!trace.toolCalls.some((call) => call.mutatedPaths.some((p) => /(?:^|\/)(?:PRODUCT|DESIGN)\.md$/.test(p))), 'must not fabricate or replace project context');
@@ -695,6 +695,34 @@ for (const modelId of resolveModelList()) {
       });
     }
 
+    it('scenario 19: denied launcher keeps planning-only work read-only without craft-floor', async () => {
+      const workspace = prepareWorkspace({ files: {
+        'PRODUCT.md': PRODUCT_MD_SAMPLE,
+        'DESIGN.md': DESIGN_MD_SAMPLE,
+        'index.html': '<!doctype html><html><body><button style="padding:2px 4px">New note</button></body></html>',
+      } });
+      try {
+        const { trace, text, stepTexts, finishReason, responseMessages } = await runTurn({
+          workspace,
+          model,
+          userPrompt: '/impeccable polish index.html. Inspect the button spacing and propose a short plan only. Do not edit any files or implement the plan yet.',
+          maxSteps: 12,
+          denyBash: true,
+        });
+        logTrace('S19', 'denied-launcher-planning', modelId, trace, { finishReason, text: stepTexts.join('\n') });
+        assert.notEqual(finishReason, 'length', 'a truncated response is not a completed plan');
+        assert.ok(trace.toolCalls.some((call) => call.name === 'bash' && call.denied && /impeccable\s+context\b/.test(call.input.command)), 'must encounter an actual denied context attempt');
+        assert.deepEqual(readsMatching(trace, 'craft-floor.md'), [], 'planning-only work must not load the editing floor');
+        assertAdviceOnly(trace, text);
+        assertPlanningFallbackWarning(responseMessages);
+        for (const filename of ['PRODUCT.md', 'DESIGN.md', 'index.html', 'reference/polish.md']) {
+          assert.ok(trace.toolCalls.some((call) => call.name === 'read' && call.succeeded && (call.input.path === filename || call.input.path.endsWith(`/${filename}`))), `${filename} must actually be read`);
+        }
+      } finally {
+        cleanupWorkspace(workspace);
+      }
+    });
+
     it('scenario 18: explicit command request takes precedence over workflow advice', async () => {
       const workspace = prepareWorkspace({ files: WORKFLOW_ADVICE_FILES });
       try {

tests/workflow-security.test.js

@@ -0,0 +1,35 @@
+import { describe, expect, test } from 'bun:test';
+import { readFileSync, readdirSync } from 'node:fs';
+
+const directory = new URL('../.github/workflows/', import.meta.url);
+const workflows = Object.fromEntries(readdirSync(directory)
+  .filter(name => /\.ya?ml$/.test(name))
+  .map(name => [name, Bun.YAML.parse(readFileSync(new URL(name, directory), 'utf8'))]));
+
+describe('workflow execution boundaries', () => {
+  test('repository actions are pinned to full commit SHAs', () => {
+    for (const [name, workflow] of Object.entries(workflows)) {
+      for (const [jobName, job] of Object.entries(workflow.jobs)) {
+        for (const step of job.steps || []) {
+          if (!step.uses || step.uses.startsWith('./')) continue;
+          expect(step.uses, `${name}: ${jobName}`).toMatch(/@[a-f0-9]{40}$/);
+        }
+      }
+    }
+  });
+
+  test('CI uses a read-only repository token without job-level escalation', () => {
+    const ci = workflows['ci.yml'];
+    expect(ci.permissions).toEqual({ contents: 'read' });
+    for (const job of Object.values(ci.jobs)) {
+      expect(job.permissions).toBeUndefined();
+    }
+  });
+
+  test('generated-output sync and sheriff retain their required write access', () => {
+    expect(workflows['sync-generated-output.yml'].permissions).toEqual({ contents: 'write' });
+    expect(workflows['sheriff.yml'].permissions).toEqual({
+      actions: 'read', checks: 'read', contents: 'read', issues: 'write', 'pull-requests': 'write',
+    });
+  });
+});