d2c-audit（d2c 审核）

Design Token Drift Audit

You are auditing this project's codebase to find violations of the design system. Your job is to detect hardcoded values that MUST use design tokens, find unused tokens, and report components that bypass the design system. Every finding is a violation of a non-negotiable rule — treat the tokens file as authoritative.

Non-negotiables

These rules hold across every phase of this skill. No exceptions.

1. Design tokens MUST be loaded before any decision. Read .claude/d2c/design-tokens.json. If it is missing, unreadable, or has d2c_schema_version < 1, STOP AND ASK the user to run /d2c-init (or /d2c-init --force if outdated).
2. NEVER use a library outside preferred_libraries.<category>.selected. The user explicitly chose which library to use for each capability. NEVER substitute an installed-but-not-selected library. If the design requires a capability not covered by preferred_libraries, STOP AND ASK.
3. NEVER hardcode color, spacing, typography, shadow, or radius values. Every visual value MUST reference a design token from design-tokens.json. No raw hex, no magic numbers, no exceptions.
4. MUST reuse existing components when an existing component can serve the need. Check the components array in design-tokens.json before creating anything new. If an existing component can do the job, MUST use it.
5. MUST follow project conventions when confidence > 0.6 and value ≠ "mixed". Project conventions (declaration style, export style, type definitions, import ordering, file naming, CSS wrapper, barrel exports, props pattern) override framework defaults.
6. NEVER re-decide a locked component or token. Read decisions.lock.json from the IR run directory at the start of every phase after Phase 2. Only nodes with status: "failed" may have their component choice or token mapping changed. If a locked decision must change, STOP AND ASK.

When any rule is ambiguous, STOP AND ASK — do not guess.

Note for d2c-audit: These are the rules whose violations this skill reports. Every audit finding is a violation of one of the rules above. When --fix is passed, fixes MUST be verifiable against these rules before being applied — STOP AND ASK on any fix that is not a deterministic, token-exact match.

Arguments

Parse $ARGUMENTS for optional flags:

• Directory path — If a path is provided (e.g., /d2c-audit src/components/), scope all audits to files within that directory only. If no path is provided, audit the entire src/ and app/ directories as usual.
• --fix — After reporting, auto-fix hardcoded values that have an exact token match (labeled definite in Audit 1). Only fixes exact matches — undocumented values are reported but not touched. Show a summary of fixes applied after the report. When combined with --dry-run, shows a preview of proposed fixes instead of applying them (see below).
• --dry-run — Run the full audit and generate the report, but do NOT apply any changes. Behavior depends on whether --fix is also provided:

- --dry-run alone (without --fix): Runs the audit as read-only and generates the report. This is the same as the default behavior but makes the read-only intent explicit. - --dry-run with --fix: Runs the full audit, identifies all fixable issues, but instead of auto-fixing, presents each proposed fix as a diff-style preview: --- src/components/Header.tsx (current) +++ src/components/Header.tsx (proposed) @@ line 42 @@ - color: #3B82F6 + color: theme.colors.primary After showing all proposed fixes, ask the user how to proceed: - "Apply all" — Apply every proposed fix. - "Review individually" — Step through each fix one by one. For each, the user can approve or skip. - "Skip all" — Discard all proposed fixes and end the audit. Group fixes by file for easier review. Show the total count: "Found X fixable issues across Y files."

Pre-flight Check

1. Check that .claude/d2c/design-tokens.json exists. If it doesn't, tell the user to run /d2c-init first and stop.
2. Schema version check: Read the d2c_schema_version field. If it is missing or less than 1 (the current version), STOP AND ASK the user: "design-tokens.json uses schema version {version or 'none'} but the current version is 1. You MUST run /d2c-init --force to regenerate before this audit can produce reliable results. Continue with the outdated schema anyway, or abort?" NEVER proceed past this check without an explicit user decision.
3. Read .claude/d2c/design-tokens.json fully into context.

Split file support: After reading design-tokens.json, check the split_files field. If true, load only the split files relevant to each audit instead of keeping the full file in context:

Load each split file only when starting its corresponding audit. When split_files: true, the split files are the source of truth — the monolithic file is a lightweight pointer. If a split file is missing, warn the user to run /d2c-init --force to regenerate.

Audit 1: Hardcoded Values That Must Be Tokens

Scan all component and source files for hardcoded values that exist in the design tokens or MUST be using them. Read the framework field from design-tokens.json to determine which file extensions to scan:

• react / solid / qwik: .tsx, .jsx, .ts, .js
• vue: .vue files (scan both <template> and <style> blocks), .ts, .js
• svelte: .svelte files (scan both markup and <style> blocks), .ts, .js
• angular: .component.ts, .component.html, .ts, .js
• astro: .astro files (scan both template and <style> blocks), .ts, .js

Also scan style files: .css, .scss, .module.css, .module.scss.

Colors

• Search for raw hex colors (#[0-9a-fA-F]{3,8}), rgb(, rgba(, hsl(, hsla( in component files and style files.
• Exclude: SVG files, image assets, config files (tailwind.config.*, postcss.config.*), node_modules, .next, test files.
• Color comparison rule (exact match only): Convert all found colors to lowercase 6-digit hex before comparison. Convert rgb() and hsl() values to hex. A match is exact hex equality only — no fuzzy matching, no "near-equivalent" matching.
• For each hardcoded color found, check if it exactly matches any value in design-tokens.json → colors. If yes, label it definite — this value MUST use the token.
• Also flag colors that do NOT exactly match any token value — label these undocumented.

Spacing

• Search for hardcoded pixel values in padding, margin, gap, width, height properties: padding: 16px, margin: 24px, gap: 8px, p-[16px], m-[24px], etc.
• In Tailwind projects, look for arbitrary value brackets: p-[...], m-[...], gap-[...], w-[...], h-[...] with pixel values that match the spacing scale.
• Flag values that match the spacing scale but use raw numbers instead of tokens.

Typography

• Search for hardcoded font sizes: font-size: 14px, text-[14px], etc.
• Search for hardcoded font weights: font-weight: 600, font-[600], etc.
• Search for hardcoded line heights in style declarations.
• Flag values that match the typography scale but use raw numbers instead of tokens.

Shadows & Borders

• Search for hardcoded box-shadow values and border-radius values.
• Flag values that match tokens but are hardcoded.

Audit 2: Unused Tokens

Check each token defined in design-tokens.json to see if it's actually used in the codebase.

Search patterns by styling approach (deterministic):

• For Tailwind projects — color tokens: Search for each color token name prefixed with bg-, text-, border-, ring-, fill-, stroke-, decoration-, accent-, outline-, shadow-. A color token is "used" if ANY of these prefixed forms appear in source files.
• For Tailwind projects — spacing tokens: Search for each spacing token name prefixed with p-, px-, py-, pt-, pr-, pb-, pl-, m-, mx-, my-, mt-, mr-, mb-, ml-, gap-, gap-x-, gap-y-, w-, h-, space-x-, space-y-, inset-, top-, right-, bottom-, left-. A spacing token is "used" if ANY of these prefixed forms appear.
• For Tailwind projects — typography tokens: Search for font family with font-, font size with text-, font weight with font-, line height with leading-.
• For Tailwind projects — shadow tokens: Search with shadow- prefix.
• For Tailwind projects — border tokens: Search with rounded- and border- prefixes.
• For CSS variable projects: Search for var(--token-name) usage. Exact string match.
• For CSS-in-JS projects: Search for theme.colors.tokenName, theme.spacing.tokenName, etc. Exact property path match.

Report tokens that exist in the token file but have zero references in the codebase. Label each as:

• unused — zero matches found across all search patterns
• build-only — if the token appears only in files under .next/, dist/, or build/ (false positive — note this)

Audit 3: Component Reuse Violations

Check if page-level files use raw HTML elements instead of existing reusable components.

Detection method (deterministic):

Determine page-level file locations based on framework:

• react / next: app/ and src/pages/ (.tsx/.jsx files)
• vue / nuxt: pages/ (.vue files — search in <template> blocks)
• svelte / sveltekit: src/routes/ (+page.svelte files — search in markup section)
• angular: src/app/*/ (.component.html files or inline templates in .component.ts)
• astro: src/pages/ (.astro files — search in template section)
• solid: src/routes/ (.tsx files)

1. Search for raw HTML elements (<button, <input, <select, <textarea, <dialog, <table) in the framework's page-level files (NOT files inside component directories).
2. For each raw HTML element found, check if a corresponding component exists in the components array of design-tokens.json. Match by element type to component name:

- <button → any component with button (case-insensitive) in its name - <input → any component with input or textfield in its name - <select → any component with select or dropdown in its name - <textarea → any component with textarea in its name - <dialog → any component with dialog or modal in its name - <table → any component with table in its name

1. If a raw element is found AND a matching component exists, flag it as a violation.
2. Do NOT flag raw elements in component files themselves (they are the implementation).

Audit 4: Preferred Library Violations (if applicable)

If preferred_libraries exists in design-tokens.json:

4a: Data fetching violations. If preferred_libraries.data_fetching exists, check that components use the selected data fetching library, not alternatives:

• If selected is @tanstack/react-query but a component uses raw fetch or useEffect + useState for data fetching, flag it.
• If selected is @tanstack/react-query but a component imports swr, flag it.
• If selected is axios but a component uses raw fetch, flag it.
• Check for missing error handling on API calls (no onError, no .catch(), no error state).
• Check for missing loading states on API calls.

4b: Other library category violations. For each category in preferred_libraries where installed has more than 1 entry, check that source files use the selected library and not a non-selected alternative:

• If dates.selected is date-fns but a file imports moment, flag it.
• If forms.selected is react-hook-form but a file imports formik, flag it.
• Apply the same check for all categories with competing libraries.

Detection method: For each non-selected library in a category, grep for its import statement (import... from '<package>' or require('<package>')). Any match in source files (excluding node_modules, config files, and lock files) is a violation.

Audit 5: Accessibility Violations

Scan component and page files for common accessibility issues using deterministic pattern matching.

5a: Images without alt text. Search for <img tags without an alt attribute. Every <img must have alt (either descriptive text or alt="" for decorative).

5b: Buttons and links without accessible labels. Search for <button and <a tags that have no text content between opening and closing tags AND no aria-label attribute. Icon-only buttons must have aria-label.

5c: Heading hierarchy violations. For each component/page file, scan for heading tags (h1 through h6). Flag files where heading levels skip (e.g., h1 followed by h3 with no h2).

5d: Click handlers on non-interactive elements. Search for onClick (React/Solid), @click (Vue), onclick (Svelte), (click) (Angular) on <div, <span, or <p elements that do NOT have role="button" or tabIndex.

5e: Form inputs without labels. Search for <input, <select, <textarea that are not preceded by a <label tag and do not have aria-label or aria-labelledby attributes.

Audit 6: Token Conflict Violations

If .claude/d2c/token-conflicts.json exists and contains resolved conflicts (status auto-resolved or user-resolved), scan source files for references to non-canonical (duplicate) token names.

For each resolved conflict entry, extract the duplicates array (these are the non-canonical token names). For each duplicate token name, search for its usage in source files:

• Tailwind projects: Search for the duplicate name in class name contexts (e.g., bg-{name}, text-{name}, border-{name}, p-{name}, m-{name}, gap-{name}, shadow-{name}, rounded-{name}, and any other Tailwind utility prefix).
• CSS Variables projects: Search for var(--{name}) or var(--{category}-{name}) in source files.
• CSS-in-JS projects: Search for theme.{category}.{name} or theme('{category}.{name}') in source files.

For each match found, report:

This is report-only — do NOT auto-fix. Both token names are valid; only one is preferred. Auto-fixing class names or variable references could break if the tokens later diverge intentionally.

If token-conflicts.json does not exist or has no resolved conflicts, skip this audit with a brief note: "Skipped — no resolved token conflicts."

Report Format

Present the audit results as a structured report:

## Design Token Drift Audit Report

### Hardcoded Values Found
| File | Line | Value | Should Be | Token Name |
|------|------|-------|-----------|------------|
| src/components/Header.tsx | 42 | #3B82F6 | Token | primary |
| src/app/dashboard/page.tsx | 87 | padding: 24px | Token | spacing-6 |

**Total: X hardcoded values across Y files**

### Unused Tokens
| Token | Category | Defined Value |
|-------|----------|---------------|
| warning | colors | #F59E0B |

**Total: X unused tokens**

### Component Reuse Violations
| File | Line | Issue | Suggested Component |
|------|------|-------|-------------------|
| src/app/settings/page.tsx | 23 | Inline button styling | Use Button component |

**Total: X potential violations**

### Preferred Library Violations (if applicable)
| File | Line | Issue | Category | Selected | Used Instead |
|------|------|-------|----------|----------|-------------|
| src/app/users/page.tsx | 15 | Uses raw fetch | data_fetching | @tanstack/react-query | fetch |
| src/utils/format.ts | 8 | Imports moment | dates | date-fns | moment |

**Total: X library violations**

### Accessibility Violations
| File | Line | Issue | WCAG Criterion |
|------|------|-------|---------------|
| src/components/Icon.tsx | 12 | img without alt | 1.1.1 Non-text Content |
| src/app/dashboard/page.tsx | 45 | div with onClick, no role="button" | 4.1.2 Name, Role, Value |

**Total: X accessibility violations**

### Token Conflict Violations (if applicable)
| File | Line | Non-Canonical Token | Canonical Equivalent | Category |
|------|------|---------------------|---------------------|----------|
| src/components/Hero.tsx | 18 | brand-blue | primary | colors |

**Total: X non-canonical token usages**

### Summary
- Hardcoded values: X (Y are direct token matches, Z are new/undocumented)
- Unused tokens: X
- Component reuse violations: X
- Preferred library violations: X
- Accessibility violations: X
- Token conflict violations: X (or "N/A — no resolved conflicts")
- **Design system adherence score: X/100**

Scoring — percentage-based per category, then weighted average.

For each category, calculate a per-category score:

1. Hardcoded values (base weight: 30%): Count all scannable values in source files (colors, spacing, typography, shadows, borders). Score = ((total_values - violations) / total_values) * 100. If total_values is 0, score is N/A (category excluded from weighted average).
2. Unused tokens (base weight: 10%): Score = ((total_tokens - unused_tokens) / total_tokens) * 100. If total_tokens is 0, score is N/A (category excluded from weighted average).
3. Component reuse (base weight: 20%): Count raw HTML elements in page-level files that have a corresponding component. Score = ((total_matchable_elements - violations) / total_matchable_elements) * 100. If total_matchable_elements is 0, score is N/A (category excluded from weighted average).
4. Preferred libraries (base weight: 15%): Count all imports in scanned files that belong to a category with a selected library. Score = ((total_categorized_imports - violations) / total_categorized_imports) * 100. If total_categorized_imports is 0 or no preferred_libraries exist, score is N/A (category excluded from weighted average).
5. Accessibility (base weight: 25%): Count all auditable elements (images, buttons, links, headings, click-handler divs, form inputs). Score = ((total_elements - violations) / total_elements) * 100. If total_elements is 0, score is N/A (category excluded from weighted average).
6. Token conflict violations (informational — not scored): Count usages of non-canonical tokens from resolved conflicts. This is informational and does NOT affect the adherence score. Report count only.

N/A handling and weight redistribution: When one or more categories have a score of N/A, exclude them from the final score calculation and redistribute their weights proportionally across the remaining categories. The redistribution formula is: effective_weight = base_weight / sum_of_active_base_weights. For example, if "Unused tokens" (10%) and "Preferred libraries" (15%) are both N/A, the remaining categories have base weights 30% + 20% + 25% = 75%. Their effective weights become: Hardcoded = 30/75 = 40%, Component reuse = 20/75 = 26.7%, Accessibility = 25/75 = 33.3%.

Final score = sum of (category_score * effective_weight) for all non-N/A categories. If ALL categories are N/A, report: "No auditable elements found — score: N/A."

Display per-category scores in the summary:

### Summary
- Hardcoded values: X violations / Y total values — **score: Z/100**
- Unused tokens: X / Y total tokens — **score: N/A** (no tokens defined)
- Component reuse: X violations / Y matchable elements — **score: Z/100**
- Preferred libraries: X violations / Y imports — **score: N/A** (no preferred libraries)
- Accessibility: X violations / Y auditable elements — **score: Z/100**
- **Design system adherence score: X/100** (weighted average of non-N/A categories)

Important Rules

• Only scan source files. Skip node_modules, .next, dist, build, coverage, test files (*.test.*, *.spec.*, __tests__/), and config files (*.config.*, postcss.*).
• Flag ALL matches from regex scans. Label each as either definite (the hardcoded value exactly matches a token value) or undocumented (the hardcoded value has no matching token). Do not skip any regex match. Do not use subjective confidence — every match is either exact-match or not.
• File modification rules: By default, this is a read-only audit — do not modify any files. When --fix is provided (without --dry-run), auto-fix definite matches only. When --dry-run --fix is provided, show diff previews and wait for user approval before modifying anything.
• If the project is small (< 10 source files), say so and note that the audit may be less meaningful.