ln-600-docs-auditor（ln 600 文档审核员）

Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.

Documentation Auditor

Audit project documentation quality. Universal for any tech stack.

Purpose

• Proactively compress - find all opportunities to reduce size while preserving value
• Eliminate meaningless, redundant, and verbose content
• Convert prose to structured formats (tables, lists)
• Verify documentation hierarchy with CLAUDE.md as root
• Detect duplication and enforce Single Source of Truth
• Ensure docs match current code state
• Semantic verification - delegate to ln-601 to verify content matches SCOPE and codebase reality

Invocation

• Direct: User invokes for documentation quality review
• Pipeline: Called by ln-100-documents-pipeline (Phase 5, if auditDocs=true)

Workflow

1. Scan: Find all.md files in project (CLAUDE.md, README.md, docs/**)
2. Build Tree: Construct hierarchy from CLAUDE.md outward links
3. Audit Categories 1-7: Run structural checks (see Audit Categories below)
4. Semantic Audit (Category 8): For each project document, delegate to ln-601-semantic-content-auditor
5. Score: Calculate X/10 per category (including semantic scores from ln-601)
6. Context Validation: Post-filter findings (see below)
7. Report: Output findings and recommended actions

Phase 4: Semantic Audit Delegation

For each project document (excluding tasks/, reference/, presentation/):

FOR doc IN [CLAUDE.md, docs/README.md, docs/project/*.md]:
    result = DELEGATE ln-601-semantic-content-auditor {
        doc_path: doc,
        project_root: project_root,
        tech_stack: detected_stack
    }
    semantic_findings.append(result.findings)
    semantic_scores[doc] = result.scores

Target documents: CLAUDE.md, docs/README.md, docs/documentation_standards.md, docs/principles.md, docs/project/*.md

Excluded: docs/tasks/, docs/reference/, docs/presentation/, tests/

Audit Categories

Context Validation (Post-Filter)

MANDATORY READ: Load shared/references/context_validation.md

Apply Rule 1 + doc-specific inline filters:

FOR EACH finding WHERE severity IN (HIGH, MEDIUM):
  # Rule 1: ADR/Planned Override
  IF finding matches ADR → advisory "[Planned: ADR-XXX]"

  # Doc-specific: Compression context
  IF Cat 3 (Compression) finding:
    - Skip if path in references/ or templates/ (reference docs = naturally large)
    - Skip if filename contains architecture/design/api_spec
    - Skip if tables+lists > 50% of content (already structured, not verbose prose)

  # Doc-specific: Actuality severity calibration
  IF Cat 5 (Actuality) finding:
    - Path/function COMPLETELY missing → CRITICAL (broken docs)
    - Path exists but deprecated/renamed → HIGH (not CRITICAL)
    - Example code outdated but concept valid → MEDIUM

Downgraded findings → "Advisory Findings" section in report.

Output Format

## Documentation Audit Report - [DATE]

### Compliance Score

| Category | Score | Issues |
|----------|-------|--------|
| Hierarchy & Links | X/10 | N issues found |
| Single Source of Truth | X/10 | N duplications |
| Proactive Compression | X/10 | N compression opportunities |
| Requirements Compliance | X/10 | N violations |
| Actuality | X/10 | N mismatches with code |
| Legacy Cleanup | X/10 | N legacy items |
| Stack Adaptation | X/10 | N stack mismatches |
| Semantic Content | X/10 | N semantic issues (via ln-601) |
| **Overall** | **X/10** | |

### Critical Findings

- [ ] **[Category]** `path/file.md:line` - Issue description. **Action:** Fix suggestion.

### Recommended Actions

| Priority | Action | Location | Category |
|----------|--------|----------|----------|
| High | Remove duplicate section | docs/X.md | SSOT |
| Medium | Add link to CLAUDE.md | docs/Y.md | Hierarchy |

Scoring Algorithm

MANDATORY READ: Load shared/references/audit_scoring.md for unified scoring formula.

Severity mapping:

Reference Files

• Size limits and targets: references/size_limits.md
• Detailed checklist: references/audit_checklist.md

Definition of Done

• All.md files in project scanned and hierarchy tree built from CLAUDE.md
• Categories 1-7 (structural) audited with score X/10 each
• Category 8 (semantic) delegated to ln-601 for each target document; scores collected
• Overall Compliance Score calculated (average of 8 categories)
• Critical Findings listed with file:line, category, and fix suggestion
• Recommended Actions table generated with priority, action, location, category

Critical Notes

• Fix content, not rules: NEVER modify standards/rules files (*_standards.md, *_rules.md, *_limits.md) to make violations pass. Always fix the violating files instead.
• Verify facts against code: Actively check every path, function name, API, config mentioned in docs. Run commands. Outdated docs mislead - they're worse than no docs.
• Compress always: Size limits are upper bounds, not targets. A 100-line file instead of 300 is a win. Always look for compression opportunities.
• Meaningless content: Remove filler words, obvious statements, over-explanations. If it doesn't add value, delete it.
• No code in docs: Documents describe algorithms in tables or ASCII diagrams. Code belongs in codebase.

- Forbidden: Code blocks, implementation snippets - Allowed: Tables, ASCII diagrams, Mermaid, method signatures (1 line) - Instead of code: "See Official docs" or "See src/file.cs:42"

• Format Priority: Tables/ASCII > Lists (enumerations only) > Text (last resort)
• Stack adaptation: Verify all documentation references match project stack..NET project must not have Python examples. Check official doc links point to correct platform (Microsoft docs for C#, MDN for JS, etc.)
• Code is truth: When docs contradict code, always update docs. Never "fix" code to match documentation.
• SSOT re-verification after fixes: After making ANY documentation change, re-check that the fix maintains Single Source of Truth. If content exists in multiple files, keep it in the canonical source only and replace other occurrences with a link to that source (e.g., See [section](path#anchor)). Never duplicate content inline — always link. Canonical source hierarchy: CLAUDE.md → docs/README.md → docs/project/*.md → docs/reference/*.md.
• Delete, don't archive: Legacy content should be removed, not moved to "archive"
• No history: Documents describe current state only; git tracks history

Version: 4.0.0 Last Updated: 2026-01-28