Documentation Auditor

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. Report: Output findings and recommended actions

Phase 4: Semantic Audit Delegation

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

Audit Categories

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

shared/references/audit_scoring.md

Reference Files

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

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.
• Delete, don't archive: Legacy content should be removed, not moved to "archive"
• No history: Documents describe current state only; git tracks history