DO
documentation-audit
Full audit and update of BioETL project documentation for v5.23+. Use when asked to review docs for staleness, reconcile docs with code, sync RULES.md and REQUIREMENTS.md, update architecture/provider/contract docs, reflect ADR-010/ADR-014/ADR-017, or identify dead documentation.
Install
mkdir -p .claude/skills/documentation-audit && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/16535" && unzip -o skill.zip -d .claude/skills/documentation-audit && rm skill.zipInstalls to .claude/skills/documentation-audit
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Full audit and update of BioETL project documentation for v5.23+. Use when asked to review docs for staleness, reconcile docs with code, sync RULES.md and REQUIREMENTS.md, update architecture/provider/contract docs, reflect ADR-010/ADR-014/ADR-017, or identify dead documentation.280 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
About this skill
Documentation Audit
Objective
Perform a full documentation audit of BioETL and bring docs in sync with code and ADRs (v5.23+). Produce a clear audit report, a prioritized plan, and updated documentation changes.
Source Of Truth
- Root runtime contract:
../../../AGENTS.md - Project rules:
../../../docs/00-project/RULES.md - Requirements:
../../../docs/01-requirements/REQUIREMENTS.md - Accepted ADRs:
../../../docs/02-architecture/decisions - Normative index:
../../../docs/00-project/NORMATIVE_SOURCES.md
Required inputs (before starting)
- Load
references/audit-checklist.md— checklist for each audit area. - Load
references/report-template.md— template for the audit report. - Treat
../../../AGENTS.md,../../../docs/00-project/ai/agents/guides/MEMORY_USAGE.md, and../../../docs/00-project/ai/agents/policy/POST_CHANGE_VALIDATION.mdas the active runtime policy entrypoints for AI-surface documentation work.
Workflow
1. Intake and scope
- Confirm repo root and target version (v5.23+).
- Identify doc entry points: README.md, mkdocs.yml.
- List files under
docs/:rg --files docs.
2. Audit (follow audit-checklist.md)
- Compare documentation to current code and configs.
- Focus: RULES.md, REQUIREMENTS.md, architecture, provider, contract docs.
- Check ADR alignment: ADR-010 (Local-Only), ADR-014 (Determinism), ADR-017 (Observability).
- Trace doc claims back to runtime/code/config/workflow sources and note mirror surfaces when AI guidance files are involved.
- Record findings with severity: Critical, High, Medium, Low.
3. Plan
- Turn findings into a concrete change list.
- Prioritize by impact (Critical > High > Medium > Low).
- Call out unknowns that need user confirmation.
4. Update (if user requests)
- Edit docs to match current code and ADRs.
- Keep versions and dates explicit in text.
- For obsolete docs: propose delete/archive, do not remove unless user asks.
5. Verify
- Check links and nav entries (mkdocs.yml).
- Confirm RULES.md and REQUIREMENTS.md are synchronized.
- Ensure ADRs are reflected in top-level docs.
- When AI runtime/docs surfaces were touched, verify runtime source first and mirror consistency second.
Practical commands
# ADR references
rg -n "ADR-010|ADR-014|ADR-017" docs README.md mkdocs.yml
# Version mentions
rg -n "v5\.23|5\.23" docs README.md
# Doc refs in nav
rg -n "docs/|\.md" mkdocs.yml README.md
# Orphan scan: list docs/ then search each filename in mkdocs.yml and other docs
rg --files docs
Output format
- Use
references/report-template.mdstructure for the audit report. - Provide a short prioritized change list and required user decisions.
- Do not change code unless explicitly requested; focus on documentation.
Constraints
MUST
- Use report-template.md for the audit report.
- Verify findings against actual code and configs.
- Record severity for each finding.
MUST NOT
- Change code unless user explicitly asks.
- Remove docs without explicit user approval.
- Document desired behavior instead of reality when code and docs diverge — flag and propose options.
SHOULD
- Prefer documenting reality; when code and docs diverge, flag and propose options.