Claude-skill-registry docs-audit

Documentation Audit Skill

install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/docs-audit" ~/.claude/skills/majiayu000-claude-skill-registry-docs-audit && rm -rf "$T"
manifest: skills/data/docs-audit/SKILL.md
source content

Documentation Audit Skill

Triggers (ANY of these should invoke this skill):

  • /docs-audit
    , "do an audit", "run audit"
  • "update the readme", "update docs", "update documentation"
  • "update the website", "sync website"
  • "update everything", "ship it", "we're done"
  • "push to git", "commit everything", "let's commit"
  • "end session", "session close", "wrap up"
  • "prepare for release", "get ready to ship"

Part of standard procedures:

  • Session close: Run docs-audit before session summary
  • Before push: Run docs-audit before git operations

Don't let stale docs slip through.

Purpose

Comprehensive documentation audit that removes cognitive burden from the user. You are NOT just updating docs - you are thinking through everything so the user doesn't have to.

The user is not an engineer, designer, or marketer. They built something cool and need YOU to:

  1. Figure out what exists in the code
  2. Figure out what's missing from the docs
  3. Think through it from multiple expert perspectives
  4. Present a clear action plan
  5. Execute it properly

The Process

Phase 1: Discovery (YOU figure out what exists)

Before ANYTHING else, audit the codebase to understand what's actually there:

1. Find all entry points (main.swift, App.swift, CLI commands)
2. Find all public features (menu items, commands, UI screens)
3. Find all configuration options (settings, preferences, flags)
4. Find all integrations (APIs, services, other apps)
5. Read the current README.md and any docs/ folder
6. Check if there's a website (docs site, landing page)

Output a Feature Inventory:

## Feature Inventory (from code)

### Commands/Actions
- [ ] /command1 - description (file:line)
- [ ] /command2 - description (file:line)

### UI Elements
- [ ] Settings panel - what it does
- [ ] Menu bar items - what they do

### Integrations
- [ ] API X - what it connects to

### Configuration
- [ ] Setting A - what it controls

Phase 2: Multi-Perspective Audit

Run these 5 specialized audits. Each one thinks deeply from their expertise:

1. Engineer Audit

Read:

prompts/engineer.md
Focus: Technical accuracy, completeness, API docs

2. Designer Audit

Read:

prompts/designer.md
Focus: Screenshots, demos, visual storytelling

3. Marketer Audit

Read:

prompts/marketer.md
Focus: Value proposition, benefits, "why should I care?"

4. User Advocate Audit

Read:

prompts/user.md
Focus: Onboarding, clarity, "can my grandma understand this?"

5. QA Audit

Read:

prompts/qa.md
Focus: Edge cases, known issues, gotchas, troubleshooting

6. Hygiene Audit

Read:

prompts/hygiene.md
Focus: Document duplication, terminology drift, memory MCP gaps

Catches: Creating new docs instead of updating existing ones. Multiple docs for "next steps", bugs in files instead of memory.

7. Security Audit (CRITICAL)

Read:

prompts/security.md
Focus: Leaked secrets, PII in screenshots, internal URLs exposed

Catches: API keys in examples, passwords in config, screenshots showing real emails, internal Slack/Jira links, file paths with usernames.

8. Freshness Audit

Read:

prompts/freshness.md
Focus: Stale examples, outdated screenshots, broken links, wrong version numbers

Catches: Code examples that don't compile, screenshots of old UI, "Requires macOS 12" when we're on 15, 404 links.

9. Completeness Audit

Read:

prompts/completeness.md
Focus: Incomplete docs, unchecked checklists, TODO placeholders, time-sensitive items

Catches: Templates never filled in, "Coming soon" from 6 months ago, certificate expiry dates left blank, 15 unchecked action items.

10. Ops Audit (CRITICAL)

Read:

prompts/ops.md
Focus: Git hygiene, certificates, dependencies, domains, code TODOs, cross-project drift

Catches: Stale branches, expiring certs, vulnerable dependencies, lapsing domains, TODO comments rotting in code, version mismatches across projects.

This is "everything humans forget to check" - trivial for AI, nightmare to track manually.

11. Brand Compliance Audit (CRITICAL)

Read:

prompts/brand-compliance.md
Reference:
/Users/sj/SaneApps/meta/Brand/SaneApps-Brand-Guidelines.md
Focus: Colors, typography, accent usage, SaneUI integration

Catches:

  • Grey-on-grey unreadable UI (using
    .gray
    instead of brand Stone/Cloud)
  • Tiny fonts (12px for body text instead of 16px)
  • No accent colors (everything monochrome)
  • Not using SaneUI shared package
  • Hardcoded colors instead of brand palette

This is WHY the apps look like generic grey developer UIs instead of the polished brand.


Phase 3: Gap Report

Present findings to user in this format:

# Documentation Audit Report

## Executive Summary
- X features in code, Y documented (Z% coverage)
- Critical gaps: [list]
- Estimated effort: [quick/medium/significant]

## Critical Gaps (Fix These First)
| Gap | Why It Matters | Recommendation |
|-----|----------------|----------------|
| Feature X undocumented | Users can't discover it | Add section with example |

## Missing Visuals
| What | Current State | Needed |
|------|---------------|--------|
| Main UI | No screenshot | Capture showing X, Y, Z |

## Stale Content
| Section | Problem | Fix |
|---------|---------|-----|
| Installation | Shows old method | Update to new flow |

## Website Status
- [ ] Exists / Doesn't exist
- [ ] Last updated: [date or unknown]
- [ ] Matches README: Yes/No
- [ ] Needs: [list of updates]

## Documentation Hygiene (Duplication/Sprawl)
| Issue | Files Involved | Fix |
|-------|----------------|-----|
| Duplicate "next steps" docs | TODO.md, ROADMAP.md | Delete, use SESSION_HANDOFF.md only |
| Bugs in files not memory | BUGS.md | Move to memory MCP, delete file |
| Terminology drift | "handoff" vs "todos" | Standardize to "handoff" |

## Security Issues (FIX IMMEDIATELY)
| File | Line | Issue | Action |
|------|------|-------|--------|
| README.md | 45 | Looks like real API key | Replace with placeholder |
| screenshot.png | - | Shows /Users/john/ path | Retake screenshot |

## Freshness Issues
| Location | Problem | Current Reality |
|----------|---------|-----------------|
| README.md:15 | Shows v1.2.0 | Actually v2.4.0 |
| install.md | `brew install foo` | Package renamed |
| screenshot.png | Old UI | Redesigned in v2.0 |

## Incomplete Documents (USER ACTION NEEDED)
| File | Issue | What You Need To Do |
|------|-------|---------------------|
| DISASTER_RECOVERY.md | 15 unchecked items | Fill in cert expiry, domain dates, contacts |
| SETUP.md | TODO placeholders | Provide actual config values |

## Time-Sensitive Items
| Item | Status | Urgency |
|------|--------|---------|
| Dev certificate | Expiry not documented | Check Keychain, document date |
| Domains | "[ ] CHECK" not filled | Look up actual expiry dates |

## Ops Issues (Git, Deps, Infrastructure)
| Category | Issue | Action |
|----------|-------|--------|
| Git | 7 stale branches | Review and delete |
| Dependencies | 3 npm vulnerabilities | Run npm audit fix |
| Code | 23 TODO comments | Review and resolve |
| Legal | Copyright says 2024 | Update to 2026 |

## Calendar Reminders Needed
| What | When | Who |
|------|------|-----|
| Dev certificate renewal | [DATE] | You |
| Domain renewal | [DATE] | You |
| Apple Developer $99 | [DATE] | You |

## Recommended Actions (Priority Order)
1. [ ] [Action] - [why] - [effort: 5min/30min/2hr]
2. [ ] [Action] - [why] - [effort]
...

Phase 4: User Approval

STOP and ask the user:

"Here's what I found. Before I update anything:

  1. Does this inventory match your understanding?
  2. Any features I missed?
  3. Which gaps should I prioritize?
  4. Do you want me to proceed with all recommendations?"

Do NOT update docs until user confirms.


Phase 5: Execute Updates

Only after approval:

  1. Update README.md with all gaps addressed
  2. Update website if it exists
  3. Create list of screenshots/demos needed (user must capture these)
  4. Commit with clear message listing what was updated

Key Principles

  1. You think, they approve - Don't make them figure out what's missing
  2. Show your work - Present the inventory so they can catch what you missed
  3. Multiple lenses - Engineer sees different gaps than Marketer
  4. Visuals matter - A README without screenshots is incomplete
  5. Website = README - If they're out of sync, that's a bug
  6. Effort estimates - Help them prioritize what to tackle
  7. Consolidate, don't multiply - Update existing docs, don't create new ones
  8. Bugs go in memory MCP - Not markdown files that go stale
  9. One source of truth - If info is in 2 places, delete one

Example Invocations

User: "update the readme" → Run full audit, present gaps, get approval, then update

User: "quick docs check" → Run audit, present gaps, stop (don't update)

User: "/docs-audit --engineering-only" → Run only technical completeness check

User: "is the website up to date?" → Compare website to README, report differences

User: "end session" / "let's wrap up" → Run audit BEFORE session summary. Flag any stale docs. Don't let them slip.

User: "push to git" / "commit and push" → Run audit BEFORE git operations. Confirm docs are current.

User: "ship it" / "we're done" → Full audit. This is the last chance to catch missing docs.


Session End Integration

When this skill is triggered by session close:

  1. Run the full audit (all 10 perspectives)
  2. Present gaps to user
  3. Ask: "Should I update docs before we close?"
  4. If yes → update, then proceed with session summary
  5. If no → note in SESSION_HANDOFF.md that docs need attention

The goal: Never end a session with hidden features or stale docs.


Quick Mode vs Full Mode

Full mode (default): All 10 audits, comprehensive report

  • Use for: Major releases, session end, "update everything"

Quick mode (add

--quick
): Engineer audit only, fast check

  • Use for: Mid-session check, "did I break anything?"

Single perspective (add

--engineering-only
,
--design-only
, etc.):

  • Use for: Targeted checks when you know what you're looking for