Claude Agents Marketplace
← Back to Skills
skillClaude Code ≥ 1.0/audit

audit

Full-system audit — settings hygiene, dependencies, secrets, types, tests, memory. Returns a prioritized action list. Use monthly or before major releases.

  • workflow

Install

~/.claude/skills/audit/SKILL.md
Run a comprehensive audit across the workspace and report issues with severity.

## Steps

1. **Settings hygiene**
   - `cat  | jq '.permissions.allow | length'` — flag if >80
   - grep for hardcoded one-shot rules (`rm <specific file>`, `grep -n "<specific string>"`) — these are cruft
   - check hooks are present: SessionStart, PreToolUse (Edit/Write protect + secrets-scan + Bash danger), PostToolUse (prettier)

2. **Dependencies**
   - for each of agent-dashboard, courseai, my-assistant: run `npm outdated` and `npm audit --audit-level=high`
   - flag high/critical vulnerabilities

3. **Secrets scan**
   - `grep -rnE "sk-ant-|sk-proj-|ntn_[A-Za-z0-9]{40}|AKIA[0-9A-Z]{16}|ghp_[A-Za-z0-9]{36}|-----BEGIN.*PRIVATE KEY-----" c:/AI --include="*.ts" --include="*.js" --include="*.json" --include=
…

Paste into ~/.claude/skills/audit/SKILL.md and Claude Code will pick it up on next session.

Definition

Run a comprehensive audit across the workspace and report issues with severity.

Steps

  1. Settings hygiene

    • cat | jq '.permissions.allow | length' — flag if >80
    • grep for hardcoded one-shot rules (rm <specific file>, grep -n "<specific string>") — these are cruft
    • check hooks are present: SessionStart, PreToolUse (Edit/Write protect + secrets-scan + Bash danger), PostToolUse (prettier)

  2. Dependencies

    • for each of agent-dashboard, courseai, my-assistant: run npm outdated and npm audit --audit-level=high
    • flag high/critical vulnerabilities

  3. Secrets scan

    • grep -rnE "sk-ant-|sk-proj-|ntn_[A-Za-z0-9]{40}|AKIA[0-9A-Z]{16}|ghp_[A-Za-z0-9]{36}|-----BEGIN.*PRIVATE KEY-----" c:/AI --include="*.ts" --include="*.js" --include="*.json" --include="*.md"
    • must return zero results

  4. Memory health

    • delegate to general-purpose agent: Agent({ subagent_type: "general-purpose", description: "Audit memory dir", prompt: "Audit read-only. Check MEMORY.md drift, duplicates, stale project facts (>30d old or dead paths), frontmatter integrity (name/description/type), Why/How structure on feedback/project entries. Report issues + proposed DELETE/MERGE/UPDATE/FIX-INDEX actions. Do not edit." })
    • or invoke /prune-memory skill directly

  5. Types & tests

    • for each project with tsconfig.json: npx tsc --noEmit (list errors)
    • for each project with vitest: npm test -- --run (pass/fail count)

  6. Agent/skill inventory

    • ls | wc -l — current count
    • `grep -L "^model:" — agents without explicit model (cost optimization opportunity)

Output

Return a structured report:

Summary

  • Overall health: 🟢 / 🟡 / 🔴
  • Critical issues: N
  • Warnings: M

Critical (fix now)

  • [severity] issue + file:line + fix

Warnings (fix this week)

  • ...

Optimizations (nice to have)

  • ...

Keep under 500 words. Actionable, not descriptive.