claude-config-audit

Audit the .claude configuration for valid frontmatter, discoverable agents and skills, missing skill references, stale slash-command plugin names, broad permissions, and hook/settings regressions.

SKILL.md

GitHub

Claude Config Audit

Use this skill before merging any change under .claude/**, CLAUDE.md, or agent-memory files.

Workflow

  1. Parse YAML frontmatter in every .claude/**/*.md file.
  2. Build inventories for agents, skills, commands, rules, imports, output styles, hooks, and memory.
  3. Verify every agent skills: entry maps to .claude/skills/<name>/SKILL.md.
  4. Search slash commands for stale agent-skills:* references and browser-only workflows that do not apply to this terminal simulator.
  5. Check orchestrators for overly broad delegation and confirm /ship requires DegenExus verification evidence.
  6. Confirm .claude/settings.json exists and contains conservative secret/destructive-command denies plus hook wiring.
  7. Report findings as BLOCKER, IMPORTANT, or SUGGESTION with exact file paths.

Required Local Commands

python .claude/hooks/validate-claude-config.py
python -m compileall -q .claude/hooks

Exit Criteria

  • Frontmatter parses cleanly.
  • Agent skill references are resolvable.
  • Commands reference local agents/skills or documented project workflows only.
  • Hooks and settings are present and auditable.