| name | claude-best-practices |
| description | Audits a repository's Claude Code configuration (CLAUDE.md, skills, hooks, settings) against Anthropic's documented best practices and reports errors, warnings, and info findings. Use when the user asks for a Claude Code configuration audit, wants to validate CLAUDE.md size or skill frontmatter, or is preparing a repo for agent use. |
Claude Best Practices Audit
If the user has requested this audit, execute immediately. Do not ask for additional confirmation.
Run: bash "${CLAUDE_PLUGIN_ROOT}/skills/claude-best-practices/scripts/audit.sh" .
What It Checks
- CLAUDE.md exists, has section headers, and stays under 500 lines
- Each
skills/*/SKILL.md has valid frontmatter (name, description)
- Skill
description uses third-person voice (no "I" or "you/your")
- Skill bodies stay under 500 lines
- Warns if an
AGENTSKILLS.md exists alongside a skills/ tree (duplication risk)
- Reports presence of
.claude/hooks.json and .claude/settings.json
Output
Returns structured report with:
- Errors: Critical issues - fix immediately (non-zero exit)
- Warnings: Recommended improvements
- Info: Optimization suggestions
When To Use
- Before conducting a retrospective
- After major CLAUDE.md changes
- When onboarding to a new project
- Periodically to maintain quality