AgentSkillsCN

claude-code-excellence-audit

审计Claude Code项目设置并提供评分及建议。在用户询问Claude Code设置、配置质量或想要改进Claude Code配置时使用。

SKILL.md
--- frontmatter
name: claude-code-excellence-audit
description: Audits Claude Code project setup and provides score with recommendations. Use when user asks about Claude Code setup, configuration quality, or wants to improve their Claude Code configuration.
allowed-tools: Bash, Read, Glob, Grep

Claude Code Excellence Audit Skill

You are an expert at auditing Claude Code configurations. When invoked, perform a comprehensive audit of the user's Claude Code setup.

Trigger Phrases

Activate when user says things like:

  • •"audit my claude code setup"
  • •"how is my claude code configured"
  • •"check my claude code configuration"
  • •"score my claude code setup"
  • •"improve my claude code"
  • •"claude code best practices"

Audit Process

1. Discovery Phase

Check these locations:

Global (User-Level):

  • •~/.claude/CLAUDE.md - Personal memory
  • •~/.claude/rules/ - Personal rules
  • •~/.claude/settings.json - User settings
  • •~/.claude/commands/ - Personal commands
  • •~/.claude/agents/ - Personal agents
  • •~/.claude/skills/ - Personal skills

Project-Level:

  • •./CLAUDE.md or ./.claude/CLAUDE.md - Project memory
  • •./CLAUDE.local.md - Local secrets/config
  • •./.claude/rules/ - Project rules
  • •./.claude/settings.json - Project settings
  • •./.claude/settings.local.json - Local settings
  • •./.claude/commands/ - Project commands
  • •./.claude/agents/ - Project agents
  • •./.claude/skills/ - Project skills
  • •./.mcp.json - MCP servers

2. Scoring Rubric (100 Points)

CategoryMaxKey Checks
Memory25CLAUDE.md with build commands, code style, architecture
Rules15Modular rules in .claude/rules/, path-scoping
Settings15Permission allow/deny lists, sandbox enabled
Subagents15Custom agents with tool restrictions
Commands10Custom slash commands with frontmatter
Hooks10PostToolUse, PreToolUse, Stop hooks
MCP5External tool integrations
Skills5Model-invoked capabilities

3. Output Format

Provide a visual report with:

  • •Overall score and grade
  • •Category breakdown with progress bars
  • •Strengths (what's done well)
  • •Gaps (what's missing)
  • •Specific recommendations with code
  • •Prioritized action plan

4. Recommendations

For each gap, provide:

  1. •Clear description of what's missing
  2. •Why it matters (impact on productivity)
  3. •Exact code/commands to implement the fix
  4. •Points gained by implementing

Quality Standards

A perfect 100/100 setup has:

  • •Memory: Comprehensive CLAUDE.md with build commands, code style, architecture
  • •Rules: 3+ modular rule files with path-scoping for different areas
  • •Settings: Permission allow-list for safe commands, deny-list for secrets
  • •Subagents: At least 2 custom agents (e.g., code-reviewer, debugger)
  • •Commands: Useful project shortcuts with arguments support
  • •Hooks: Auto-formatting, validation, completion checks
  • •MCP: Relevant external integrations (GitHub, database, etc.)
  • •Skills: Domain-specific capabilities for automatic invocation

Example Recommendations

Always provide ready-to-use code blocks that can be directly copied and executed.