audit-claude-config

name	audit-claude-config
description	Analyze .claude configuration for gaps, improvements, and needed updates
user-invocable	true
argument-hint

Audit Claude Configuration Skill

Performs comprehensive analysis of the .claude/ directory and related configuration files, identifying gaps, suggesting improvements, and proposing updates based on project evolution.

Usage

/audit-claude-config

What It Analyzes

1. Rules (.claude/rules/)

• ✅ Are rules still relevant to current work?
• ✅ Do rules contradict each other?
• ✅ Are there new patterns that should be documented?
• ✅ Are rules loading at the right times (path patterns)?

2. Skills (.claude/skills/)

• ✅ Are there repeated workflows that should be skills?
• ✅ Can existing skills be consolidated?
• ✅ Are skill descriptions accurate?
• ✅ Are there gaps in available commands?

3. Agents (.claude/agents/)

• ✅ Are agents being used effectively?
• ✅ Should new specialized agents be created?
• ✅ Do agent tool restrictions make sense?
• ✅ Is agent memory growing effectively?

4. Agent Memory

• ✅ What have agents learned?
• ✅ Is memory becoming too large?
• ✅ Are patterns documented in rules?
• ✅ Should memory be compressed/archived?

5. Workflows

• ✅ Are generated workflows reusable?
• ✅ Should any be saved as named workflows?
• ✅ Are we using workflows appropriately?

6. Settings (.claude/settings.json)

• ✅ Are permissions appropriate?
• ✅ Should hooks be added/modified?
• ✅ Are environment variables set correctly?

7. MCP Servers (.mcp.json)

• ✅ Should any be enabled?
• ✅ Are we manually doing what MCP could automate?

8. EditorConfig & Build Props

• ✅ Are code style rules being followed?
• ✅ Is Directory.Build.props up to date?
• ✅ Should new warnings be enforced?

Output Format

# 🔍 Claude Configuration Audit Report

Generated: [Date]

## 📊 Summary
- Rules: X active, Y need updates
- Skills: X available, Y suggestions
- Agents: X defined, Y utilized, Z suggestions
- Overall Health: ✅ Excellent | ⚠️ Good | ❌ Needs Attention

---

## ✅ What's Working Well

- [Strengths in current configuration]

---

## ⚠️ Identified Gaps

### Rules
1. **Missing Pattern Documentation**
   - Gap: We've fixed nullable warnings 10 times with same pattern
   - Recommendation: Add to `.claude/rules/modernization.md`
   - Priority: Medium

### Skills
1. **Repeated Workflow**
   - Gap: Manually running build → test → coverage check frequently
   - Recommendation: Create `/pre-commit` skill
   - Priority: High

### Agents
1. **Underutilized Agent**
   - Gap: test-writer agent only used 2 times
   - Recommendation: Delegate more test writing tasks
   - Priority: Low

---

## 💡 Suggested Improvements

### High Priority
1. **Add nullable pattern to modernization.md**
   ```markdown
   Pattern: private readonly IConfig _config = null!;
   Reason: Constructor initialization, guaranteed non-null

2. Create /pre-commit skill Automates: build + test + coverage check

Medium Priority

1. Update code-reviewer agent Add: Check for var keyword usage

2. Enable GitHub MCP server Use case: Automating issue creation

Low Priority

1. Consolidate similar skills Combine: audit-build + verify-coverage into /pre-commit

---

📈 Agent Learning Summary

code-reviewer

• Reviews completed: 5
• Patterns learned: 8
• Common issues: Nullable warnings, missing var keyword
• Effectiveness: ✅ High

test-writer

• Tests written: 12
• Coverage achieved: 87% average
• Patterns learned: 6
• Effectiveness: ✅ High

---

🎯 Recommended Actions

Immediate (Do Now)

• Add nullable pattern to modernization.md
• Create /pre-commit skill

Short-term (This Week)

• Update code-reviewer checklist
• Compress agent memory if > 100 entries

Consider (Optional)

• Enable GitHub MCP server
• Create nullable-fixer specialized agent

---

📋 Configuration Health Checklist

• [✅] Rules load correctly and are relevant
• [✅] Skills are documented and user-invocable
• [⚠️] Agents could be used more frequently
• [✅] Agent memory is learning effectively
• [✅] Settings permissions are appropriate
• [❌] MCP servers might be beneficial but disabled
• [✅] EditorConfig rules are enforced
• [✅] Git conventions documented

---

📊 Metrics

• Total .claude files: X
• Rules: X (Y with path restrictions)
• Skills: X available
• Agents: X defined, Y with memory
• Agent memory size: X KB
• Workflows generated: X
• Last audit: [Previous date] or Never

---

🔄 Next Audit Recommendation

Run /audit-claude-config again:

• After 50 commits
• After 2 weeks
• When starting new feature area
• If configuration feels stale

## Analysis Process

I will:

1. **Read all configuration files**
   - Rules, skills, agents, memory
   - Settings, MCP, editorconfig

2. **Analyze git history**
   - Look for repeated patterns
   - Identify manual workflows
   - Check commit frequency

3. **Review agent memory**
   - What patterns have been learned?
   - Is learning transferring to rules?

4. **Check for gaps**
   - Missing documentation
   - Underutilized features
   - Automation opportunities

5. **Compare to best practices**
   - Are we following our own standards?
   - Are configurations optimal?

6. **Generate recommendations**
   - Prioritized by impact
   - Specific, actionable
   - With examples

## Use Cases

### Weekly Review

User: /audit-claude-config → Review what's changed → Update configurations → Plan improvements

### After Major Milestone

User: "We just finished nullable fixes. /audit-claude-config" → Document patterns learned → Update rules with discoveries → Archive completed work notes

### New Team Member Onboarding

User: /audit-claude-config → Shows current state → Explains why things are configured this way → Identifies what's actively used

### Feeling Lost

User: "Configuration feels messy. /audit-claude-config" → Identifies what's stale → Suggests cleanup → Proposes reorganization

## Output Saved To

Report saved to: `.claude/audits/audit-{timestamp}.md`

## Related Skills

- `/verify-coverage` - Check test coverage
- `/audit-build` - Analyze build errors
- `/create-modernization-issue` - Create issues from findings

## Important Notes

- This skill is **read-only analysis** - I suggest changes but don't apply them
- You review recommendations and approve updates
- Agent memory is referenced but not modified
- Safe to run anytime - no side effects

## Implementation

When invoked, I will:

1. Read all `.claude/` files
2. Analyze patterns and usage
3. Compare to best practices
4. Generate comprehensive report
5. Save report to `.claude/audits/`
6. Present findings with recommendations
7. Ask: "Which improvements should I implement?"