| name | harness-audit |
| description | Audit Claude Code harness configuration for optimization opportunities |
| triggers | ["harness audit","audit harness","check setup","optimize harness"] |
Harness Audit
Comprehensive audit of Claude Code configuration quality.
When to Use
- After initial setup or major configuration changes
- Periodically (monthly) to identify optimization opportunities
- When experiencing slow sessions or high costs
- Before starting a new project
How It Works
-
Read Configuration
- Parse
~/.claude/settings.json (global)
- Parse
~/.claude/CLAUDE.md (instructions)
- Scan
~/.claude/hooks/ (hook scripts)
- Scan
~/.claude/rules/ (rule files)
- Scan
~/.claude/skills/ (skill definitions)
- Scan
~/.claude/agents/ (agent definitions)
-
Score Dimensions (0-100 each)
- Token efficiency (model routing, compaction, thinking budget)
- Security posture (permissions, hooks, secret protection)
- Workflow coverage (skills, agents, commands for common tasks)
- Hook quality (performance, error handling, coverage)
-
Generate Report
- Overall score with letter grade
- Per-dimension breakdown
- Prioritized recommendations
- Quick wins highlighted
Output Format
## Harness Audit Report
Overall: **B+ (82/100)**
| Dimension | Score | Grade |
|-----------|-------|-------|
| Token Efficiency | 90 | A |
| Security | 85 | B+ |
| Workflows | 75 | B |
| Hooks | 78 | B |
### Recommendations
1. [Quick Win] Add CLAUDE_CODE_SUBAGENT_MODEL=haiku
2. [Strategic] Add database-reviewer agent
3. [Nice-to-have] Add Rust/Go language rules