Harness Diagnostics

Harness Diagnostics -- Agent Harness Self-Diagnostic Skill

12-principle based harness maturity diagnostics and improvement suggestion tool. Read-only: Performs diagnostics + suggestions only; does not directly modify files.

Input

$ARGUMENTS

Mode Selection

Default to Audit mode if no keywords match.

12-Principle Scoring System

Each principle scores 0-8, total 100 (12 x 8 = 96, remaining 4 points as overall bonus).

Maturity Levels

Execution Procedure

Setup Mode

1. Explore project root (Explore sub-agent)
2. Check existence of these items:
   • CLAUDE.md or .claude/CLAUDE.md
   • rules/ or .claude/rules/
   • docs/ directory
   • Package manager lock files
   • Linter/formatter configuration
   • Test framework configuration
   • CI/CD configuration
3. Recommend creation for missing items (do not create directly)
4. Recommendation priority: P0 (immediate) -> P1 (recommended) -> P2 (improvement)

Audit Mode

1. Run checks for each of the 12 principles:
   • File/directory existence (Glob)
   • Configuration file content analysis (Read)
   • Pattern search (Grep)
   • Hook checks cover all hooks registered in settings.json — however, notchi-hook.sh is optional (notification-only), so its absence is not penalized
2. Score each principle (0-8)
3. Calculate overall bonus (0-4):
   • Synergy between principles (e.g., hooks + rules + docs all present: +2)
   • Consistency (naming conventions, structural uniformity: +2)
4. Output report

Maintenance Mode

1. Detect stale items:
   • docs/ files unmodified for 30+ days
   • Dependencies in package.json not imported anywhere
   • Collect TODO/FIXME comments
   • Empty directories
   • Unused configuration files
2. Detect drift:
   • CLAUDE.md content vs actual project structure mismatch
   • rules/ rules vs code reality mismatch
3. Output cleanup suggestion list (do not modify directly)

Output Format

# Harness Diagnostics Report

## Project: {project name}
## Mode: {Setup | Audit | Maintenance}
## Date: {YYYY-MM-DD}

---

## Per-Principle Scores

| # | Principle | Score | Rationale |
|---|-----------|-------|-----------|
| 1 | Agent Entry Point | X/8 | ... |
| 2 | Map, Not Manual | X/8 | ... |
| ... | ... | ... | ... |
| 12 | Self-Documentation | X/8 | ... |

**Overall Bonus**: X/4
**Total Score**: XX/100

## Maturity Level: LX {level name}

---

## Top 3 Improvement Suggestions

### 1. {suggestion title} (P{0-2})
- **Current**: {current state}
- **Improvement**: {specific action}
- **Impact**: {expected score increase}

### 2. ...

### 3. ...

---

## Detailed Analysis
{per-principle detailed description -- as collapsible sections}

Constraints

• Read-only: No file creation/modification/deletion
• Diagnostic tools: Glob, Grep, Read, Bash (read-only commands only)
• Output provided as markdown report only
• Instead of direct modification, suggest in "doing X will improve score by Y" format
• Leverage sub-agents to protect main context