Skill Pruner

Keeps the skill surface tight by recommending removals based on real usage data.

When to use

• Monthly cadence
• Before adding a new skill (check for an existing one to extend)
• When skill-master routing feels slow

Process

1. Pull 30-day invocation report:

   python -c "from tools.eval_tracker import skill_usage_report; import json; print(json.dumps(skill_usage_report(days=30), indent=2))"
   

2. Categorize skills:
   • Hot (>10 invocations) — keep
   • Warm (1–10) — keep, possibly merge with neighbors
   • Cold (0 invocations) — propose removal or consolidation
3. For each cold skill, check git log skills/<name>/ to see if it's brand new (<14 days = give it more time)
4. Output: ranked recommendation list with reasoning

Output format

• Markdown table: skill | invocations | last modified | recommendation
• One-paragraph summary