netanel-abergel 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-masterrouting feels slow
Process
- 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))" - Categorize skills:
- Hot (>10 invocations) — keep
- Warm (1–10) — keep, possibly merge with neighbors
- Cold (0 invocations) — propose removal or consolidation
- For each cold skill, check
git log skills/<name>/to see if it's brand new (<14 days = give it more time) - Output: ranked recommendation list with reasoning
Output format
- Markdown table: skill | invocations | last modified | recommendation
- One-paragraph summary