skillbench
par g9pedro
Git & GitHubRésumé
Track skill versions, benchmark performance, compare improvements, and get self-improvement signals.
Description
Ce que fait ce skill
skillbench is a CLI tool for tracking AI agent skill performance across versions. It records task outcomes, scores skills on a graded scale, and surfaces which skills are declining, stale, or need attention. Benchmarks can be collected manually or pulled automatically from tasktime.
Combines version-aware benchmarking, graded scoring, regression detection, and improvement suggestions in one tool built around the AI agent skill workflow rather than general software testing.
Quand l'utiliser
- Comparing a skill's success rate before and after an update
- Flagging which skills have dropped below 70% success rate
- Auto-recording task duration via tasktime integration without manual timing
- Running baseline checks in CI to catch skill regressions
- Generating a performance dashboard to review agent health across all skills
Exemple de workflow
User asks: Check if my github skill improved after updating it to v1.1.0
- 1
Sets active version with
skillbench use github@1.1.0 - 2
Records a task outcome after completing work:
skillbench record "Create PR" --success - 3
Runs a version comparison:
skillbench compare github@1.0.0 github@1.1.0 - 4
Reviews the score breakdown across success rate, duration, consistency, and trend
- 5
Checks improvement suggestions with
skillbench improve githubif grade is still low
Side-by-side grade comparison showing whether v1.1.0 improved or regressed across each scoring dimension
Installation
clawhub install g9pedro/skillbenchExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.