SkillProof tested Claude skills

Optimize Skill

Eval-gated loop to improve a skill's SKILL.md without regressions

di Bennyoooo · Bennyoooo/skillmaxxing

Tested · Didn't pass

Optimize Skill — Eval-gated loop to improve a skill's SKILL.md without regressions

Cosa fa

Runs a rollout-score-reflect-apply-validate-promote loop against an existing skill's eval.yaml, capping edits per step and blocking promotion unless a candidate strictly beats the current version with no held-out regression. Triggers when a skill underperforms or the user asks to optimize/tune/improve a skill that already has an eval set.

Rapporto di test

Copying just the skill folder (the only GitHub-only install path available) leaves scripts/optimize.sh unresolved — verified with a real install: `bash scripts/optimize.sh score ...` from inside the copied skill dir exits 127, 'No such file or directory', because the actual script lives one level up at skill-maxing-plugin/scripts/, not inside optimize-skill/.

Testato il: 2026-07-15 · Claude Code 2.x (agent harness)

Installazione

git clone https://github.com/Bennyoooo/skillmaxxing
cd skillmaxxing
mkdir -p ~/.claude/skills
cp -r skill-maxing-plugin/skills/optimize-skill ~/.claude/skills/optimize-skill

Comandi e prompt di esempio

  • /optimize-skillEval-gated loop to improve a skill's SKILL.md without regressions

Gli skill si attivano con richieste in linguaggio naturale, senza comandi da ricordare. Dopo l'installazione, prompt come questi lo attivano (in inglese):

  • My PR-review skill has an eval.yaml but keeps failing one test case
  • Optimize this skill's prompt against its eval set without breaking passing cases
  • This skill scores badly on its evals, improve it without regressing the rest

Rapporto completo →