Run static and category test suites, diagnose failures, and apply targeted edits.
npx skills add https://github.com/WindXRan/y3td --skill skill-improve-windxran
Please help me install this Agent Skill. Skill: skill-improve Source: https://github.com/WindXRan/y3td/tree/main/.claude/skills/skill-improve Command: npx skills add https://github.com/WindXRan/y3td --skill skill-improve-windxran
This Skill reduces the time and risk of improving an existing skill by guiding an AI through a controlled test → diagnose → targeted edit → retest loop, then keeping only changes that improve the score.
Run the improvement loop for the tech-debt skill by issuing: /skill-improve skill-improve
High-intent search queries and answers about installing and using this skill.
You can improve an existing skill by running a test-fix-retest workflow that executes static and category test suites, diagnoses failures, applies targeted file edits, and reverts changes if the combined failure score does not improve.
The test-fix-retest workflow diagnoses failures by parsing exact check gaps from test suites, proposing targeted file edits to fix those specific gaps, and asking permission before writing changes to the skill file.
Yes, the skill improvement workflow runs static checks that validate YAML frontmatter and other structural requirements, then proposes scoped edits to resolve those specific warnings before re-testing the skill.
The best way is a score-based keep-or-revert decision process that re-runs tests after edits and uses git revert to discard changes if the combined failure count does not decrease, ensuring only measurable quality gains are kept.
Git is required for the revert mechanism, which discards proposed file edits if the combined test failure count does not improve after re-running static and category test suites on the modified skill.
Automated skill improvement loops are limited to skills located at .claude/skills/[name]/SKILL.md and rely on test failure counts to judge quality, meaning edits that pass tests but degrade overall behavior may still be kept.
lawless-mOrchestrate test-fail-issue-fix loops with GitHub issues and Claude Code.
solace-illRun iterative test-fix-retest loops to improve local skill definitions.
erenisiklarCreate Semgrep rules with YAML and test files for validation.
easonlaoRun static and category tests on SKILL.md, diagnose failures, apply edits, and retest.
