skill-improve

To fix failing static checks in a SKILL.md file, use a deterministic test-fix-retest loop that identifies specific frontmatter and body convention failures, proposes targeted fixes, and verifies the improvement through re-testing. Only changes that improve the overall validation score are kept.

A test-retest loop for skill improvement is an automated workflow that runs static validation checks on a skill file, applies targeted patches to failing sections, and then re-runs the tests to confirm the overall score has improved before permanently keeping the changes.

You can automatically revert skill file changes if validation scores drop by running an improvement loop that compares combined test scores before and after writing patches. If the score does not improve, the workflow executes a git revert to safely restore the previous version.

Yes, you can automate workflow checks for missing required verdict keywords in skill files by running guided diagnostics that detect specific gaps in required body conventions. The workflow then patches only those missing parts and verifies the fix through automated re-testing.

Yes, the skill improvement process requires explicit approval before writing changes to your SKILL.md file. This safe change control mechanism ensures you review targeted fixes, while offering an easy git revert if the re-test score does not improve.