skill-improve

Automating iterative SKILL.md improvements requires running a test, diagnose, propose, write, and retest workflow. This process applies to individual skill directories under .claude/skills to reduce static and category test failures.

To reduce static analysis failures in a skills catalog, run baseline static tests to record failure counts, apply targeted patches to fix exact gaps, and rerun tests to verify score improvements.

Git version control secures automated skill testing by recording the original file before applying edits, then reverting changes via git operations if the combined test score does not improve after retesting.

Yes, you can revert SKILL.md changes if test scores do not improve. The workflow records the original file, applies approved edits, reruns tests to compare scores, and reverts via git if scores worsen.

Yes, you need read and write access to the skill file, the ability to run /skill-test static and /skill-test category commands, and git operations to save or revert changes during test-fix-retest cycles.

Maintainers should use automated diagnosis for skill file gaps when baseline tests identify failing checks, allowing the system to pinpoint exact gaps and propose minimal before/after edits for approval.