skill-improve

Automate continuous improvement for a SKILL.md file by running an iterative test-fix-retest loop that identifies gaps in static checks, generates targeted fixes, and retests to decide whether to keep or revert changes.

The best way to diagnose failing static checks is to run an automated baseline test, identify gaps against category rubrics, and apply targeted fixes that rewrite only the failing areas in the SKILL.md frontmatter.

Yes, automated skill testing requires a root SKILL.md file with valid frontmatter present in the directory, along with adherence to allowed-tools and safety constraints to run the baseline checks.

Yes, you can use optional scripts, references, and assets directories when improving a skill, as the automated process parses the skill name and processes these components during the baseline static checks.

The automated skill improvement loop reverts changes to SKILL.md when end-to-end retesting determines that the targeted fixes did not produce the desired quality gains or failed the category rubrics.

Limitations of running targeted fixes include strict adherence to allowed-tools and safety constraints, requiring valid frontmatter in the root SKILL.md, and the process only rewriting specific failing areas rather than the entire file.