skill-improve

Automate iterative skill improvement by running a structured test-fix-retest loop that surfaces gaps in frontmatter and body structure, then applies targeted edits to drive quality. The workflow measures baseline test results, diagnoses deviations, applies fixes, and reruns tests.

Diagnose structural gaps in a SKILL.md file by capturing baseline static test results and identifying exact deviations in frontmatter, body structure, and category rubric. The diagnosis phase isolates specific issues before generating targeted edits.

You can run quality assurance tests on skills within CI-like or manual review workflows. The improvement loop applies to any Skill under .cursor/skills, enabling automated baseline testing, diagnosis, and retesting to confirm quality improvements.

Optional scripts and references are used to perform deterministic testing tasks on a skill. While the core workflow relies on the target SKILL.md and uses phases 1-7, these additional assets provide the necessary context for accurate diagnosis and retesting.

Targeted edits may fail to improve a skill if the baseline diagnosis missed structural deviations or if optional scripts lack deterministic references. The test-fix-retest loop requires re-running tests after edits to confirm that the applied fixes actually resolved the captured gaps.