skill-improve

To fix failing Claude Code skills, you use an automated test-fix-retest loop that runs static test suites, proposes targeted edits for missing frontmatter or weak structure, and validates improvements through re-testing. The workflow keeps changes only if the combined failure count decreases.

A test-fix-retest loop for skill refactoring measures baseline failures, applies minimal targeted edits to resolve specific gaps, and re-runs tests to validate that quality scores improved. It ensures passing sections remain untouched while fixing structural quality and rubric conformance issues.

Static analysis and category rubric validation are run by executing the skill testing framework, which optionally applies category rubrics when a skill declares a category. It checks frontmatter completeness, argument-hint correctness, and end-to-end rubric conformance across iterative workflows.

Yes, change discipline and rollback mechanisms automatically revert skill updates if the combined failure count does not improve after re-testing. This ensures only modifications that demonstrably enhance structural quality and rubric conformance are kept in the final skill file.

Yes, the improvement workflow runs static test suites on any skill and optionally applies category rubrics only when a category is declared. Skills without a category still undergo frontmatter completeness, argument-hint correctness, and structural quality checks during the test-fix-retest cycle.