skill-improve

Automated baseline testing runs static and optional category checks to capture current failure counts, then analyzes each failing check to explain the exact gap in the skill definition before proposing fixes.

Run the test-fix-retest loop by executing a single command targeting your skill directory; the tool proposes precise edits, asks for user confirmation, applies changes, and retests to verify improvement.

Bash tools are optionally used to read and write skill files during improvement. The process requires executing test commands and modifies files in the .Codex/skills directory to apply targeted fixes.

If new edits do not improve the combined score, safe reversion automatically restores the original skill file, ensuring your skill definition is never degraded by a failed automated fix attempt.

Yes, automated improvement applies to any existing skill in the .Codex/skills directory, targeting static and category checks to iteratively improve the skill definition's quality.