Openkata Optimize Skills

Optimize Skills

Improve skills to 95%+ review score with passing evals.

Usage

optimize skills

Workflow

1. Select targets — Ask the user: "All skills below 95%, or a specific skill?" If all, read the score field from each skills/*/.tessl-plugin/plugin.json and list those below 95%. If specific, use the named skill.
2. Optimize — For each target, follow openkata-review-skill (lint, review, optimize, checklist) then openkata-eval-runner (generate scenarios, run evals). Iterate until both pass 95%+, maximum 3 iterations. If still below after 3 rounds, report the final score and stop. When multiple skills are selected, run them as parallel subagents.
3. Learn — If you discovered a pattern that consistently improves scores, update skills/create-skill/SKILL.md so future skills benefit.
4. Commit — One commit per skill.

Constraints

• Run tessl commands without asking for confirmation.
• Maximum 3 optimization iterations per skill.
• Do not add repo-internal boundaries (publishing, releasing, tagging) to distributable skills. Those belong in openkata-skill-conventions.
• After optimization, persist the final score to .tessl-plugin/plugin.json.

Boundaries

DOES:

• Read scores from plugin.json to find targets
• Run tessl lint, review, optimize
• Generate eval scenarios and run evaluation
• Update create-skill with learned patterns
• Commit improvements per skill

Does NOT:

• Create new skills
• Publish or release skills
• Modify skills scoring 95%+