skill-creator

To create and refine Claude skills, draft a SKILL.md file with structured frontmatter and body, then run trigger evaluations and benchmarking. Iterate by refining the content and re-running tests until triggering accuracy and performance meet your goals.

The best way to benchmark workflow automation prompt performance is to run trigger evaluations using built-in tooling that generates quantitative metrics, then iterate on the prompt engineering to optimize skill descriptions until constraints are satisfied.

An evaluation loop improves skill triggering accuracy by executing benchmark tests, measuring results, and feeding those metrics back into the iteration process. You refine the SKILL.md trigger description and re-run tests until performance targets are met.

Yes, pyyaml is required as a dependency to build and evaluate skills. It is needed to parse and manage the structured YAML frontmatter that defines the skill metadata within the SKILL.md entrypoint.

Yes, you can modify existing automation workflows. The skill creator supports iteratively improving existing skills by updating the SKILL.md body and frontmatter, then running the evaluation loop to measure the impact of your changes.

A skill description may not trigger correctly during benchmarking due to vague frontmatter or misaligned trigger descriptions. You must iterate by refining the SKILL.md content and re-running the evaluation scripts until the triggering accuracy meets your defined goals.