skill-creator

You can automate Claude skill creation by running an eval-driven loop that drafts SKILL.md files, runs benchmarks, and iteratively refines content until triggering accuracy is maximized. This structured workflow enables measurable improvement through repeated testing.

An eval-iterate loop for prompt engineering is a repeatable workflow that drafts a skill, runs controlled evaluations, generates benchmarks, and refines the prompt based on metrics. It enables rapid experimentation and measurable improvement for Claude skills.

You can benchmark and evaluate existing Claude skills by running controlled evaluations that compare performance against baselines. The process generates metrics, extends test sets, and updates SKILL.md frontmatter based on feedback to maximize accuracy.

Yes, pyyaml is required as a dependency to create and test Claude skills. It is needed to parse and manage the SKILL.md frontmatter configuration that controls skill triggering and evaluation behavior.

Yes, you can automatically extend test sets, read references, and manage bundled assets to scale skill quality during the iteration loop. This allows you to broaden evaluation coverage while maintaining organized skill resources.

The best way to refine a SKILL.md file is to update its frontmatter and body content based on evaluation feedback and benchmarking metrics. Repeating this draft-eval-improve loop stabilizes results and maximizes triggering accuracy.