skill-creator

To create and refine Claude skills, you draft an initial SKILL.md and run a structured evaluation loop that tests and iteratively improves skill prompts and descriptions for better triggering reliability.

Improving skill triggering reliability requires running structured evaluations, comparing iteration results against baselines, and refining the YAML frontmatter description based on actionable feedback.

You need a SKILL.md file with YAML frontmatter containing a name and description, alongside optional directories like scripts/, references/, and assets/ to support tooling and reporting.

You can use this evaluation loop for versioned experimentation to track iterations, compare results against baselines, and surface measurable improvements across AI tooling and product workflows.

Claude skills fail to trigger consistently when descriptions lack refinement; applying an evaluation-driven iteration loop drafts, tests, and refines prompts to produce robust, triggerable skills.

Structured evaluations for skill engineering require the pyyaml dependency to parse the YAML frontmatter in SKILL.md files and support the evaluation and reporting tooling.