skill-creator

To create and evaluate Claude skills, you draft prompts and SKILL.md files with YAML frontmatter, run trigger evaluations, collect results, and analyze them against baselines to drive iterations until the skill is optimized. This structured workflow ensures repeatable skill creation and refinement.

Prompt engineering iteration for higher triggering accuracy involves running trigger evaluations, comparing results with prior iterations or baselines, and refining the skill description. This cycle of testing and analysis drives optimized performance for the skill.

Yes, pyyaml is required to handle the YAML frontmatter specifying the skill name and description. This dependency supports parsing and packaging the SKILL.md files, alongside optional folders for scripts, references, and assets, into distributable formats.

Yes, the skill creation workflow supports building from scratch or improving existing skills. You can load an existing skill, run trigger evaluations, analyze the benchmark results, and iterate on the prompts and descriptions to refine its performance.

A packaged skill includes a SKILL.md file with required YAML frontmatter for name and description, alongside optional folders for scripts, references, and assets. These components support on-demand tooling and distributable skill formats.

Inconsistent trigger evaluation results often indicate an unoptimized skill description. By running evaluations, collecting results, and comparing them with baselines or prior iterations, you can identify description gaps and iterate to achieve consistent triggering accuracy.