skill-creator

To create and evaluate Claude skills, provide a high-level goal to draft an initial SKILL.md, generate test prompts, and run benchmarks for variance analyses. Iterate based on quantitative and qualitative feedback.

The best way to structure SKILL.md frontmatter is to include the required name and description fields at the root directory. This validates the skill unit and allows metadata extraction for categorization and triggering optimization.

Run benchmarks and variance analyses by executing evaluation plans generated during skill creation. This measures effectiveness through quantitative metrics and qualitative feedback to iterate and refine the prompt engineering.

Skill unit validation fails when required fields are missing because the system locates SKILL.md frontmatter to extract metadata. Units missing a valid frontmatter body or required name and description fields are flagged as invalid.

Yes, you can optimize skill descriptions to improve triggering accuracy. Iterate on the SKILL.md metadata block based on evaluation benchmarks to refine how effectively the skill is categorized and triggered.

Skill-creation works without external dependencies. The process drafts structured prompts, validates frontmatter, and assembles metadata blocks entirely within the environment to reduce time to a polished capability.