skill-creator

To create and test AI skills, you draft the SKILL.md file, run evaluation prompts, compare outputs against baselines, and refine trigger descriptions based on pass rates. This lifecycle takes a rough idea and validates it into a distributable skill package.

The best way to benchmark AI skills is to run evaluation prompts, compare skill-assisted results against baselines, and review quantitative pass rates and timing. This quantifies quality improvements during trigger optimization and instruction refinement.

Packaging and validating AI skills requires checking the skill structure, enforcing frontmatter validation, and bundling components like scripts and references into a distributable file. This ensures the skill triggers correctly before release.

Yes, skill-creator requires the pyyaml dependency to parse and enforce frontmatter validation during skill creation. This ensures your SKILL.md metadata meets structural requirements before benchmarking and packaging.

AI skill triggers fail to activate when description wording is too loose or exceeds description limits. You can fix this by iterating on trigger accuracy, tightening the description, and running benchmarks to validate improvements.

Yes, you can use skill-creator to refine existing skills by editing instructions, running evals, and comparing outputs. It coordinates SKILL.md authoring and benchmark generation for both new drafts and existing skill improvements.