skill-creator

To create and validate SKILL.md frontmatter, draft metadata fields like name, description, and compatibility, then run automated validation to enforce correct formatting. This ensures your Claude skill is properly structured before testing and packaging.

A trigger-evaluation loop tests skill descriptions using train and test splits to measure performance. By running evaluations and tracking history, you iteratively refine descriptions to improve how accurately Claude triggers the correct skill.

You package Claude skills into distributable .skill files after drafting and validating the SKILL.md content. This bundles validated metadata, evaluation history, and optional resources like scripts, references, and assets into a single distributable file.

Yes, YAML is a required dependency for automating the Claude skill lifecycle. It manages SKILL.md content and metadata throughout the drafting, evaluation, iteration, and packaging workflows.

Yes, you can generate lightweight benchmark reports to compare different skill iterations. The evaluation workflow tracks history and results, allowing you to compare performance across iterations and select the best performing skill description.

The main limitation is that automated evaluation workflows rely on train and test splits, meaning trigger accuracy depends heavily on test data quality. Additionally, packaging requires strict frontmatter validation, which can block distribution if formatting errors exist.