ck:skill-creator

You optimize Claude skill descriptions by running trigger evaluations, comparing results, and iterating on descriptions using train and test splits. This eval-driven loop refines descriptions against real test data to improve trigger reliability.

The best way to package a Claude skill for distribution is to validate frontmatter, check folder structure, and verify distribution readiness. This ensures your skill structure, metadata, and bundled scripts are ready for reliable deployment.

Benchmarking for skill authoring works by aggregating run metrics, generating reports, and inspecting outputs with viewer tools. This process helps evaluate trigger performance and iterate on structured instructions.

Yes, you need the Anthropic dependency to create and test Claude skills. The skill authoring workflow depends on Anthropic to run trigger evaluations, benchmark outputs, and iterate on metadata optimization.

Yes, you can initialize a skill folder with the correct structure and metadata automatically. The skill authoring process sets up the required folders, structured instructions, bundled scripts, and references for reliable distribution.