ck:skill-creator

Automating Claude skill creation involves using an eval-driven workflow to initialize skill templates, run parallel baseline evaluations, and iterate on descriptions to ensure consistent trigger behavior. This process validates frontmatter and structure before packaging.

Benchmarking Claude skills requires running parallel with-skill and baseline evaluations, aggregating metrics, and generating human-readable reports. Post-hoc analyzer and comparator agents identify improvements for trigger precision and deterministic behavior.

Packaging Claude skills for distribution uses validation scripts to check frontmatter and structure integrity. You initialize a skill template, validate the folder structure, and run packaging scripts to produce a distributable skill package.

Yes, the anthropic dependency is required to run model-driven description refinement and evaluation workflows. It provides the necessary framework to execute parallel evaluations, grade outputs, and optimize trigger behavior.

Claude skills fail to trigger correctly when descriptions lack precision or structure is invalid. Running trigger tests, grading outputs against baselines, and applying model-driven description refinement improves trigger accuracy and deterministic execution.

Yes, you can optimize skill descriptions automatically through model-driven description refinement. The improvement loop runs trigger tests, evaluates outputs, and iterates descriptions to achieve benchmarked quality and security criteria.