skill-creator

To create and iterate on Claude Skills, draft a SKILL.md file and run automated evaluation loops to measure triggering accuracy and resource usage. This process refines prompt engineering workflows and outputs packaged skill metadata for deployment.

A SKILL.md file formalizes Claude Skill prompting and workflows. It requires name and description frontmatter, with optional dependencies and components directories, to produce packaged skill metadata that passes toxicity checks.

Yes, skill-creator supports benchmarking and description optimization. It runs multi-stage evaluation loops that reveal triggering accuracy, timing, and resource usage, guiding improvements for a more trigger-friendly skill.

Yes, you can use yaml dependencies when building prompt engineering workflows. The skill lifecycle supports yaml configurations to manage evaluation sets and iterate toward reliable skill triggering.

You measure triggering accuracy and resource usage by running automated evals and benchmarks within an evaluation loop. This reveals timing metrics and guides iterations to optimize the skill's triggering description.

The best way to package skill and evaluation metadata is to formalize the skill in a SKILL.md file, run evaluation loops, and include optional scripts, references, and assets directories to output packaged metadata for deployment.