skill-creator

You create and iterate on Claude skills by drafting a SKILL.md file with frontmatter, running evaluation loops to test performance, and updating metadata and assets based on benchmark outcomes. This process automates end-to-end skill creation and continuous improvement.

The best way to automate skill evaluation is to run an evaluation-improvement loop that executes prompts, reviews outcomes, and updates skill descriptions. This generates eval schemas, history, and benchmark-ready artifacts for measurable performance tracking.

Frontmatter metadata in skill packaging enforces a standardized SKILL.md structure with required name and description fields. This metadata controls triggering descriptions and supports optional bundles like scripts, references, and assets for reusable tooling.

Yes, you need yaml to handle frontmatter parsing and metadata serialization when drafting SKILL.md files. This dependency is essential for structuring skill configurations and packaging optional bundles like scripts and references.

Yes, you can bundle scripts, references, and assets when building a new skill. These optional components accelerate future skill development by providing reusable tooling alongside the core SKILL.md file and evaluation schemas.

Your skill description may not trigger correctly if the frontmatter metadata is not optimized through evaluation loops. Iterative improvement reviews outcomes and updates the triggering description to ensure accurate skill activation.