skill-evaluator

You can audit Claude Code skills by validating frontmatter fields, checking required files like SKILL.md, and evaluating content against a deterministic rubric across 8 quality dimensions to generate actionable findings.

A deterministic rubric for skill evaluation applies fixed criteria across 8 quality dimensions to validate metadata, structure, and content, ensuring consistent and repeatable quality audits for Claude Code skills.

Validating frontmatter fields in Claude Code skills involves checking for required metadata keys including name, description, argument-hint, allowed-tools, and effort, automatically identifying missing or invalid entries in the skill files.

Yes, you can generate a markdown findings report and a JSON sidecar for skill quality audits, providing a prioritised fixes list and a concise chat-ready summary suitable for CI pipelines and code reviews.

No, the skill evaluation audit keeps outputs read-only to the source skill, ensuring the original skill files within plugins/*/skills directories remain unmodified during the quality validation process.

Fast mode in skill quality audits is an optional setting that skips qualitative review to accelerate the validation process, focusing only on deterministic checks for metadata, structure, and required files.