skill-creator

Skill development requires defining user intent, drafting the SKILL.md frontmatter and body, and orchestrating evaluation loops. You iterate on the skill by measuring trigger accuracy and performance against benchmarks to generate data-driven improvements for distribution.

A SKILL.md file structures a Claude skill through frontmatter and procedural instructions loaded into context upon activation. It works with automation by integrating with evaluation tooling to measure trigger accuracy and generate data-driven improvements.

You evaluate trigger accuracy by running built-in evaluation scripts that aggregate results. Benchmarking compares these results against baselines or prior iterations to measure performance and guide data-driven improvements.

Yes, skill development requires the yaml dependency to parse the SKILL.md frontmatter. This ensures the skill configuration is correctly formatted for compatibility and proper context loading when activated.

You package skills by validating the finalized SKILL.md and exporting it as a distributable .skill file. This enables reuse and distribution across teams within different environments.

You should iterate on a skill when evaluation results show poor trigger accuracy or performance below baselines. Use the benchmark data to refine the SKILL.md instructions and improve overall skill performance.