skill-creator

To create and iterate on Claude skills, draft a SKILL.md file with YAML frontmatter specifying the name and description, then run an evaluation workflow to test performance and refine the skill based on benchmark results.

Building a new skill requires a SKILL.md file containing YAML frontmatter with name and description fields. You can optionally add scripts, references, and assets directories to support the skill's core workflow and evaluation processes.

Benchmark and optimize skill triggering descriptions by running the built-in evaluation workflow, analyzing the results, and iteratively refining the description in the SKILL.md frontmatter to improve activation accuracy and performance.

Yes, you can edit and improve existing skills by modifying their SKILL.md files, updating scripts and references, and running the evaluation workflow to measure performance changes and ensure the updated skill meets benchmark targets.

No external dependencies are required to run skill evaluations. The skill includes built-in scripts like the eval-viewer to present evaluation results and streamline the feedback loop for iterative improvement.

The best way to measure AI skill performance is to apply an iterative cycle of drafting, testing, and evaluating using the built-in eval workflow, then analyze the benchmark results to optimize the skill's triggering description and capabilities.