skill-creator

To design and evaluate Claude skills, draft a YAML frontmatter with name and description, write a contextual SKILL.md body, define test prompts, and run iterative eval loops to validate triggering, performance, and stability across baselines.

Iterating on prompt performance requires running controlled experiments with defined eval prompts and benchmarks, then rewriting the skill based on results to optimize triggering accuracy and overall stability.

Yes, you need pyyaml installed to parse and enforce the YAML frontmatter structure required for skill packaging, which standardizes the name, description, and compatibility notes in the SKILL.md file.

You structure a SKILL.md file by writing a YAML frontmatter block containing the skill name and description, followed by a Markdown body that provides context, compatibility notes, and links to optional scripts, references, and assets.

Yes, you can run benchmark tests and generate reports by integrating evaluation tooling into the workflow, applying controlled test prompts to validate triggering criteria and performance before deployment.

You should update an existing Claude skill when triggering criteria misalign with practical workflows or benchmark results show performance drops, applying iterative eval loops to refine the description and SKILL.md body.