skill-creator

To create and refine Claude Code skills, draft a concept, generate SKILL.md with frontmatter, wire bundled resources, and run trigger-evaluation loops to iteratively improve prompts based on quantitative metrics.

Writing a SKILL.md file involves drafting frontmatter metadata, writing the Markdown body instructions, and wiring bundled resources. The skill package orchestrates this creation process to capture user intent and generate testable skill components.

Prompt engineering evaluation works by running trigger-evaluation loops that collect metrics, compare results with baselines, and refine descriptions and prompts iteratively. This generates structured artifacts like eval prompts and performance metadata for repeatable benchmarking.

Yes, PyYAML is required to generate Claude Code skill packages. This dependency supports the parsing and creation of SKILL.md frontmatter, enabling the orchestration of discovery, evaluation, and iterative improvement workflows.

The best way to iteratively improve skill descriptions and prompts is by running trigger-evaluation loops, collecting quantitative metrics, comparing with baselines, and refining based on edge cases and user feedback to ensure benchmark readiness.

Claude skills may not trigger correctly if descriptions and prompts lack precision. Running trigger-evaluation loops identifies edge cases, collects metrics, and compares with baselines to iteratively refine prompts until trigger conditions match reliably.