skill-creator

To create a Claude skill, you draft a SKILL.md file containing frontmatter metadata and a structured markdown body. This skill provides a guided workflow to capture intent and progressively build out the body instructions.

You improve skill triggering accuracy by running iterative evaluation loops and benchmarking test prompts. This workflow supports built-in testing patterns to refine the description and ensure repeatable, production-grade performance.

Production-grade skills rely on bundled resources including scripts, references, and assets. You wire up these optional directories within the skill structure to enable on-demand execution and provide supporting context.

Yes, skill development and evaluation require the anthropic and pyyaml dependencies. These libraries support the underlying design, testing, and markdown frontmatter parsing needed for the workflow.

The best way to structure SKILL.md frontmatter is by defining the skill name and description metadata. This metadata drives progressive disclosure and accurate triggering before users read the detailed body.

Inaccurate triggering usually means the SKILL.md description metadata is misaligned with user intent. You must iterate through evaluation loops, benchmark test prompts, and refine the frontmatter to fix the issue.