skill-creator

To create a new Claude skill, define your intent, constraints, and expected outputs, then generate a SKILL.md file with a practical workflow. This process drafts the skill instructions and sets up the initial trigger description for invocation.

Benchmarking improves triggering accuracy by running baseline and with-skill comparisons across test prompts. It grades expectations with evidence and aggregates results to refine the SKILL.md frontmatter description for higher trigger precision and recall.

Yes, you can optimize an existing skill description by evaluating should-trigger and should-not-trigger query sets. This updates the SKILL.md frontmatter description to ensure the skill is invoked at the right times without altering core behavior.

The evaluation loop for iterating on a Claude skill involves generating test prompts, executing baseline runs, grading expectations with evidence, and rewriting instructions based on qualitative feedback and quantitative benchmark results to expand coverage.

No prompt engineering experience is required to build skills with Claude Code. The skill authoring process translates your stated goals and constraints into structured SKILL.md instructions and automatically generates the necessary evaluation prompts.

If your Claude skill is not triggering, the SKILL.md frontmatter description lacks sufficient precision. Running trigger optimization across should-not-trigger and should-trigger query sets refines the description to improve recall and correct invocation.