skill-creator

Creating a new skill from scratch involves drafting a SKILL.md file with trigger-focused metadata and workflow instructions. This skill authoring process generates self-contained skills with trigger-focused instructions that activate accurately on relevant user requests.

Skill evaluation creates test prompts and runs baseline-versus-skill comparisons to grade results with quantitative assertions. This benchmarking process measures whether a skill performs better than the default baseline.

Improving trigger accuracy through description tuning involves revising the frontmatter description so the skill triggers more accurately on relevant user requests. This optimization ensures better activation matching for intended queries.

Yes, you can iterate on skill instructions by revising them based on feedback and rerunning tests to measure improvement. This iteration workflow supports benchmark-driven refinement to optimize skill performance quantitatively.

Yes, skill authoring supports evaluation workflows with structured grading, comparison analysis, and packaged skill delivery. These benchmarking features enable quantitative assertions and baseline-versus-skill comparisons.

The best way to package a self-contained skill after iteration is using the packaged skill delivery feature, which bundles the refined SKILL.md, scripts, and assets together for deployment after benchmark validation.