Build and refine skills with evidence from real tests.
skill development
Creates, evaluates, benchmarks, and improves skills, including their triggering descriptions.
When to use it
Use when creating, editing, evaluating, benchmarking, or optimizing a skill.
Work with it in turns to define, draft, test, review, and iteratively improve a skill in the project.
This skill
Claude
Sends optimization data to Claude
documentation
Researches documentation through MCPs
evaluation viewer server
Starts a local review server
default browser
Opens the eval review page
Python runs the benchmark aggregation, review viewer, optimization, and packaging scripts.
The claude CLI, specifically claude -p, enables the optional description-optimization workflow.