Language:JapaneseEnglish

skill-creator Claude Code Skill

Author
Anthropic
Topics
Claude Code Customization & Workflow · Code Quality, Review & Testing
First cataloged
2026-07-09 (UTC)
Explanation last updated
2026-09-03 (UTC)
Source (GitHub) last updated
2026-04-20 (UTC) (138 days ago)

The explanation below is AI-generated. Please verify it against the sources.

This skill helps users create new Claude skills and iteratively improve existing ones through a structured test-and-review process. According to SKILL.md, it guides drafting a SKILL.md file based on an interview about intended behavior and triggering conditions, then generates test prompts that are run via subagents to compare skill-assisted output against a baseline. Outputs are graded against defined assertions and aggregated into benchmark data covering pass rate, execution time, and token usage. A review viewer is launched so the user can inspect results and leave feedback, which feeds back into further revisions. The skill also includes a separate procedure for optimizing a skill's description to improve triggering accuracy, and it adapts its steps depending on whether it is running in Claude Code, Claude.ai, or Cowork.

What you can do with skill-creator

  • Draft a new skill's SKILL. md based on an interview covering intended behavior, triggering conditions, and expected output format
  • Generate realistic test prompts and run them through subagents, comparing skill-enabled output against a baseline
  • Grade outputs against defined assertions and aggregate results into benchmark data (pass rate, timing, token usage)
  • Launch a review viewer to collect qualitative feedback on outputs and iterate on skill improvements
  • Package a finished skill into a distributable . skill file

Sources

Back to list