skill-creator

Automating Claude skill creation and evaluation involves drafting skills, running structured evals with train/test splits, and packaging them for distribution. This workflow tracks iteration history and analyzes results to inform improvements.

Benchmarking Claude skills across iterations requires running structured evaluations that establish a baseline and track performance history. Analyzing these benchmark results identifies areas for refinement and triggers necessary safety checks.

Yes, you need YAML frontmatter to package Claude skills for distribution. The workflow strictly enforces including name and description fields within the YAML frontmatter before validating and packaging the skill for reuse.

Yes, you can run toxicity and safety checks during skill optimization. The iteration workflow analyzes results and evaluates toxicity as part of the automated cycle to ensure skills meet safety standards before packaging.

Tracking iteration history when refining Claude skills is handled automatically by the workflow. It records each drafting, evaluating, and tuning cycle, allowing you to analyze past results and measure improvements against the baseline.