Build and optimize AI skills with evals

Create and iteratively improve skills with eval loops and performance benchmarking

Building new skills when you need structured eval loops before production deployment.

• · Skill intent and trigger conditions
• · Test prompts
• · Evaluation criteria

• · Skill implementation
• · Eval results with metrics
• · Optimized skill description

• · vitest
• · eval-viewer script

Understand what the skill should do; ability to define test cases; eval framework installed

Skill intent too vague; insufficient test coverage; eval metrics don't match actual performance; description doesn't trigger

• · Iterative refinement loop documented
• · Quantitative evals run in background
• · Skill description optimizer included