skillgrade

mgechev/skillgrade

Skillgrade is a CLI tool for evaluating Agent Skills, essentially providing "unit tests" that verify whether AI agents correctly discover and use your skills. It scaffolds an eval.yaml with AI-powered tasks and graders, then runs trials against agents like Gemini, Claude, or Codex via Docker or locally, producing CLI or browser reports. It suits developers building and maintaining Agent Skills who need reliable pass-rate estimates or regression detection in CI.