waza

Why we included this project

Waza is a Go CLI for the stage of agent development where demos stop being convincing: once a skill works on a happy path, you still need to prove it works reliably. It scaffolds a compliant skill structure, runs agentic test loops against real LLM execution, and scores the results with eleven pluggable validators covering code assertions, text matching, JSON schema, tool-call sequences, and LLM-powered grading. Test cases are defined in YAML, so the same suite can run against several models like GPT-4o and Claude, and the compare command shows which model handles your skill best. A local web dashboard shows transcripts and trends alongside per-task metrics, and a pre-configured GitHub Actions workflow makes it easy to run the same evaluations in CI. If you maintain agent skills or frameworks and want repeatable, cross-model quality checks rather than ad-hoc manual testing, this is a practical fit.