verdict

Provides structured, sandboxed test and lint feedback for coding agents, returning compact typed verdicts with failure fingerprints instead of raw runner output