A type system for LLM prompts. Define schemas, write behavioural contracts, stress-test with adversarial AI cases or your own ground-truth CSV, and diff fingerprints across versions. Catch prompt regressions before they hit production. Built on Claude Code. No API keys. No dependencies.