A model-agnostic, Jev-first Responsible AI harness for assessing AI systems for prompt injection, secret/PII leakage, unsafe tool use, and policy bypass. Runs portable tests, captures evidence, scores risk with calibrated uncertainty, and routes serious or unclear findings to human review.
Stars measure the whole repository, including work unrelated to Jev.
Documented records the linked documentation or source. JevHunt has not independently run or benchmarked this project.
Non-macOS or CI: set `TYPESAFE_API_KEY` in your own environment/secret manager instead — `assess-jev-live.ts` prefers that env var whenever it's explicitly set, and only falls back to Keychain when it isn't.
Evidence commit: b1fb72ab44ecead5700a46b3d49063c352118646
Discovered through: github-search.
How we review →