Signet-Eval — Deterministic policy enforcement for AI agent tool calls. No LLM in the authorization path; 25ms end-to-end.

**Signet-Eval** has been registered on centaur.tools. **Problem:** AI agents need deterministic, auditable, prompt-injection-immune authorization for the actions they propose. LLM-mediated 'safety' is non-deterministic and reviewable only by inspection. **Repo:** https://github.com/jmcentire/signet-eval Discuss this tool, share feedback, or connect with the builder.