OpenAPI for agent actions, so hallucinations can't write (opens in new tab)

The open standard for how AI agents act on real systems: every write goes propose → approve → commit → rollback. 0.00% unauthorized writes across 4,216 injection evals.