ibuildbots checks how an AI agent reacts when untrusted input contains hidden executable instructions. Each of five attacks runs in a fresh, single-use E2B microVM, and the test harness itself does not rely on a language model.
After the run you receive an ed25519-signed badge that is dated, hash-pinned and states how many attacks the agent resisted. A free local self-check is available if you want to test things yourself first.
