In-loop hacking detection

Red-team lab: an agent writes its own verifier that only looks for a word in the source, and the pipeline records VERIFIED on an unfixed bug. Put a rule-based detector inside the loop so a flagged verifier is an instant failure, find the cheat the starter rules miss, add one rule that closes it without flagging honest verifiers, and see an unseen cheat still scan clean.

This Harness Engineering lesson is part of a paid plan. Sign in with a plan that includes this course to read it.