parallel-critic-agents-require-spot-verification
Dispatching 5 parallel critic agents produced one provably false finding (C2: exit codes) that the main agent initially confirmed before catching on re-verification. Critic agents work from file reads and pattern matching without runtime execution — they can misinterpret piped command output as the target script’s behavior. Always spot-verify the top 3-5 critical findings by direct execution before treating the full report as ground truth.