An autonomous attack hypothesis generation system that reasons about target architecture from limited observations, generating and prioritizing testable vulnerability hypotheses using local LLM agents that compete and cooperate in structured reasoning tournaments.