Andrej Karpathy's LLM Council sends one question to several models, has them rank each other's answers anonymously, then a chairman model writes the final answer. Karpathy calls the repo a Saturday hack he won't support, so treat it as a method to copy; a Claude skill version runs the same flow inside Claude with five advisors (Contrarian, First Principles, Expansionist, Outsider, Executor). Neither repo has a license, and the Claude version costs about eleven agent runs per question.