Multi-Agent Step Race Benchmark: LLM Collaboration and Deception Under Pressuregithub.com/lechmazur7 pointszone411a year ago