Multi-agent cooperation through in-context co-player inference | ResearchPod