Collaborative Multi-Agent Scripts Generation for Imperfect-Information Reasoning in Murder Mystery Games

arXiv·low signalXBlueskyLinkedInCopy link

Zhong et al. (arXiv 2604.11741, April 14) address multi-agent collaboration under imperfect information using murder mystery games as a testbed. Vision-language models degrade in complex multi-hop reasoning under multiplayer game conditions, and the paper proposes collaborative script generation to enhance reasoning. While the domain is games, the underlying challenge — agents reasoning under information asymmetry with hidden state — maps directly to enterprise multi-agent scenarios where agents have partial views of organizational knowledge.