Aiming at the bottleneck problems such as the time-consuming analysis and representation of massive observation requirements, the low matching efficiency of large-scale Aerospace & Space sensing nodes, and the insufficient autonomous decision-making capability of Aerospace & Space resources in complex game-theoretic environments, an Aerospace & Space Task Planning Method Based on Large Language Model Agents is proposed. By deeply integrating large language models with cloud-edge collaborative technologies, the method enhances the efficiency of user requirement understanding and representation, as well as online task planning, thereby providing technical support for the optimization design and application of large-scale Aerospace & Space observation systems.