Abstract: Focusing on the intelligent decision-making needs enabled by large language models in command and control, the difficulties and challenges of course of action in command and control are analyzed. A generation method of course of action based on mixture-ofagents is proposed. To address the capability limitations of a single large language model in the course of action, in the framework, it is suggested that multiple large language models are utilized to process different aspects of input information and to integrate them into the effective intelligence prompts. Then, multiple large language models are used to generate various proposals for the course of action based on the prompts, and finally, a single large language model summarizes multiple operational schemes. It is validated that the method has super feasibility and effectiveness.