FoH：基于反思自动演化分布式多智能体harness
原标题：Federation over Harness
AI 摘要
论文提出 Federation over Harness（FoH）框架，用于自动演化分布式多智能体推理流程中本地客户端与中央服务器两端的 harness。服务器智能体演化客户端 harness 以完成任务与自我反思，同时演化自身 harness 以聚合分布式反思、维护共享知识并更新后续客户端 harness。该方法无需真实标签或验证数据，仅依赖智能体自生成的反思进行学习。在多个基准上，FoH 相比固定 harness 的单智能体基线提升 53.57%，相比 harness 自演化提升 29.76%，并将任务完成 token 减少 34.09%。
正文节选
Federation over Harness Abstract Modern LLM-powered agents increasingly improve through experience sharing, where distributed agents distill local reasoning experience into reusable knowledge, such as a shared insight library. Existing approaches, however, still heavily rely on manually designed prompts and workflows that determine both how local agents extract useful reasoning traces and how a central/global agent aggregates them to obtain common patterns and insights. We propose Federation ove