Harness-S：从兄弟 Harness 学习可泛化 Agent 训练
原标题：Paper page - Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training
AI 摘要
该论文提出 Harness-S，一种选择性 harness 救援训练方法：当基础执行系统下的强化学习 rollout 全部失败时，调用由方向性 harness 演化构建的兄弟 harness 池生成新轨迹来替换失败组，并通过自适应 harness 路由器选择 Top-k 变体。在 AppWorld、SWE-Gym 和 DBBench 上，训练后的策略回到原始基础 harness 时显著优于仅策略训练，表明 harness 激发的能力可迁移进策略本身。
正文节选
Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training Abstract A selective harness-rescue training method uses specialized sibling harnesses to replace failed reinforcement learning rollouts and transfer their benefits into a single policy that performs better under the original base harness. Harness evolution from a common base exhibits Locality, Complementarity, and Nonstationarity: variants repair different failures, and their utility changes as the policy learns. Rather