Google 论文提出 OO-Harness：按任务分化专用 agent harness，基准成绩提升 15.79 个百分点
Google 把 agent harness 写成类继承，一个任务一个专用版，AutomationBench 上比全局方法高了 13 分多，还能兜底陌生任务。
Google 发表论文，提出用面向对象方式演化 agent harness：OO-Harness 将 harness 写成代码类，各专用版本从父类继承、只改动所属任务簇需要的部分。路由器读取基座模型的一次普通运行结果，决定交给专用 harness 还是保留基座输出。在 AutomationBench 业务工作流上，OO-Harness 得分 77.41%，高于基座 harness 的 61.62% 和全局方法 Meta-Harness 的 63.92%。在陌生应用上它保留基座 harness，通过率维持 59.18%，而 Meta-Harness 跌至 44.90%。
New Google paper shows that agents score higher when each task gets a specialized harness from a shared family, not 1 global harness.
An agent harness is the prompts, tools and control code around a model. Most tuning methods evolve 1 global harness, though a fix for 1 task type can hurt another.
OO-Harness evolves harnesses as code classes, and each specialist inherits from a parent and changes only what its cluster needs. A router reads a normal base run and picks a specialist or keeps the base result.
On AutomationBench business workflows, it scored 77.41% versus 61.62% for the base harness and 63.92% for Meta-Harness, a global method. On unfamiliar apps, it kept the base harness and held 59.18% pass rate, while Meta-Harness's global harness fell to 44.90%.
For mixed workloads, keep the base harness as a fallback and route to specialists only when a run shows a known problem.
