Google 开源 EnvHarness：让智能体在动态环境中自进化学习
原标题：Google EnvHarness Enables Self-Evolving Agents to Learn in Dynamics Environments
AI 摘要
Google 研究人员开源了 EnvHarness 框架，它通过标准 reset()/step() 接口包裹静态环境，在不修改智能体、环境或评估系统的前提下动态调整训练环境。该框架包含 setup、rule、link 三类可组合组件，并引入 EnvRigger 自动分析智能体弱点并生成针对性修改。在 WebArena、SWE-bench Verified 和 ALFWorld 等基准上，多个模型性能提升 2.9 至 3.7 个百分点。
正文节选
Google researchers open sourced EnvHarness, a framework that dynamically adapts agent-training environments to an agent's weaknesses without modifying the agent, environment, or evaluation system. According to the researchers, this approach can improve both learning effectiveness and efficiency while helping agents overcome their weaknesses. According to the research authors, self-evolving agents typically improve by rewriting their prompts and acquiring new skills, but rarely adapt the environm