Google FlowAgent:CI中基于ReAct的自动程序修复实践
Great paper from Google on program repair agents inside CI.
这是AI Agent在软件工程领域落地的硬核一手实践,给出了完整的架构设计与大规模生产环境数据,做DevOps或Agent开发的同学值得收藏参考。
Great paper from Google on program repair agents inside CI.
Google 关于 CI 中程序修复智能体的优秀论文。
(bookmark it)
(收藏它)
They propose FlowAgent, which runs a ReAct-style generate-and-validate loop on pre-submit test failures and shows the fix inside Google's code review tools.
他们提出了 FlowAgent,它在预提交测试失败时运行 ReAct 风格的生成与验证循环,并在 Google 的代码审查工具中展示修复方案。
Two abstention filters decide whether a fix is worth showing. One runs before execution and one after, so weak suggestions never reach the developer.
两个弃权过滤器决定修复方案是否值得展示。一个在运行前执行,另一个在运行后执行,因此弱建议永远不会到达开发者手中。
In a manual review of 195 real failures, 67.18% of its fixes were correct.
在对 195 个真实失败的 manual review 中,其 67.18% 的修复是正确的。
After the Google-wide launch, it suggested fixes on 295,508 changes. Developers previewed 65,069 and applied 28,554.
在 Google 全公司范围发布后,它在 295,508 个变更中建议了修复方案。开发者预览了 65,069 个并应用了 28,554 个。
Paper: https://arxiv.org/abs/2610.07289
论文:https://arxiv.org/abs/2610.07289
Chat with Paper: https://academy.dair.ai/papers/catching-developers-in-the-flow-low-latency-agentic-program-repair-at-google-sca-2610.07289
与论文对话:https://academy.dair.ai/papers/catching-developers-in-the-flow-low-latency-agentic-program-repair-at-google-sca-2610.07289
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力