AI 智能体双域 PR 的因果研究：部署基础设施共修改与审查绕过
原标题：Co-Modifying Application Code and Deployment Infrastructure: A Causal Study of AI Agent Pull Requests
AI 摘要
该研究基于 AIDev v5 与 MOSAIC-3M 两个大规模数据集，对 26,999 个 AI 智能体提交的 Pull Request 进行因果分析，考察同时修改应用代码与部署基础设施的「双域 PR」。结果显示双域 PR 仅占智能体 PR 的 4.48%–4.55%，但产生两极分化后果：被显式审查的贡献中位延迟增加 13.13 小时，而 43.1%–56.2% 的双域 PR 绕过人工审查、中位不到 31 分钟即被静默合并，智能体身份使人工审查几率降低 63.7%。作者指出这会造成循环验证失败并危及软件供应链安全，呼吁建立部署感知基准与基于路径的访问控制。
正文节选
Co-Modifying Application Code and Deployment Infrastructure: A Causal Study of AI Agent Pull Requests Abstract. Current software engineering agents are predominantly evaluated inside isolated sandboxes against synchronous, local tests. However, deploying features in collaborative repositories requires modifying continuous integration pipelines, container configurations, and provisioning manifests alongside application logic. Because agents lack deployment awareness, attempting these cross-domain