验证工具覆盖面决定AI编码智能体价值：一项受控研究
原标题：The reach of a verification tool decides its value: A controlled study of verification surface, artifact quality, and cost in AI coding agents
AI 摘要
该研究通过构建一个工具列表为唯一变量的最小化编码智能体，在六个模型和八种工具配置下实现了1116个Web应用，以探究验证工具覆盖面（如linter、启动探针、shell、截图）对AI编码智能体输出软件质量的影响。研究发现，验证工具的最廉价收益是确保应用能启动，仅一个启动探针即可消除大部分启动失败，且成本仅为完整shell的35%；完整shell将无工具成本提高2.35倍，而截图仅在可见错误（如元素布局）上有微小提升，对仅可测量的失败（如滚动性能）无帮助。结论是验证工具仅在覆盖实际失败模式时才能改善输出。
正文节选
[Page 1] The reach of a verification tool decides its value: A controlled study of verification surface, artifact quality, and cost in AI coding agents Achint Mehta Corresponding author: Achint Mehta (e-mail: achintmehta@gmail.com) This work received no external funding. ABSTRACT Modern artificial-intelligence coding agents can be equipped with tools for checking their own work e.g. a linter, a boot probe, a shell, a screenshot tool. We call this set the agent's verification surface. This s