LLM 软件工程中的测试驱动方法：阶段、任务与智能体技能综述
原标题：Test-Driven Approaches to Software Engineering with Large Language Models: A Survey of Phases, Tasks, and Agent Skills
AI 摘要
一篇 arXiv 综述论文系统梳理了大语言模型驱动软件工程中的测试驱动方法，围绕「测试改变了什么决策」这一核心问题，整合了 87 条研究记录（其中 83 条做了方法或协议级提取）及 5 项实践资源。论文区分了 Red–Green–Refactor 循环与测试条件生成、执行引导精炼、测试中介分析和仅评估测试等不同机制，并跨代码生成、修复、翻译、重构、克隆检测、代码搜索、定位、训练数据构建和形式化规范验证等任务进行比较。作者强调测试通过本身不能证明行为等价、反馈有效或流程合规，并提出 oracle 验证、因果评估、长周期维护和可复用测试驱动智能体能力等研究方向。
正文节选
Test-Driven Approaches to Software Engineering with Large Language Models: A Survey of Phases, Tasks, and Agent Skills Abstract Tests increasingly participate in the decisions made by large language models and software engineering agents. They specify intended behavior, guide program construction and repair, select candidates, constrain transformations, and provide execution evidence for software analysis. These uses draw on test-driven development, yet differ substantially in test order, oracle