多智能体工作流的变异测试
原标题：Mutation Testing for Multi-Agent Workflows
AI 摘要
该论文将变异测试引入多智能体 LLM 工作流，定义了四类共 14 个变异算子（委派、允许工具、限制、智能体/指令）和三个杀死判据（K-Struct、K-Outcome、K-Safety），并在 10 个基于 OpenAI Agents SDK 的工作流上执行 29,601 次智能体运行、评估 1,084 个变异体。结果显示，最佳基线测试套件 T-Dev⋆ 的变异得分仅为 0.717（K-Struct）和 0.559（K-Safety），限制移除类故障的检出率低至 0.19 至 0.28，说明任务成功型测试套件只能捕获约一半的权限违规。
正文节选
Mutation Testing for Multi-Agent Workflows Abstract Multi-agent large language model (LLM) systems are defined by explicit workflows, agents, tools, tool-access rules, and delegation paths. Most evaluations, however, report end-to-end task success rather than whether a test suite can detect faults in this coordination structure, such as a removed delegation, a silently removed restriction, a misrouted tool edge, or a faulty tool implementation. We adapt mutation analysis to this setting, fourtee