实证研究：Agentic GUI软件中工具介导的交互失败
原标题：Understanding Tool-Mediated Interaction Failures in Agentic GUI Software: An Empirical Study
AI 摘要
该论文对40个开源agentic GUI软件项目进行实证研究，构建了包含335个真实缺陷的Malva缺陷库，归纳出11种根因于状态表示、状态转换与同步问题的缺陷模式。研究发现90%的应用存在多种缺陷类型，86.6%的失败表现为静默错误，其中119例无失败停止信号却产生错误行为。作者还提出部署前静态契约检查器MalvaGuard，在20个应用语料上达到57.3%精确率和76.8%召回率。
正文节选
Understanding Tool-Mediated Interaction Failures in Agentic GUI Software: An Empirical Study Abstract Large language model (LLM) agents increasingly complete tasks through live graphical user interfaces (GUIs). Because the model cannot directly operate the GUI, these systems often rely on tools to capture screenshots or UI trees, translate model outputs into GUI actions, and report whether the interface changed. We study failures where this tool-mediated interaction makes the agent’s view of the