一句话：提出基于 LLM 多智能体协作的即时兼容性缺陷检测方法 MAJIT，在代码变更上检测系统版本、设备与屏幕尺寸问题，精确率与召回率均超 96%。
移动应用常因系统版本、设备与屏幕尺寸差异产生兼容性问题，在代码变更时及时发现这类缺陷对开发者至关重要。本文提出 MAJIT，一种面向 Android 与 iOS 兼容性问题的多智能体即时检测方法。给定代码变更，它先用 diff 增强分析构造聚焦且可靠的上下文，再通过可扩展的基于大语言模型的多智能体缺陷检测流程定位兼容性问题，最终返回被预测存在问题的变更代码块，并附带判断依据与可能原因。作者在自行收集的真实世界兼容性问题数据集上开展实验，针对系统版本问题、设备问题和屏幕尺寸问题三类缺陷，精确率与召回率均超过 96%。
In this paper, we introduce a Multi-Agent Just-In-Time detection approach: MAJIT, which is tailored for Android and iOS compatibility issues. Given a code change as input, our approach adopts diff-augmented analysis to provide targeted and reliable context, and further employ a scalable LLM-based multi-agent bug detection approach for compatibility issues. Our approach finally return changed code blocks predicted to have compatibility issues, along with their corresponding evidence and possible causes. We conducted our experiments on a self-collected real-world compatibility issue dataset. Experimental results show that when detecting three types of compatibility issues: system version issue, device issue, and screen size issue, our approach demonstrates over 96% on both precision and recall score.
@inproceedings{WangCLMQGL26,
title = {MAJIT: Just-in-Time Detection of Compatibility Issues in Android and iOS Apps through Large Language Model-based Multi-Agent Collaboration},
author = {Jiaqi Wang and Di Cui and Shenghan Liu and Qiankang Mao and Xiangxing Qian and Qiaoyin Gan and Rui Li},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3786582.3786819}
}