一句话：对 CrewAI 与 LangGraph 的 998 条 issue 进行实证研究，构建 LLM agent 框架缺陷的根因—症状分类体系。
LLM agent 已在真实应用中广泛部署，其生命周期执行与多 agent 协作依赖 agent 框架。随着系统规模扩大，理解底层框架中的缺陷变得至关重要，但既有工作多关注 agent 层面的失败，忽视了框架层面的缺陷。本文对 CrewAI 与 LangGraph 的 998 条 issue 报告开展实证研究，在 Agent Initialization、Perception、Self-Action、Mutual Interaction 与 Evolution 五个生命周期阶段上，构建了包含 15 类根因与 7 类可观察症状的分类体系。研究发现，框架缺陷主要源于 API 误用、API 不兼容与文档不同步，并高度集中于 Self-Action 阶段；症状多表现为 Functional Error、Crash 与 Build Failure，反映出任务推进与控制流受到的破坏。
LLM agents have been widely adopted in real-world applications, relying on agent frameworks for lifecycle execution and multiagent coordination. As these systems scale, understanding bugs in the underlying agent frameworks becomes critical. However, existing work mainly focuses on agent-level failures, overlooking framework-level bugs. To address this gap, we conduct an empirical study of 998 issue reports from CrewAI and LangGraph, and construct a taxonomy of 15 root causes and seven observable symptoms across five agent lifecycle stages: 'Agent Initialization', 'Perception', 'Self-Action', 'Mutual Interaction' and 'Evolution'. Our findings show that agent framework bugs mainly arise from 'API Misuse', 'API Incompatibility', and 'Documentation Desync', largely concentrated in the 'Self-Action' stage. Symptoms typically appear as 'Functional Error', 'Crash', and 'Build Failure', reflecting disruptions to task progression and control flow.
@inproceedings{ZhuWZLMZSFL26,
title = {Bugs in Modern LLM Agent Frameworks: An Empirical Study},
author = {Xinxue Zhu and Jiacong Wu and Xiaoyu Zhang and Tianlin Li and Yanzhou Mu and Juan Zhai and Chao Shen and Chunrong Fang and Yang Liu},
booktitle = {FSE},
year = {2026},
doi = {10.1145/3803437.3805536}
}