信息网络安全

信息网络安全 ›› 2026, Vol. 26 ›› Issue (9): 1369-1382.doi: 10.3969/j.issn.1671-1122.2026.09.004

• 入选论文 • 上一篇    下一篇

仓库级静态漏洞检测综述：从规则驱动到大语言模型与智能体

宋铮1(), 杨芮淇1,2, 李化成1   

  1. 1 公安部第三研究所， 上海 201204
    2 武汉大学国家网络安全学院， 武汉 430072
  • 收稿日期:2026-05-10 出版日期:2026-09-10 发布日期:2026-09-30
  • 通讯作者: 宋铮 E-mail:songzheng@gass.cn
  • 作者简介:宋铮（1986—），男，江苏，副研究员，硕士，CCF会员，主要研究方向为软件安全、网络安全|杨芮淇（2002—），女，湖南，硕士研究生，主要研究方向为二进制分析、系统安全|李化成（1995—），男，山东，助理研究员，博士，主要研究方向为内容安全、网络空间安全、人工智能与网络安全
  • 基金资助:
    公安部第三研究所人才培育项目(CWS25105)

Survey of repository-level static vulnerability detection: from rule-driven to LLM and agent

Song Zheng1(), Yang Ruiqi1,2, Li Huacheng1   

  1. 1 The Third Research Institute of Ministry of Public Security, PRC, Shanghai 201204, China
    2 School of Cyber Science and Engineering, Wuhan University, Wuhan 430072, China
  • Received:2026-05-10 Online:2026-09-10 Published:2026-09-30
  • Contact: Song Zheng E-mail:songzheng@gass.cn

RichHTML

2

PDF (PC)

13

摘要：

现代软件系统规模与复杂度激增，使传统静态分析面临严峻挑战。为此，文章系统地综述仓库级静态漏洞检测技术的演进脉络、核心挑战及前沿方法。首先，梳理从规则驱动的传统静态分析、数据驱动的机器学习与深度学习，到语义驱动的大语言模型（LLM）及智能体的技术代际演进，剖析中间表示在跨文件建模中的作用。然后，针对仓库级场景，归纳上下文规模超限、跨文件依赖断裂和误报控制困难等6类核心挑战，系统评估静态分析工具与LLM的深度集成、智能体协同检测等技术路线的能力边界。最后，在此基础上，深入剖析概率推理与形式化约束冲突而引发的可信性瓶颈，指出现有数据集在结构上的若干局限，提出面向工程效能的评估指标重构思路，并展望未来研究方向。

关键词: 仓库级静态漏洞检测, 大语言模型, 智能体, 静态分析工具集成

Abstract:

The surging scale and complexity of modern software systems posed severe challenges to traditional static analysis. To address this, this paper systematically reviewed the evolutionary trajectory, core challenges, and state-of-the-art methods in repository-level static vulnerability detection. First, it traced the generational technological evolution from rule-driven traditional static analysis and data-driven machine learning/deep learning to semantics-driven large language model(LLM) and agent, and analyzed the critical role of intermediate representations in cross-file modeling. Then, for repository-level scenarios, it summarized six core challenges, including excessive context scale, fragmented cross-file dependencies, and difficulty in controlling false positives. It systematically evaluated the capability boundaries of technical approaches such as the deep integration of static analysis tools with LLM and collaborative agent-based detection frameworks. Finally, on this basis, it deeply dissected the trustworthiness bottleneck caused by the conflict between probabilistic reasoning and formal constraints, identified several structural limitations in existing datasets, proposed an approach to reconstructing evaluation metrics for engineering effectiveness, and outlined future research directions.

Key words: repository-level static vulnerability detection, large language model, agent, static analysis tool integration

中图分类号: