跳转至

基于大语言模型的智能体化软件缺陷修复:综述

文章背景与核心概要

本文探讨了人工智能与软件工程的前沿交叉领域,聚焦于用于自动化软件缺陷修复(Software Issue Resolution)的基于大语言模型(LLM)的智能体系统。随着大模型在推理和生成能力上的飞跃,如何利用AI处理现实世界中的软件维护任务已成为行业焦点。

通过对近期 242 篇相关研究的系统性梳理,该综述深入分析了智能体框架如何通过长程推理(long-horizon reasoning)、迭代式探索以及基于反馈的决策机制,突破传统单步LLM交互的局限。文章建立了一个包含基准测试(Benchmarks)、核心技术(Techniques)和实证研究(Empirical Studies)三个维度的分类体系,凸显了强化学习在软件工程智能体训练中的重要地位,并指出了该领域的未来挑战与发展方向。


执行摘要 / Executive Summary

本综述探讨了人工智能与软件工程的前沿交叉领域,重点关注旨在实现自动化软件缺陷修复的基于LLM的智能体系统。通过分析 242 项最新研究,本文考察了智能体框架如何超越传统的单步LLM交互,通过整合长程推理、迭代式探索和基于反馈的决策机制,来应对复杂的现实世界软件维护任务。

This survey explores the cutting-edge intersection of artificial intelligence and software engineering, focusing on LLM-based agentic systems designed for automated software issue resolution. Analyzing 242 recent studies, the paper examines how agentic frameworks move beyond traditional single-step LLM interactions by incorporating long-horizon reasoning, iterative exploration, and feedback-driven decision-making to tackle complex real-world software maintenance tasks.


文章详情 / Article Details

  • arXiv 标识符: arXiv:2512.22256 [cs.SE]
  • 学科分类: 软件工程 (cs.SE);人工智能 (cs.AI)
  • 作者:
  • Zhonghao Jiang
  • David Lo
  • Zhongxin Liu
  • 时间线:
  • 提交时间:2025年12月24日 (v1)
  • 最后修订:2026年8月6日 (v2)
  • DOI: 10.48550/arXiv.2512.22256
  • arXiv Identifier: arXiv:2512.22256 [cs.SE]
  • Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI)
  • Authors:
  • Zhonghao Jiang
  • David Lo
  • Zhongxin Liu
  • Timeline:
  • Submitted: 24 December 2025 (v1)
  • Last Revised: 6 August 2026 (v2)
  • DOI: 10.48550/arXiv.2512.22256

摘要 / Abstract

软件缺陷修复旨在根据用户提供的自然语言描述,解决软件代码库中的现实问题,它是软件维护的关键环节。随着大语言模型(LLM)在推理和生成能力方面的迅猛发展,基于LLM的方法在自动化软件缺陷修复方面取得了显著进展。

然而,解决现实世界的软件缺陷本质上十分复杂,需要长程推理、迭代式探索以及基于反馈的决策机制,这些都对超越传统单步方法的智能体能力提出了要求。近年来,基于LLM的智能体系统已成为软件缺陷修复领域一个充满前景的研究方向,相关文献也随之呈现爆发式增长。

智能体化软件缺陷修复的进展不仅能极大地提升软件维护的效率和质量,还能为评估智能体系统的推理、规划和执行能力提供一个逼真的环境,从而将人工智能与软件工程紧密连结起来。

本工作对处于LLM驱动的智能体化软件缺陷修复研究最前沿的 242 项最新研究 进行了系统性综述。文章概述了该任务的一般工作流程,并从以下三个维度建立了一个分类体系: 1. 基准测试(Benchmarks) 2. 核心技术(Techniques) 3. 实证研究(Empirical Studies)

此外,文章强调了强化学习作为软件工程中智能体系统日益重要的训练范式,最后总结了关键挑战以及对未来研究充满希望的方向。

Software issue resolution aims to address real-world issues in software repositories based on natural language descriptions provided by users, and represents a key aspect of software maintenance. With the rapid development of large language models (LLMs) in reasoning and generation, LLM-based approaches have made significant progress in automated software issue resolution.

However, resolving real-world software issues is inherently complex and requires long-horizon reasoning, iterative exploration, and feedback-driven decision-making, which demand agentic capabilities beyond conventional single-step approaches. Recently, LLM-based agentic systems have emerged as a promising research direction for software issue resolution, accompanied by rapid growth in the relevant literature.

Advances in agentic software issue resolution can not only greatly improve the efficiency and quality of software maintenance, but also provide a realistic environment for evaluating the reasoning, planning, and execution capabilities of agentic systems, thereby bridging artificial intelligence and software engineering.

This work presents a systematic survey of 242 recent studies at the forefront of research on LLM-based agentic software issue resolution. It outlines the general workflow of the task and establishes a taxonomy across three dimensions: 1. Benchmarks 2. Techniques 3. Empirical Studies

Furthermore, it highlights reinforcement learning as an increasingly important training paradigm for agentic systems in software engineering, concluding with a summary of key challenges and promising directions for future research.


访问与资源 / Access & Resources