Agentic Software Issue Resolution with Large Language Models: A Survey
2025/12/24 by Zhonghao Jiang, David Lo, Jiang, Zhonghao +3
Computer Science · #Advanced Software Engineering Methodologies #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Software Engineering (cs.SE) #Software Engineering Research #Software System Performance and Reliability
paper · doi:10.48550/arxiv.2512.22256
openalex publication_date 2025/12/24 · openalex created_date 2025/12/31 · openalex updated_date 2026/07/28
Abstract
Software issue resolution aims to address real-world issues in software repositories (e.g., bug fixing and efficiency optimization) based on natural language descriptions provided by users, representing a key aspect of software maintenance. With the rapid development of large language models (LLMs) in reasoning and generative capabilities, LLM-based approaches have made significant progress in automated software issue resolution. However, real-world software issue resolution 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 become mainstream for software issue resolution. Advancements in agentic software issue resolution not only greatly enhance software maintenance efficiency and quality but also provide a realistic environment for validating agentic systems' reasoning, planning, and execution capabilities, bridging artificial intelligence and software engineering. This work presents a systematic survey of 126 recent studies at the forefront of LLM-based agentic software issue resolution research. It outlines the general workflow of the task and establishes a taxonomy across three dimensions: benchmarks, techniques, and empirical studies. Furthermore, it highlights how the emergence of agentic reinforcement learning has brought a paradigm shift in the design and training of agentic systems for software engineering. Finally, it summarizes key challenges and outlines promising directions for future research.
Citations
- When Old Meets New: Evaluating the Impact of Regression Tests on SWE Issue Resolution
- A Comprehensive Survey on Benchmarks and Solutions in Software Engineering of LLM-Empowered Agentic System
- Saving SWE-Bench: A Benchmark Mutation Approach for Realistic Agent Evaluation
- Retrieval-Augmented Code Generation: A Survey with Focus on Repository-Level Approaches
- Improving Code Localization with Repository Memory
- CWM: An Open-Weights LLM for Research on Code Generation with World Models
- Kimi-Dev: Agentless Training as Skill Prior for SWE-Agents
- A Benchmark for Localizing Code and Non-Code Issues in Software Projects
- SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks?
- An Empirical Study on Failures in Automated Issue Solving
- The Landscape of Agentic Reinforcement Learning for LLMs: A Survey
- When Agents go Astray: Course-Correcting SWE Agents with PRMs
- Automated Generation of Issue-Reproducing Tests by Combining LLMs and Search-Based Testing
- Devstral: Fine-tuning Language Models for Coding Agent Applications
- GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models
- Training Long-Context, Multi-Turn Software Engineering Agents with Reinforcement Learning
- SE-Agent: Self-Evolution Trajectory Optimization in Multi-Step Reasoning with LLM-Based Agents
- RepoForge: Training a SOTA Fast-thinking SWE Agent with an End-to-End Data Curation Pipeline Synergizing SFT and RL at Scale
- Trae Agent: An LLM-based Agent for Software Engineering with Test-time Scaling
- SWE-Exp: Experience-Driven Software Issue Resolution
- SWE-Debate: Competitive Multi-Agent Debate for Software Issue Resolution
- Prometheus: Unified Knowledge Graphs for Issue Resolution in Multilingual Codebases
- NoCode-bench: A Benchmark for Evaluating Natural Language-Driven Feature Addition
- AssertFlip: Reproducing Bugs via Inversion of LLM-Generated Passing Tests
- SWE-Perf: Can Language Models Optimize Code Performance on Real-World Repositories?
- SWE-MERA: A Dynamic Benchmark for Agenticly Evaluating Large Language Models on Software Engineering Tasks
- SPICE: An Automated SWE-Bench Labeling Pipeline for Issue Clarity, Test Coverage, and Effort Estimation
- Agent KB: Leveraging Cross-Domain Experience for Agentic Problem Solving
- SACL: Understanding and Combating Textual Bias in Code Retrieval with Semantic-Augmented Reranking and Localization
- Skywork-SWE: Unveiling Data Scaling Laws for Software Engineering in LLMs
- Understanding Software Engineering Agents: A Study of Thought-Action-Result Trajectories
- PAGENT: Learning to Patch Software Engineering Agents
- SemAgent: A Semantics Aware Program Repair Agent
- MCTS-Refined CoT: High-Quality Fine-Tuning Data for LLM-Based Repository Issue Resolution
- The SWE-Bench Illusion: When State-of-the-Art LLMs Remember Instead of Reason
- SWE-Bench-CL: Continual Learning for Coding Agents
- EXPEREPAIR: Dual-Memory Enhanced LLM-based Repository-Level Program Repair
- Agent-RLVR: Training Software Engineering Agents via Guidance and Environment Rewards
- UTBoost: Rigorous Evaluation of Coding Agents on SWE-Bench
- SWE-Dev: Building Software Engineering Agents with Training and Inference Scaling
- Coding Agents with Multimodal Browsing are Generalist Problem Solvers
- CoRet: Improved Retriever for Code Editing
- GSO: Challenging Software Optimization Tasks for Evaluating SWE-Agents
- Satori-SWE: Evolutionary Test-Time Scaling for Sample-Efficient Software Engineering
- SWE-rebench: An Automated Pipeline for Task Collection and Decontaminated Evaluation of Software Engineering Agents
- Co-PatcheR: Collaborative Software Patching with Component(s)-specific Small Reasoning Models
- Code Graph Model (CGM): A Graph-Integrated Large Language Model for Repository-Level Software Engineering Tasks
- InfantAgent-Next: A Multimodal Generalist Agent for Automated Computer Interaction
- Qwen3 Technical Report
- TRAIL: Trace Reasoning and Agentic Issue Localization
- OmniGIRL: A Multilingual and Multimodal Benchmark for GitHub Issue Resolution
- SWE-smith: Scaling Data for Software Engineering Agents
- SWE-Synth: Synthesizing Verifiable Bug-Fix Data to Enable Large Language Models in Resolving Real-World Bugs
- SWE-PolyBench: A multi-language benchmark for repository level evaluation of coding agents
- R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agents
- Thinking Longer, Not Larger: Enhancing Software Engineering Agents via Scaling Test-Time Compute
- Enhancing repository-level software repair via repository-aware knowledge graphs
- SEAlign: Alignment Training for Software Engineering Agent
- Are "Solved Issues" in SWE-bench Really Solved Correctly? An Empirical Study
- DARS: Dynamic Action Re-Sampling to Enhance Coding Agent Performance by Adaptive Tree Traversal
- LocAgent: Graph-Guided LLM Agents for Code Localization
- Automated Benchmark Generation for Repository-Level Coding Tasks
- FEA-Bench: A Benchmark for Evaluating Repository-Level Code Generation for Feature Implementation
- SoRFT: Issue Resolving with Subtask-oriented Reinforced Fine-Tuning
- SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution
- Ambig-SWE: Interactive Agents to Overcome Underspecificity in Software Engineering
- Otter: Generating Tests from Issues to Validate SWE Patches
- Agentic Bug Reproduction for Effective Automated Program Repair at Google
- OrcaLoca: An LLM Agent Framework for Software Issue Localization
- Qwen2.5-1M Technical Report
- Learn-by-interact: A Data-Centric Framework for Self-Adaptive Agents in Realistic Environments
- SWE-Fixer: Training Open-Source LLMs for Effective and Efficient GitHub Issue Resolution
- Training Software Engineering Agents and Verifiers with SWE-Gym
- Repository Structure-Aware Training Makes SLMs Better Issue Resolver
- CodeV: Issue Resolving with Visual Data
- TDD-Bench Verified: Can LLMs Generate Tests for Issues Before They Get Resolved?
- CoRNStack: High-Quality Contrastive Data for Better Code Ranking
- AEGIS: An Agent-based Framework for General Bug Reproduction from Issue Descriptions
- A Survey on LLM-as-a-Judge
- An Empirical Study on LLM-based Agents for Automated Bug Fixing
- Infant Agent: A Tool-Integrated, Logic-Driven Agent with Cost-Effective API Usage
- Lingma SWE-GPT: An Open Development-Process-Centric Language Model for Automated Software Improvement
- SWE-Search: Enhancing Software Agents with Monte Carlo Tree Search and Iterative Refinement
- Evaluating Software Development Agents: Patch Patterns, Code Quality, and Issue Complexity in Real-World GitHub Scenarios
- SWE-Bench+: Enhanced Coding Benchmark for LLMs
- SWE-bench Multimodal: Do AI Systems Generalize to Visual Software Domains?
- RepoGraph: Enhancing AI Software Engineering with Repository-level Code Graph
- TestGenEval: A Real World Unit Test Generation and Test Completion Benchmark
- Qwen2.5-Coder Technical Report
- SuperCoder2.0: Technical Report on Exploring the feasibility of LLMs as Autonomous Programmer
- HyperAgent: Generalist Software Engineering Agents to Solve Coding Tasks at Scale
- Large Language Model-Based Agents for Software Engineering: A Survey
- MarsCode Agent: AI-native Automated Bug Fixing
- SWE-bench-java: A GitHub Issue Resolving Benchmark for Java
- Diversity Empowers Intelligence: Integrating Expertise of Software Engineering Agents
- CodexGraph: Bridging Large Language Models and Code Repositories via Code Graph Databases
- SpecRover: Code Intent Extraction via LLMs
- The Llama 3 Herd of Models
- BLAZE: Cross-Language and Cross-Project Bug Localization via Dynamic Chunking and Hard Example Learning
- OpenHands: An Open Platform for AI Software Developers as Generalist Agents
- SWT-Bench: Testing and Validating Real-World Bug-Fixes with Code Agents
- MASAI: Modular Architecture for Software-engineering AI Agents
- Alibaba LingmaAgent: Improving Automated Issue Resolution via Comprehensive Repository Exploration
- CodeR: Issue Resolving with Multi-Agent and Task Graphs
- A Survey on Large Language Models for Code Generation
- SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering
- From Local to Global: A Graph RAG Approach to Query-Focused Summarization
- MAGIS: LLM-Based Multi-Agent Framework for GitHub Issue Resolution
- RepairAgent: An Autonomous, LLM-Based Agent for Program Repair
- Bug Priority Change: An Empirical Study on Apache Projects
- Bug priority change: An empirical study on Apache projects
- CrashTranslator: Automatically Reproducing Mobile Application Crashes Directly from Stack Trace
- Persistent memory for AI coding agents: a pre-registered SWE-bench Verified benchmark
- Qwen Technical Report
- RepoFusion: Training Code Models to Understand Your Repository
- GPT-4 Technical Report
- Large Language Models are Few-shot Testers: Exploring LLM-based General Bug Reproduction
- Self-Consistency Improves Chain of Thought Reasoning in Language Models
- AutoFL: Enabling Heterogeneity-Aware Energy Efficient Federated Learning
- Attention Is All You Need
- An orchestrated survey of methodologies for automated software test case generation
- PatchPilot: A Cost-Efficient Software Engineering Agent with Early Attempts on Formal Verification
- A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications
- Issue Localization via LLM-Driven Iterative Code Graph Searching
Related