SWE-Exp: Experience-Driven Software Issue Resolution

2025-07-31

AI Summary

Existing LLM-based software repair agents lack memory mechanisms, preventing reuse of historical experience and leading to redundant trial-and-error and knowledge waste. Method: We propose the first LLM repair framework supporting continual learning: it constructs a multi-dimensional structured experience repository that explicitly stores and distills knowledge—including problem understanding, reasoning paths, and code modifications—from successful and failed repair trajectories; integrates multi-agent collaboration, Monte Carlo tree search, and experience distillation to enable cross-task dynamic retrieval and knowledge transfer. Contribution/Results: Our approach achieves 41.6% Pass@1 on SWE-bench-Verified, significantly outperforming existing open-source agents. It is the first to realize systematic, experience-informed decision-making in software engineering tasks, establishing an experience-driven paradigm for automated software maintenance.

Technology Category

• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Multiagent Systems: Multiagent Learning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Recent advances in large language model (LLM) agents have shown remarkable progress in software issue resolution, leveraging advanced techniques such as multi-agent collaboration and Monte Carlo Tree Search (MCTS). However, current agents act as memoryless explorers - treating each problem separately without retaining or reusing knowledge from previous repair experiences. This leads to redundant exploration of failed trajectories and missed chances to adapt successful issue resolution methods to similar problems. To address this problem, we introduce SWE-Exp, an experience - enhanced approach that distills concise and actionable experience from prior agent trajectories, enabling continuous learning across issues. Our method introduces a multi-faceted experience bank that captures both successful and failed repair attempts. Specifically, it extracts reusable issue resolution knowledge at different levels - from high-level problem comprehension to specific code changes. Experiments show that SWE-Exp achieves state-of-the-art resolution rate (41.6% Pass@1) on SWE-bench-Verified under open-source agent frameworks. Our approach establishes a new paradigm in which automated software engineering agents systematically accumulate and leverage repair expertise, fundamentally shifting from trial-and-error exploration to strategic, experience-driven issue resolution.

Problem

• LLM agents lack memory for reusing past repair experiences
• Redundant exploration of failed trajectories in issue resolution
• Need for continuous learning across similar software issues

Innovation

• Experience-enhanced approach with actionable knowledge
• Multi-faceted experience bank for repair attempts
• Continuous learning across software issue resolutions