Execution guided ai improves fixing and locating github issues
XAgent: eXecution-guided Agentic AI for Effective Localization and Resolution of GitHub Issues
Software Engineering
Summary
Fixing bugs on GitHub often depends only on the problem descriptions, which can miss important details and lead to incomplete fixes. The authors created XAgent, an AI system that looks at how code actually runs and gathers more context to better find and fix issues. Their tests show XAgent fixes more problems and spots the right parts of code more accurately than other methods. This approach moves beyond just reading problem descriptions to using real code behavior, making fixes more reliable.
What this means in practice
- •For software maintenance teams: Improve automated fixing of GitHub repository bugs by using dynamic code execution to guide issue localization and patch validation.
- •For code review engineers: Enhance validation of code patches by integrating execution feedback for more accurate issue resolution during reviews.
Authors
Hieu Huynh, Patanamon Thongtanunam, Michael Fu, Bach Le, Kla Tantithamthavorn
Abstract
Agentic AI has enabled capabilities in leveraging Large Language Models (LLMs) to autonomously resolve repository-level GitHub issues. However, due to the reliance on limited static description of issues, existing agentic approaches suffer from incorrect localization and incomplete validation. Solely relying on this information can bias LLM reasoning toward the narrow scope of the issue description, leading to incomplete patches that fail to address the underlying issue. In this paper, we present XAgent, an execution-guided agentic framework that analyzes dynamic behavior and additional program context to localize and validate issues. The experimental results on the SWE-bench-lite dataset demonstrate that XAgent outperforms other existing approaches, achieving a resolve rate of 62.0% and a function localization accuracy of 72.8%, while maintaining cost efficiency. Our analysis further shows that XAgent successfully resolves 7 additional issues that the top existing baselines fail to address. This work highlights a shift from static, description-oriented patch generation toward dynamic execution-guided issue resolution, opening new opportunities for LLM-based coding agents to achieve more robust and generalizable software maintenance.