LocAgent: Graph-Guided LLM Agents for Code Localization

By Zhaoling Chen, Xiangru Tang, Gangda Deng, Fang Wu, Jialong Wu, Zhiwei Jiang, Viktor K. Prasanna, Arman Cohan, Xing-Yao Wang. 2025

Subfields: LLM Agents & Tool Use, Graph Learning, Software Engineering

Citations in the corpus

• SweRank: Software Issue Localization with Code Ranking (2025, unknown): “More details in Appendix A. Evaluation Datasets: We utilize SWE-Bench-Lite (Jimenez et al., 2023) and LocBench (Chen et al., 2025) Baselines: Our primary comparison is against prior agent-based localization methods.”
• SweRank+: Multilingual, Multi-Turn Code Ranking for Software Issue Localization (2025, unknown): “…we show that S WE R ANK M ULTI establishes new state-of-the-art performance for issue localization across multiple programming languages (Rashid et al., 2025; Zan et al., 2025; Yang et al., 2025b), while maintaining competitive results on Python benchmarks (Jimenez et al., 2024; Chen et al., 2025).”
• A Benchmark for Localizing Code and Non-Code Issues in Software Projects (2025, unknown): “Overall, both SWE-Bench Lite and LocBench remain narrow in scope: they emphasize specific issue types, focus heavily on project-side defects, restrict locations to in-project files, and are largely code-centric in artifact coverage.”
• RepoNav: From Snippet Retrieval to File-Centered Repository Navigation for Code Agents (2026, unknown): “Repo-Graph (Ouyang et al., 2025) and CodexGraph (Liu et al., 2025) map entire codebases into graph databases; LocAgent (Chen et al., 2025) constructs directed code graphs for multi-hop traversal; Ling-maAgent (Ma et al., 2024) builds repository-level knowledge representations with Monte Carlo tree…”
• A^2Agent: Action-Aware Reinforcement Learning for Repository-Level Code Localization Agents (2026, unknown): “First, SFT generalizes poorly to unseen code-bases (7.11%p drop in file-level F1), while GRPO assigns the same advantage to every turn, leaving individual tool call contributions unattributed and treating partially successful trajectories identically to complete failures.”
• SHERLOC: Structured Diagnostic Localization for Code Repair Agents (2026, unknown): “…2025]; OrcaLoca uses priority-based action scheduling and distance-aware context pruning [Yu et al., 2025]; CoSIL and LocAgent guide LLM search through code-graph structure [Jiang et al., 2025, Chen et al., 2025]; RepoSearcher/ToolTrain uses tool-integrated reinforcement learning [Ma et al., 2025].”
• SWE-Adept: An LLM-Based Agentic Framework for Deep Codebase Analysis and Structured Issue Resolution (2026, unknown): “In addition, our framework mostly consumes fewer tokens than graph-based approaches (RepoGraph, OrcaLoca, LocAgent) due to effective context management (§ 3.2).”
• Learning Adaptive Parallel Execution for Efficient Code Localization (2026, unknown): “Agent-based methods enable flexible multi-step exploration: graph-guided approaches like LocAgent (Chen et al., 2025) and CoSIL (Jiang et al., 2025) represent code as static or dynamic graphs for navigation, while general agents like OpenHands (Wang et al., 2025) and SWE-agent (Yang et al., 2024) us”
• GraphLocator: Graph-Guided Causal Reasoning for Issue Localization (2025, unknown): “A critical yet costly initial step in this process is issue localization, which maps a natural language issue description to the specific code snippets that require modification [6, 45, 47, 49].”
• Agentic Software Issue Resolution with Large Language Models: A Survey (2025, unknown): “CoSIL [75] achieves concise yet effective performance improvements through a two-phase call graph approach that balances the search space and iteratively walks the call graph to retrieve relevant contexts.”
• SpIDER: Spatially Informed Dense Embedding Retrieval for Software Issue Localization (2025, unknown): “Some methods exploit graph representations of code repositories along-side the reasoning capabilities of LLMs (Ouyang et al., 2025; Chen et al., 2025; Jiang et al., 2025), mostly using sparse retrieval such as BM25 (Robert-son et al., 1994) to identify relevant graph nodes or subgraphs for agent…”
• Improving Code Localization with Repository Memory (2025, unknown): “We evaluate file-level localization performance via Accuracy@k (following prior work (Chen et al., 2025b)), defined as the percentage of examples where the set of top-k predicted files completely covers the ground-truth files.”