# A^2Agent: Action-Aware Reinforcement Learning for Repository-Level Code Localization Agents

arXiv: https://arxiv.org/abs/2608.29831 · HTML: https://mlsanity.com/paper/2608.29831/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 23 of its 32 references that are in the map.
- First posted on arXiv 30 Aug 2026.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)

## Builds on (most cited first, 23 of 23)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [ZeRO: Memory Optimizations Toward Training Trillion Parameter Models](https://mlsanity.com/paper/1910.02054/) (2019)
- [Advantage-Weighted Regression: Simple and Scalable Off-Policy Reinforcement Learning](https://mlsanity.com/paper/1910.00177/) (2019)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [AWAC: Accelerating Online Reinforcement Learning with Offline Datasets](https://mlsanity.com/paper/2006.09359/) (2020)
- [Reinforced Self-Training (ReST) for Language Modeling](https://mlsanity.com/paper/2308.08998/) (2023)
- [Group-in-Group Policy Optimization for LLM Agent Training](https://mlsanity.com/paper/2505.10978/) (2025)
- [Training Software Engineering Agents and Verifiers with SWE-Gym](https://mlsanity.com/paper/2412.21139/) (2024)
- [SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution](https://mlsanity.com/paper/2502.18449/) (2025)
- [RepoGraph: Enhancing AI Software Engineering with Repository-level Code Graph](https://mlsanity.com/paper/2410.14684/) (2024)
- [LocAgent: Graph-Guided LLM Agents for Code Localization](https://mlsanity.com/paper/2503.09089/) (2025)
- [Hierarchy-of-Groups Policy Optimization for Long-Horizon Agentic Tasks](https://mlsanity.com/paper/2602.22817/) (2026)
- [OrcaLoca: An LLM Agent Framework for Software Issue Localization](https://mlsanity.com/paper/2502.00350/) (2025)
- [CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking](https://mlsanity.com/paper/2412.01007/) (2024)
- [SweRank: Software Issue Localization with Code Ranking](https://mlsanity.com/paper/2505.07849/) (2025)
- [Improving Code Localization with Repository Memory](https://mlsanity.com/paper/2510.01003/) (2025)
- [Tool-integrated Reinforcement Learning for Repo Deep Search](https://mlsanity.com/paper/2508.03012/) (2025)
- [SACL: Understanding and Combating Textual Bias in Code Retrieval with Semantic-Augmented Reranking and Localization](https://mlsanity.com/paper/2506.20081/) (2025)
- [Learning to Refuse: Refusal-Aware Reinforcement Fine-Tuning for Hard-Irrelevant Queries in Video Temporal Grounding](https://mlsanity.com/paper/2511.23151/) (2025)
- [EXPO-SQL: Execution-based Clause-level Policy Optimization for Text-to-SQL](https://mlsanity.com/paper/2606.23693/) (2026)

## Abstract (arXiv)

Localizing issue-relevant code regions is a critical step in automated software engineering. However, due to their reliance on sparse trajectory-level signals, existing methods cannot identify which per-turn actions are effective and often discover correct code regions during exploration but fail to commit them. To address these limitations, we propose an action-aware reinforcement learning method that combines a per-turn reward sequence rewarding both the discovery and commitment of gold code regions with an action-level advantage estimation scheme that isolates each action's credit by grouping turns sharing the same exploration context. Extensive evaluations show that our method improves the average F1 over the state-of-the-art (SOTA) by 1.58% on SWE-Bench Verified and 8.55% on SWE-Bench Pro, with our 4B model outperforming baselines up to 8x larger. Our code is available at https://github.com/donian00/A2Agent.