One Tool Is Enough: Reinforcement Learning of LLM Agents for Repository-Level Code Navigation

Zhaoxi Zhang, Yitong Duan, Yanzhi Zhang, Yiming Xu, Zhixiang Wang, Kun Liang, Weikang Li, Jiahui Liang, Deguo Xia, Jizhou Huang, Jiyan He, Shuxin Zheng, Yunfang Wu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:155168-155194, 2026.

Abstract

Locating files and functions requiring modification in large software repositories is challenging due to their scale and structural complexity. Existing LLM-based methods typically treat this as a repository-level retrieval task and rely on multiple auxiliary tools, which often overlook code execution logic and complicate model control. We propose RepoNavigator, an LLM agent equipped with a single execution-aware tool: jumping to the definition of an invoked symbol. This unified design reflects the actual flow of code execution while simplifying tool manipulation. RepoNavigator is trained end-to-end via Reinforcement Learning (RL) directly from a base pretrained model, without relying on closed-source distillation. Experiments demonstrate that RL-trained RepoNavigator achieves state-of-the-art performance, with the 7B model outperforming 14B baselines, the 14B model surpassing 32B competitors, and the 32B model exceeding closed-source models such as GPT-5 on most metrics. These results confirm that integrating a single, structurally grounded tool with RL training provides an efficient and scalable solution for repository-level issue localization.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-zhang26an, title = {One Tool Is Enough: Reinforcement Learning of {LLM} Agents for Repository-Level Code Navigation}, author = {Zhang, Zhaoxi and Duan, Yitong and Zhang, Yanzhi and Xu, Yiming and Wang, Zhixiang and Liang, Kun and Li, Weikang and Liang, Jiahui and Xia, Deguo and Huang, Jizhou and He, Jiyan and Zheng, Shuxin and Wu, Yunfang}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {155168--155194}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/zhang26an/zhang26an.pdf}, url = {https://proceedings.mlr.press/v306/zhang26an.html}, abstract = {Locating files and functions requiring modification in large software repositories is challenging due to their scale and structural complexity. Existing LLM-based methods typically treat this as a repository-level retrieval task and rely on multiple auxiliary tools, which often overlook code execution logic and complicate model control. We propose RepoNavigator, an LLM agent equipped with a single execution-aware tool: jumping to the definition of an invoked symbol. This unified design reflects the actual flow of code execution while simplifying tool manipulation. RepoNavigator is trained end-to-end via Reinforcement Learning (RL) directly from a base pretrained model, without relying on closed-source distillation. Experiments demonstrate that RL-trained RepoNavigator achieves state-of-the-art performance, with the 7B model outperforming 14B baselines, the 14B model surpassing 32B competitors, and the 32B model exceeding closed-source models such as GPT-5 on most metrics. These results confirm that integrating a single, structurally grounded tool with RL training provides an efficient and scalable solution for repository-level issue localization.} }
Endnote
%0 Conference Paper %T One Tool Is Enough: Reinforcement Learning of LLM Agents for Repository-Level Code Navigation %A Zhaoxi Zhang %A Yitong Duan %A Yanzhi Zhang %A Yiming Xu %A Zhixiang Wang %A Kun Liang %A Weikang Li %A Jiahui Liang %A Deguo Xia %A Jizhou Huang %A Jiyan He %A Shuxin Zheng %A Yunfang Wu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-zhang26an %I PMLR %P 155168--155194 %U https://proceedings.mlr.press/v306/zhang26an.html %V 306 %X Locating files and functions requiring modification in large software repositories is challenging due to their scale and structural complexity. Existing LLM-based methods typically treat this as a repository-level retrieval task and rely on multiple auxiliary tools, which often overlook code execution logic and complicate model control. We propose RepoNavigator, an LLM agent equipped with a single execution-aware tool: jumping to the definition of an invoked symbol. This unified design reflects the actual flow of code execution while simplifying tool manipulation. RepoNavigator is trained end-to-end via Reinforcement Learning (RL) directly from a base pretrained model, without relying on closed-source distillation. Experiments demonstrate that RL-trained RepoNavigator achieves state-of-the-art performance, with the 7B model outperforming 14B baselines, the 14B model surpassing 32B competitors, and the 32B model exceeding closed-source models such as GPT-5 on most metrics. These results confirm that integrating a single, structurally grounded tool with RL training provides an efficient and scalable solution for repository-level issue localization.
APA
Zhang, Z., Duan, Y., Zhang, Y., Xu, Y., Wang, Z., Liang, K., Li, W., Liang, J., Xia, D., Huang, J., He, J., Zheng, S. & Wu, Y.. (2026). One Tool Is Enough: Reinforcement Learning of LLM Agents for Repository-Level Code Navigation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:155168-155194 Available from https://proceedings.mlr.press/v306/zhang26an.html.

Related Material