Agentic Reinforcement Learning for Real-World Code Repair
This work addresses the problem of building reliable code-fixing agents for real-world software development, but it is incremental as it builds on existing methods and highlights limitations in generalization.
The paper tackled the challenge of training reliable code-fixing agents in real repositories by developing a verifiable pipeline that improved reproducibility across about 1,000 real issues, and introduced a scalable simplified pipeline for reinforcement learning, where the supervised fine-tuned model performed on par with GPT-4.1 while being 56x smaller and reinforcement learning added 7-20% absolute gains.
We tackle the challenge of training reliable code-fixing agents in real repositories, where complex builds and shifting dependencies make evaluation unstable. We developed a verifiable pipeline with success defined as post-fix build validation and improved reproducibility across ~1K real issues by pinning dependencies and disabling automatic upgrades. Building on this, we introduced a scalable simplified pipeline for large-scale reinforcement learning (RL). Using this setup, we supervised fine-tuned Qwen3-32B in the full pipeline and applied RL on top of the SFT model in the simplified environment. The SFT model distilled from GPT-4.1 trajectories performs on par while being 56x smaller, and RL added 7-20% absolute gains under matched train-test conditions. "Thinking mode" was on par or worse in our experiments. Both SFT and RL models failed to generalize across environments, highlighting the importance of matching train-test environments for building reliable real-world code-fixing agents.