Agentic Reinforcement Learning for Real-World Code Repair

TL;DR

This paper presents a scalable reinforcement learning pipeline for real-world code repair, demonstrating improved performance and reproducibility, but highlighting challenges in environment generalization.

Contribution

Introduces a verifiable, scalable RL pipeline for code fixing and shows that RL improves performance over supervised fine-tuning in real-world settings.

Findings

RL improves code repair accuracy by 7-20%.

Reproducibility is enhanced by dependency pinning.

Models struggle to generalize across different environments.

Abstract

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.