Dockerless: Environment-Free Program Verifier for Coding Agents

2026-06-26

AI Summary

This work addresses the high configuration overhead of traditional execution-based program repair verification, which relies on specific environments such as Docker. The paper introduces the first fully execution-free intelligent patch validator, which leverages an agent to actively explore code repositories and gather evidence to assess patch correctness—entirely through static analysis without executing unit tests. The approach supports both supervised fine-tuning via trajectory filtering and reinforcement learning with reward signals. Evaluated on a standard verifier benchmark, the method achieves an AUC score 14.3 points higher than the strongest open-source baseline. Furthermore, it attains solution rates of 62.0%, 50.0%, and 35.2% on the SWE-bench Verified, Multilingual, and Pro subsets, respectively, matching the performance of post-training methods that depend on execution environments.

Abstract

Program verifiers play a central role in training coding agents, including selecting trajectories for supervised fine-tuning (SFT) and providing rewards for reinforcement learning (RL). Standard execution-based verification requires running unit tests inside per-repository environments such as Docker images, incurring substantial environment setup costs. We propose Dockerless, an environment-free agentic patch verifier that evaluates generated code patches without executing them. Rather than simply matching candidate patches to references, Dockerless judges patch correctness using evidence gathered through agentic repository exploration. On a verifier evaluation benchmark, Dockerless outperforms the strongest open-source verifier by 14.3 AUC points. Using Dockerless as both the SFT trajectory filter and the RL reward enables a fully environment-free post-training pipeline. The resulting model reaches 62.0%, 50.0%, and 35.2% resolve rate on SWE-bench Verified, Multilingual, and Pro, respectively. It surpasses the Qwen3.5-9B baseline by 2.4, 8.7, and 2.9 points, matching environment-based post-training.

Problem

• program verification
• coding agents
• environment setup
• execution-based verification
• Docker

Innovation

• environment-free verification
• agentic code verification
• Dockerless
• program repair
• code agent training