Training Versatile Coding Agents in Synthetic Environments

2025-12-13

AI Summary

Current software engineering agent training approaches are constrained by reliance on GitHub repositories and limited task diversity. This paper introduces SWE-Playground, an end-to-end synthetic environment generation pipeline that constructs diverse programming projects and multi-stage task trajectories—such as unit test generation and library implementation—from scratch, eliminating external codebase dependencies. Leveraging strong language models and agent collaboration mechanisms, it enables automated synthetic data generation, trajectory construction, and extraction of dense supervision signals. To our knowledge, this is the first framework to realize a fully synthetic, broad-spectrum, and high-density training paradigm for software engineering tasks. Experiments demonstrate that SWE-Playground achieves state-of-the-art performance on three benchmarks using significantly fewer training trajectories, empirically validating both the high quality and efficiency of its synthetic data.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Prior works on training software engineering agents have explored utilizing existing resources such as issues on GitHub repositories to construct software engineering tasks and corresponding test suites. These approaches face two key limitations: (1) their reliance on pre-existing GitHub repositories offers limited flexibility, and (2) their primary focus on issue resolution tasks restricts their applicability to the much wider variety of tasks a software engineer must handle. To overcome these challenges, we introduce SWE-Playground, a novel pipeline for generating environments and trajectories which supports the training of versatile coding agents. Unlike prior efforts, SWE-Playground synthetically generates projects and tasks from scratch with strong language models and agents, eliminating reliance on external data sources. This allows us to tackle a much wider variety of coding tasks, such as reproducing issues by generating unit tests and implementing libraries from scratch. We demonstrate the effectiveness of this approach on three distinct benchmarks, and results indicate that SWE-Playground produces trajectories with dense training signal, enabling agents to reach comparable performance with significantly fewer trajectories than previous works.

Problem

• Overcomes reliance on existing GitHub repositories for training software agents.
• Expands training scope beyond issue resolution to diverse coding tasks.
• Synthetically generates projects and tasks to enhance agent versatility.

Innovation

• Synthetic project generation with language models
• Wide variety of coding tasks without external data
• Dense training signals from generated trajectories