Reproduce “From Repositories to Training Environments: Scaling Long-Horizon Coding Agents with Environment Evolution Loop” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yang Tian, Tianyu Zheng, Xiyuan Yang, Yaxin Du, Chengran Yang, Bo Zhao, Jian Yang, Bryan Dai
RepoFactory creates testable tasks from Python repositories, where an agent rebuilds a complete project from a natural-language specification. It revises reference implementations, specifications, and tests using validity and code-structure signals, without solver rollouts for feedback. Training raises Qwen3.5-35B-A3B’s NL2Repo score from 23.8% to 46.4%. It focuses on Python; its structural proxies do not identify individual task capabilities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.