Reproduce “When Requirements Interact: Scaling Long-Horizon Coding via Interdependent Specifications” 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
Yishuo Cai, Minzheng Wang, Liang Zhao, Linghao Zhang, Lei Li, Shicheng Li, Xu Sun, Fuli Luo
CodeSpecWeave creates repository coding tasks in which requirements constrain one another, and builds executable tests alongside each task. Its dataset averages 8.36 specifications across 4.08 behavioral categories per task, compared with 2.15 and 1.49 for R2E-Gym. On DeepSWE-v1.1, MiMo-V2.6-Flash-SFT reports 62.98% success versus 56.64% with SWE-PI, though this comparison does not isolate coupling as the cause.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.