Reproduce “Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-Refine” 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
Junyao Yang, Zongxia Li, Yucheng Shi, Zhongzhi Li, Ruhan Wang, Ruiwen Zhou, Haitao Mi, Shuicheng Yan and 1 more
Terminal-RSI trains terminal agents to reproduce verified task solutions by alternating Recursive Self-Rewrite, which converts verified trajectories into demonstrations, with Recursive Policy-Refine, which updates the policy while retaining and recalibrating its critic. Across four benchmark suites, its four-suite mean success rises from 20.68% to 50.58%. On LHTB, process reward rises from 0.21 to 0.37 and strict completions from 0/46 to 5/46.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.