Reproduce “ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement” 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
Siwei Wu, Jincheng Ren, Yizhi Li, Haau-Sing Li, Chengran Yang, Xudong Ma, Yuxuan Zhang, Weicheng Gu and 7 more
ModularRSI evolves an AI agent’s harness by comparing successful and failed runs of the same task, aggregating recurring findings, and updating five modules independently before integration. On TerminalBench 2.0, using DeepSeek-V4-Flash-Preview, accuracy increased from 47.57% to 52.43% against the Terminus-2 baseline. Evolution tasks excluded benchmark instances and task-specific content, but benchmark categories guided coverage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.