Reproduce “PluginRSI: Recursive Improvement of Agent Harnesses with Reusable Plugins” 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
Yaorui Shi, Yuchun Miao, Yuxin Chen, Jiayuan Zhang, Yueqing Sun, Xierui Song, Xiang Wang, An Zhang
PluginRSI improves AI-agent harnesses by evolving plugins separately, retaining promising variants, and recombining them into harnesses without changing model weights. Its search combines plugin mutation with workflow recomposition, using model calls for evaluation. On SWE-bench Verified, its selected harness resolves 69% of 100 held-out tasks, versus 63% for Meta-Harness; this is task-level generalization within the benchmark, not unseen-repository transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.