Reproduce “RSIFlow: Adaptive Self-Improvement Workflows with an Evolving Meta-Agent” 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

Huan Zhang, Mingju Chen, Dongxu Zhou, Can Lv, Sen Cui, Faguo Wu, Shiji Zhou

RSIFlow adapts LLM-agent improvement by using execution traces and environmental feedback to choose among model updates, reusable harness changes, and task-output repairs. Its meta-agent learns from intervention outcomes to refine its harness and skills while keeping its base-model weights fixed. Across six benchmarks and two task-agent backbones, it beats the best baseline in 11 of 12 settings but trails HarnessForge on Qwen3-8B ACEBench.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.