Reproduce “Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams” 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

Zewen Liu, Zhan Shi, Yisi Sang, Bing He, Minhua Lin, Tianxin Wei, Dakuo Wang, Benoit Dumoulin and 2 more

The authors study Adaptive Auto-Harness, a framework for improving agents on task streams without changing solver weights. It separates a workspace for revising experience from the harness used on each task, through four phases: analyze, research, build, and verify. On PolyBench, it reaches 80.9% accuracy versus 22.2% for static Sonnet; this recorded-stream result does not establish profit or durability.

Summary by alphaXiv.

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