Reproduce “StateHarness: Maintaining Evolving Project State for Long-Horizon Coding Agents” 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
Chuan Zou, Yujun Zhou, Xiaonan Luo, Xiangliang Zhang
StateHarness is a runtime that carries project records and a compact working brief into later coding decisions, including hypotheses, findings, affected-file candidates, and unfinished checks. Against Claude Code, it raises RefactorBench cleanup-check accuracy from 87.0% to 95.4%, while ChainSWE issue resolution rises from 35.1% to 39.5%. However, ChainSWE pass-to-pass regressions also increase from 16 to 23.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.