Reproduce “HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs” 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
Zhuoshi Pan, ZhiYuan Feng, Yukuo Cen, Honglin Lin, Junru Lu, Zhongxia Chen, Shuo Yang, H. Vicky Zhao and 1 more
HarnessPro evolves agent harnesses by analyzing execution traces to identify failures and redundant work. It uses generated tests to compare candidates and selects Lite, Medium, and Heavy options balancing pass rate and billed cost. With GPT-5.5 and Codex, it averages a 2.0-point pass-rate gain and 26.1% lower per-task cost than RHO across three benchmarks, unweighted benchmark-level differences excluding evolution overhead.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.