Reproduce “Evolving to Match Capabilities: Improving Harness Self-Evolution via Skill Adaptation” 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
Zhihao Yang, Haojin Wang, Juncheng Wu, Kuan Zhang, Yukun Chen, Run Luo, Yaojie Lu, Yuyin Zhou and 2 more
The authors evaluate harness self-evolution, which adapts agent skills to the model that must use them without changing model parameters. They diagnose executor failures and rewrite skill instructions, using an external rewriter or the executor itself. Across 103 selected task–skill pairs from three benchmarks, skills raise GPT-5.6-luna’s score by 31.1 points but lower Gemma-4-E4B’s by 8.8, on this selected testbed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.