Reproduce “EvoSkill-S1: Evolving Reusable Skills with Genetic-Inspired Operators” 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
Yujun Wang, Jiqiang Li, Yinuo Lu, Jinhe Bi, Aniri, Wenxuan Ye, Boliang Liu, Sikuan Yan and 5 more
EvoSkill-S1 improves reusable skills by using failed problems to train a separate skill evolver, while keeping the base solver fixed. It tests revised skills and combines five operator-specific banks under a shared size budget. On Qwen3-4B-Instruct’s held-out AIME24, the large fused bank reaches 67.08% accuracy versus 61.67% for the initialized bank, a result limited to this benchmark.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.