Reproduce “AtomSearch: Evidence-Grounded Tree Search for Agent Skill Optimization” 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
Jia Tong, Senyou Deng, Min Peng, Jianfeng Pan
AtomSearch adapts a frozen agent through editable skill instructions. It proposes evidence-linked edits and searches executable skill states, testing later compositions despite weak earlier scores. On LiveMathematicianBench, deepseek-v4-flash scores 52.42% versus SkillOpt’s 42.74%, leading on three of four tasks. In 20 of 30 searches, the best validation state had a non-top-scoring ancestor; this does not show every weak state helps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.