Reproduce “Program-Aided Patchable Skill Evolution” 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

Jiahao Wang, Yuqi Xiong, Qing Yang, Zhipeng Xu, Zhenghao Liu, Zulong Chen, Yubin Bao

PASE separates agent knowledge into natural-language guidance for task strategy and an executable program for repeatable procedures, allowing either component to be updated separately. The authors evaluate it on spreadsheet-centered tasks, with evolution using 80 training and 40 validation examples and 280 held out for evaluation. On SpreadsheetBench-Verified with Qwen3.6-27B, it reaches 82.98% verified accuracy versus EvoSkill’s 68.10%, in this setup.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.