Reproduce “SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision” 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
Yuxuan Liu, Zhaochen Su, Lingyun Xie, Yuhao Zhang, Qing Zong, Jiahe Guo, Zhongwei Xie, Yiyan Ji and 6 more
SkillRevise revises reusable agent skills for multi-step tasks using execution evidence. It runs a skill, diagnoses failed and preserved checks, applies repair principles, and tests revisions, selecting the highest-utility version. On SkillsBench, GPT-5.5 verifier-defined success rose from 31/86 without a skill to 53/86 after up to three revisions; repair adds cost, and selection cannot choose no skill.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.