Reproduce “SkillDuet: Co-Evolving Models and Skills through Response Feedback” 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
Xingguang Ji, Ziqiang Dong, Yubo Zhang, Xi Leng, Yang Yang, Guanjun Jiang
SkillDuet couples policy learning with the selection of persistent written skills for simulated-task agents. Skills guide on-policy distillation, while the model’s response shifts on existing trajectories rank recent skills for consolidation without new environment interactions. On WebShop with Qwen3-4B, it reaches 92.97% complete-task success versus 87.50% for SEED-API; response shifts indicate change, not future benefit.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.