Reproduce “Just-in-Time Skill Retrieval via Long-Horizon State Tracking” 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
Xin Lin, Ziheng Wang, Zerui Chen, Jiayao Ma, Zhimin Wei, Bingshuai Liu, Buchao Zhan, Qinggang Zhang and 1 more
JitSkill helps long-running agents use reusable skills by tracking task progress and using that state to retrieve skills and guide execution. It maintains explicit and recurrent latent states updated from actions and feedback. Ablations indicate dynamic retrieval and progress-conditioned execution contribute to completion. On ALFWorld, it reaches 92.38% success versus CUA-Skill’s 90.48% under shared conditions; this measures task completion, not efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.