Reproduce “When Agent Skills Go Stale: Runtime Behavior and Selective Maintenance in Evolving Environments” 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
Ömer Faruk Akgül, Shuli Jiang, Yuting Zhang, Yi Zhang, Xiaohan Fei, Wei Xia, Stefano Soatto
The authors evaluate selective maintenance for agent skills across controlled changes. It replays checks from skill references and executions, verifies whether changed observations conflict with content, and repairs affected state. In a cache-heavy IRS condition, field accuracy rises from .305 to .975 as turns fall from 7.0 to 4.7. Evidence covers four settings and two Claude models, not broader deployments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.