Reproduce “Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries” 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
Yubo Li
This survey paper from Carnegie Mellon University provides a systematic framework for understanding dynamic agent skills by introducing a six-sense skill taxonomy, an eight-stage lifecycle architecture, and a ten-operator vocabulary for managing evolving skill libraries. It synthesizes evidence-graded patterns on system design, verification, and maintenance, identifying key challenges for scalable and safe LLM agent deployments.
Key insights
- Dynamic skill systems require robust, write-time verification and admission gates, as unverified self-generated skills lead to performance degradation.
- Effective management of skill libraries necessitates dedicated maintenance operations (e.g., pruning, merging, refining) as libraries grow, rather than relying solely on initial skill acquisition.
- The challenges of dynamic skill systems extend beyond individual skill performance to systemic software supply-chain problems, requiring comprehensive safety and governance mechanisms across the entire lifecycle.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.