Reproduce “@skills: Attention is all you have” 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
Li Yin, Zhi Li, Zhan Shi, Haoran Zhang, Haebin Seong, Zhangyang, Wang
The `@skills` protocol is introduced to address the inefficient management of AI agent skills, which currently overburdens Large Language Models' (LLMs) prompt context. This protocol decouples skill content delivery, persistence, and auto-triggering into a three-tiered system, leveraging existing file system conventions and Git for efficient, scalable, and transparent skill management across an ecosystem of 56,804 identified skills.
Key insights
- The traditional 'install-only' approach for agent skills creates a fundamental mismatch between the physical limitations of LLM context and the growing volume of procedural knowledge.
- Decoupling skill functions into content access, persistence, and auto-triggering is crucial for creating a scalable and efficient ecosystem that respects LLM context as a finite resource.
- Leveraging established software engineering principles like file system organization and Git for versioning and sharing can provide a robust and agent-neutral foundation for skill management.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.