@skills: Attention Is All You Have — A New Protocol for Agent Skill Delivery

 New paper: Attention is All You Have.
56,804 agent skills exist. Your model only has ~100 reliable trigger slots.
Every install dumps a skill’s description into the permanent system prompt. They all fight for the same scarce attention. The long tail dies. Your team’s private playbooks compete with random public ones. Token tax every message. Dilution. Reliability collapse.
Installation was never the right primitive. It bundled three things that don’t belong together: content, persistence, and auto-triggering. Only the last one needs to live in the prompt.
The new skill protocol separates them.
- A path addresses any skill, subtree, or collection.
- Reading it is using it. Zero residency.
- :save vendors a clean copy into your project’s git tree so you own and adapt it.
- :install adds one .gitignore-style line. That’s the only thing that costs prompt space.
Any agent that can read files + run commands becomes a client with a single instruction file. Optional free hub for search + ranking across the entire corpus.
Install less. Use more.
Attention is all you have.
A collaboration between AdaL and The University of Texas at Austin (Atlas Wang)

This post checks out almost entirely. There's a real arXiv paper titled "@skills: Attention is all you have" (arXiv:2608.12610) by Li Yin and five co-authors, and its abstract confirms the exact figures cited in the post — there are 56,804 public agent skills today, with the dominant installation model leaving a skill's description resident in the system prompt and competing for fewer than 100 reliable trigger slots. The paper's own framing matches the post closely: installation bundles three separable functions — content, persistence, and automatic triggering — when only triggering actually requires prompt residency, and the protocol lets a path address any skill or collection, with reading alone sufficient to use it, so nothing is installed or made resident by default. Li Yin is a real, identifiable AI professional (affiliated with AdaL) who posted this exact content to LinkedIn, so the attribution is genuine rather than fabricated. Atlas Wang (Zhangyang Wang) is also a real, verifiable figure — a tenured Associate Professor in ECE at UT Austin (currently on leave, serving as Research Director at XTX Markets), which makes the "collaboration with UT Austin" claim credible, though the post doesn't specify exactly what form that collaboration took (co-authorship, advising, etc. — worth a closer look at the paper's actual author list for that detail). The one thing outside independent verification is whether the paper's engineering claims (that the protocol actually solves reliability collapse in practice) hold up — that's a research claim, not a checkable fact, and would need the field's response over time to assess.