Open Source arXiv cs.AI

@skills: Attention is all you have

@skillsagent skillsprotocolAdaL

There are 56,804 public agent skills today, plus many private ones, but the dominant installation model forces skill descriptions into the system prompt, competing for fewer than 100 reliable trigger slots. The long tail of skills has no practical path to use, and teams' own playbooks compete for the same scarce prompt space.

@skills is an open protocol that separates installation into content, persistence, and automatic triggering — only triggering requires prompt residency. Any skill, subtree, or collection is addressed by a path; reading a skill is sufficient to use it, so nothing is installed or made resident. The 'vendor' operation copies a skill from a path into a project's Git-tracked tree for adaptation, while the 'install' operation adds a single .gitignore-style line, the only element that costs prompt residency. A directory acts as a menu, making bundles ordinary directories rather than all-or-nothing units. No manifest, lockfile, or registration is required, and SKILL.md files remain unchanged.

The protocol is additive, ships as an installable package, and turns any agent that can read files and run commands into a client via a single instruction file. Its open specification is at GitHub, with an implementation in the AdaL CLI. Because paths address skills but cannot find them, the protocol pairs with an optional free hub for corpus-wide search and ranking, repository-free hosting, private and team collections, and one-screen authoring. gh: and local paths resolve without the hub, and indexed GitHub skills retain their gh: identities. The takeaway: install less, use more.

Read original →

← Back to home