Agent Skills Overview - Agent Skills

Open original

Agent Skills is an open standard where a skill is a folder with a SKILL.md file that packages instructions, scripts, and reference material so an AI agent can load them when a task calls for it. It was originally developed by Anthropic and is now open to the wider ecosystem.

  • A skill is a folder built around a SKILL.md file, with required name and description metadata plus instructions, and it can bundle scripts, templates, and reference files.
  • Agents use progressive disclosure: they load only each skill's name and description at startup, then read the full instructions only when a task matches.
  • Skills are meant to be portable and version-controlled, so one skill can be reused across any skills-compatible agent product.

Agent Skills are a lightweight, open format for giving AI agents specialized knowledge and repeatable workflows. A skill is just a folder with a SKILL.md file at its center. That file carries metadata (at minimum a name and a description) plus instructions telling an agent how to handle a particular task. A skill can also bundle scripts, reference documents, and templates. The appeal is portability: the folder is version-controlled, can be shared across teams, and can be reused across any agent that supports the format, rather than being locked to one product. The problem the format targets is familiar to anyone who has tried to get an agent to do real work. Agents are capable, but they often lack the company-, team-, or user-specific context needed to perform reliably. Skills package that context, whether it's a legal review process, a data analysis pipeline, or house rules for presentation formatting. The article argues the payoff comes in three forms: domain expertise captured as reusable instructions, multi-step tasks turned into consistent and auditable procedures, and a build-once, use-anywhere model across compatible clients. The loading mechanism is worth understanding because it explains how this scales. Agents use progressive disclosure in three stages. At startup they read only each skill's name and description, enough to judge relevance. When a task matches a description, they pull the full SKILL.md into context. During execution they follow the instructions and load bundled code or referenced files only as needed. This keeps context windows lean even when many skills are installed. The format was originally developed by Anthropic, released as an open standard, and is open to outside contributions through GitHub and Discord. The page points to a Client Showcase of supporting tools, a quickstart for building a first skill, and a full specification for anyone who wants the details.

Ask about this article