Skip to main content
The Skill Master is where you author Skills — reusable tasks in the open agentskills.io format.

​
What a skill is on disk

Creating a skill makes a folder with a single SKILL.md inside it. That file is the skill: YAML frontmatter (name, description, and the tools the skill is allowed to use) followed by the instructions the agent reads at run time. The folder on disk is the source of truth — everything else is layered on top of it.
The working surface today is deliberately small: create a skill and edit its SKILL.md. Importing skills, testing, and version history build on that core (below) and are still stabilizing.

​
Create or import

  • Create — give the skill a name and description; the Hub scaffolds the folder and a starter SKILL.md.
  • Import — point at an existing skill folder. Bundled scripts/, data/, and references/ are preserved as read-only assets (scripts are not executed in V1).
  • Example skills — start from a worked example instead of a blank file.

​
Choose only the tools the skill needs

In the SKILL.md frontmatter you list the tools the skill may call, referencing each by name with @ autocomplete. Those names resolve at run time to:
  • built-in tools — current_time, web_fetch, and web_search (web_search needs a Tavily key; the other two need no setup), and
  • connector-served tools — actions from a Connector you’ve authorized.
Don’t add every available tool. Declare only the tools the skill actually uses. A tight tool list keeps the agent focused, makes runs cheaper and faster, and stops the model from reaching for an unrelated action. Add a tool when the skill needs it — not up front “just in case.”

​
Test and version

You can run a skill against a model to see what it does, and commit a snapshot of the SKILL.md to keep a version history. Skills use the same diff-and-restore model as prompts — a commit tracks the SKILL.md only — so see Comparing versions for how it works.
Screenshots and a worked example (authoring a skill end to end) are being added to this page.