Skip to content

Extensions

Install or build packages that ship skills, agents, automation, layouts, hooks, MCP servers, provide surfaces, and subprocess tools.

For people running agent work11 pages in this section

An extension is one installable unit that ships runtime behavior and resources behind a single manifest. Installation enables the extension by default for every profile. A per-profile disable keeps the package installed while hiding its resources from that profile.

Use extensions when runtime behavior or resources should travel between machines or teams with one install command. If the work is only a local instruction set, a skill may be enough. If it includes multiple resources, provide surfaces, or a daemon-facing Host API service, an extension gives that package a manifest, trust tier, per-profile enablement state, and operational lifecycle.

Authors of executable extensions start at Build your first extension: three commands from an empty directory to a tool you can invoke. Executable authoring is code-first, while resource-only authors can hand-write extension.toml and use the same build and workspace development lifecycle.

Operators start at the install page for adding, enabling, disabling, inspecting, updating, or removing an extension.

In this section

On this page