Skills in Code

Turn on skills for Code#

Skills & Plugins in the Code sidebar has Marketplace and Installed, as in Chat. Installed skills are shared between Chat and Code, but each has its own on/off switch. A skill can be on for Chat and off for Code, or the reverse. Chat sign-ins aren’t shared with the coding runtime.

Open a plugin to turn individual skills on or off. A skill that needs a connector works after the connector is installed and connected. If the same skill is in two plugins, it has one switch.

To add a key, use the Code control on the connector’s card. If a connector’s sign-in returns to the desktop app, finish signing in there. The card then shows whether the connector is installed, enabled, signed in, or waiting for a key, with the matching action: Sign in, Add key, Disconnect, Clear key, or Remove.

Insert a skill or reference#

At the start of a message, type / to list the skills this runtime can use. Use the arrow keys to move, Enter to insert, and Esc to close the list. Type @ to add chats, files, or folders as context.

You can use both in one message. For example, type “Review the current change”, add a skill with /, then add the changed folder with @. The runtime determines which of its own commands are available. Some open an interactive view instead of running in the background.

An Agent without a skill list uses every enabled skill. A skill list on the Agent’s profile limits it to those skills. The task’s permission still applies, and a connector can still ask before it sends or changes anything.

If a skill doesn’t appear#

Check these in order: the skill is installed, it’s turned on for Code, the individual skill is on, the connection is ready, and the runtime supports it. A badge in the catalog doesn’t grant permission.

The catalog includes plugins for Codex, Claude, and Google Workspace workflows, and Skill Creator for procedures you repeat. AI Integrator Government and AI Integrator Finance are the data plugins. Ask in plain language. The assistant finds the official path or series ID before it queries the source. Cursor Cloud Agents runs work on Cursor’s service and asks you to confirm before it dispatches.

For workflows, see Government, Finance, and Law. For charts made from that data, see Charts.