Skill
Makes a skill available to an AI Agent or a Supervisor: its instructions, its reference files, its templates and, where applicable, its scripts.
This node is declarative: it receives nothing, transforms nothing and is not part of the document flow. It simply declares what the connected agent is allowed to draw on.
Ports
- "skill" output (right, skill type) — connect it to the "skills" port of an AI Agent or a Supervisor. Several Skill nodes can feed the same agent, and one node can feed several agents.
Configuration
- Skill — the one to make available, among those created in Manager → Skills. A disabled skill remains selectable but is never passed to the agent.
The panel recalls the skill's description, its file count and whether script execution is allowed for it, with a direct link to its page.
What the agent does with it
The agent initially receives only the name and description of each connected skill. If it judges one of them relevant, it reads its full instructions, then the files those instructions point to. That progression is what lets you connect several skills to one agent without flooding its context window.
Three tools are opened to it for this: list a skill's files, read one, and run a script.
Connecting a skill turns on the agent's tool loop, as an MCP connector does. The agent therefore makes extra model calls to consult the skill: this is visible in the execution trace and in AI costs.
A skill connected to a Supervisor
A skill connected to a Supervisor's "skills" port is shared by all of its subagents. That is how to attach shared know-how once rather than once per subagent — each subagent still being free to have its own skills.
Scripts
A skill's scripts never run on the application server, but in an isolated container with no network access and no credentials, with a workspace destroyed after every run. They must also be approved by an administrator. See Skills for details.