Agent: Slug & Assigned Prompt

The single most important field on the Agent node is the agent slug. It's the link between the node you place and the agent the operator configured — and setting it correctly is what makes everything else fall into place.

What the slug does

When you set the agent slug, the node looks into the run's workflow_config_snapshot and finds the agent assignment whose agent_slug matches. That assignment is the operator's decision about this role: which model it uses, what parameters, and — importantly — which prompt. So by naming an agent, you're pulling in its entire configured setup, including the assigned prompt (the active version of it), resolved and ready.

Think of it in the employees framing: the slug says which employee is doing this step, and the operator has already given that employee their job (the prompt) and their tools (the model). You point at the employee; they arrive fully briefed.

The prompt is the operator's, not yours

This is the crucial habit. The prompt text is not something you write in n8n — it comes from the snapshot, because the operator owns it. They author it, version it, and activate the version they want, all in the Circus UI. Your Agent node never contains prompt wording; it just names the agent, and the right prompt rides along in the assignment. That's why the operator can rewrite or roll back a prompt without you touching the workflow.

Match the slug the operator configured

Because the slug is the join key, it has to match an agent the operator actually put in the workflow's configuration. If it doesn't, the node can't resolve an assignment — it logs the mismatch, terminates the run, and errors, rather than guessing. So coordinate slugs with the operator: the name you place must be one they've configured. When it matches, the model, prompt, and parameters simply appear.