agent

​

When to use it

• When which tool to call next depends on what an earlier call returned.
• When the order of calls is known in advance, write plain steps instead. They make no model calls.

​

How it works

• tools limits what the step can call, by name or glob. Omit it for the whole catalog.
• maxIterations caps model rounds. The default is 8.
• systemPrompt is added after the built-in prompt.
• Each round is a model call, and inside map it runs once per item.

​

Input

​

Result

- id: E1
  tool_name: agent
  input:
    prompt: "{{input.prompt}}"
    outputSchema: "{{input.outputSchema}}"

{
  "final": true,
  "iterations": 3,
  "output": {
    "classifications": [
      {
        "issue_id": "123",
        "reason": "…",
        "severity": "high"
      }
    ]
  },
  "tools_called": [
    {
      "round": 1,
      "tool": "linear__list_issues"
    }
  ]
}