Skip to main content
Start from the data boundary, then add runtime dependencies and agents.

​
1. Define input

input {
    project_id: number
    task_id: number
}

​
2. Define secrets

secrets {
    api_key: string
}
Keep secrets small and explicit. Pass provider keys and MCP tokens through the executor request.

​
3. Add provider and model

provider llm from openai {
    endpoint: "https://api.openai.com/v1"
    api_key: secrets.api_key
}

model fast from llm {
    id: "gpt-4.1-mini"
}

​
4. Add schemas

schema task_summary {
    summary: string
    next_action: string
}

​
5. Add agents

agent summarize_task {
    model: model.fast
    instruction: "Summarize task {{ input.task_id }} in project {{ input.project_id }}."
    output: schema.task_summary
}

​
6. Add final output

output {
    task_id: input.task_id
    summary: agent.summarize_task.summary
    next_action: agent.summarize_task.next_action
}

​
Review checklist

  • Inputs and secrets match the executor payload.
  • Provider and model declarations use the new provider ... from ... and model ... from ... syntax.
  • Agent outputs are object-shaped or schema references.
  • Tool runtime values use bindings {}.
  • uses includes every tool, prompt, and resource an agent needs.
  • Dependencies are clear from references.