main.agent

// example1: a minimal storefront support-snippet workflow, authored entirely in
// .agent (ADR 003, issue #430). A native `helper` tool supplies the return-policy
// context; the `support_writer` agent (openai/gpt-4o-mini) drafts the customer
// email fields. No project.yaml — the built-in `mock`/`openai` providers and the
// default `local` runtime are implicit.
//
// Note (issue #440): the original YAML workflow produced a *flat* multi-field
// output (`output.value: {product, subject, line}`) composed across two steps.
// `.agent` has no object-literal return, so the agent now emits the whole object
// (it already receives `product`) and the workflow returns it — the same fields,
// wrapped in the standard `.agent` `value` envelope.

policy default {
    execution {
        maxTotalCostUsd 5
        maxWallClockSeconds 300
    }
}

tool helper {
    type native
    safety {
        sideEffects false
    }
}

agent support_writer {
    model openai/gpt-4o-mini
    policy default
    constraints {
        timeoutSeconds 60
    }
    instructions """
    You draft short customer-facing email lines for a storefront.
    You receive JSON in the user message: product name and a return-policy line from internal systems.
    Respond with one JSON object only (no markdown, no code fences).
    Use exactly this shape: {"product": "<the product name, echoed back>", "subject": "<=8 words>", "line": "<=25 words, friendly>"}
    """
}

workflow support_snippet(input: any) policy default {
    context = helper.echo(product: input.product, policy_line: "30-day returns on all SKUs; free outbound shipping on defects.")
    snippet = support_writer(product: context.echo.product, return_policy: context.echo.policy_line)
    return snippet
}