Launch-day run of a Sume schedule: one key per model, input as data

To check a new model the day it launches without waiting for the weekly cron, enable the API trigger on the schedule and call POST /v1/actions/{action_id}/runs. Send the vendor and model as input, and build the Idempotency-Key from the model name so a repeated announcement replays the same run instead of starting a second one.

What the call needs

The call needs a key with actions:read and actions:write, a schedule whose status is active, and api_trigger_enabled set to true. A cron schedule can accept both triggers, so the weekly cadence keeps running. Keys created before the trigger shipped lack these scopes and fail with 403 insufficient_scope. You cannot add scopes to an existing key, so mint a new one.

Input is data

Sume serializes input into a fenced JSON block and gives it to the agent as data, not instructions. The saved instructions stay in charge. That suits a launch check well. The instructions say what to verify for any model, and input says which one, for example MAI-Voice-2.1-Flash from Microsoft. Do not design the schedule so that input changes what the schedule does.

Overlap on a busy day

Only one run of a schedule is active at a time, and on_active_run decides what a second request gets. The default skip returns 200 with a receipt whose status is skipped and skip_reason is previous_run_active. The value reject returns 409 action_run_in_progress. On launch day, when two vendors announce within the hour, reject makes the dropped trigger visible to your caller.

The trigger

Branch on the receipt's status, not the HTTP code, since 200 can mean either a replay or a skip. The script below builds a stable key from the model name, asks for rejection on overlap, and prints what happened. Replace the schedule id with the opaque aut_... id and store that, not the vanity URL.

import json, os, re, urllib.request, urllib.error

def main():
    vendor, model = "Microsoft AI", "MAI-Voice-2.1-Flash"
    key = "launch-" + re.sub(r"[^a-z0-9]+", "-", model.lower())
    body = {"input": {"vendor": vendor, "model": model},
            "on_active_run": "reject"}
    url = "https://api.sume.com/v1/actions/" + os.environ["ACTION_ID"] + "/runs"
    req = urllib.request.Request(url, data=json.dumps(body).encode(), method="POST",
      headers={"Authorization": "Bearer " + os.environ["SUME_API_KEY"],
               "Content-Type": "application/json", "Idempotency-Key": key})
    try:
        with urllib.request.urlopen(req) as r:
            d = json.load(r)["data"]
            print(r.status, d["status"], d["idempotency_hit"])
    except urllib.error.HTTPError as e:
        print(e.code, json.load(e)["error"]["code"])

main()

Why the key comes from the model

Because a key is bound to its payload, a later run for the same model with different input gets a conflict. That is the behavior you want, since it forces you to choose a new key on purpose when you really want a second look.