Skip to main content
A cron trigger starts a task on a schedule with no one present. This guide creates one, proves it fired, and sets the limits that keep an unattended agent from running away.

​
Prerequisites

  • An agent with a working model — a scheduled task against a misconfigured agent fails on every tick. See create and configure an agent.
  • A running Temporal worker. Cron triggers are Temporal schedules; without a worker the schedule fires and nothing picks it up.
  • An access token, exported as AGENTAREA_TOKEN, and the API base URL as AGENTAREA_URL.

​
Steps

1

Create the trigger

agent_id, name, and trigger_type are the only required fields. cron_expression is required in practice for a cron trigger — creation is rejected without it.
The schedule is registered in Temporal as cron-trigger-<trigger_id>, so it also appears in the Temporal UI and can be paused there. Pausing it there does not update the trigger record — prefer the API below.
2

Fire it once, without waiting for the schedule

Do not wait until 09:00 to find out the agent’s model is unset.
This runs the same path the schedule runs, including conditions, and records a TriggerExecution exactly as a scheduled tick would.
3

Put a ceiling on it

A schedule multiplies whatever one run costs. An agent that costs a little and runs hourly costs a lot by the end of the month, and nobody is watching at 03:00.Set the budget on the agent or the workspace rather than on the trigger — the trigger has no budget of its own. See set a budget.
If the trigger uses the default model-evaluated conditions, every candidate event pays for an inference before a task exists, so that spend is not attributed to the agent’s task budget.

​
Verify

Read the execution history. This is the record of what the schedule actually did, including ticks that created no task:
status: "success" with a non-null task_id means the tick created a task. status: "success" with task_id: null means the trigger ran and its conditions declined to create one — that is a working trigger, not a broken one.
Follow task_id into debug a failed task if the task itself went wrong. A trigger’s job ends when the task exists.

​
Troubleshooting

Cron triggers are Temporal schedules. Check that the Temporal service is reachable and that a worker is running — the schedule fires into a task queue and a tick with no worker produces nothing. Look for cron-trigger-<trigger_id> in the Temporal UI: if the schedule is absent, creation did not reach Temporal; if it is present and firing, the problem is downstream.
timezone defaults to UTC, not to the server’s local zone or the workspace’s. A 0 9 * * * trigger created without timezone fires at 09:00 UTC. Set an explicit IANA name.
Crossing failure_threshold consecutive failures disables the trigger. That is the safety mechanism working. Nothing re-enables it when the cause is fixed — read the recent failures, fix the cause, then re-enable explicitly:
The conditions are declining every event. A condition whose type is left unset is evaluated by a model, so its verdict is not a bug in your expression — read trigger_data on the execution record to see what the model was given.
polling is accepted by the API and has no implementation behind it. Use a cron trigger with data_extractor for periodic fetching. See the limits in triggers.

Triggers and channels

Why scheduling lives in Temporal, and what a channel does.

Debug a failed task

When the trigger worked and the task did not.

Set a budget

The ceiling an unattended schedule needs.

Durable execution

The Temporal machinery underneath.
Last modified on September 17, 2026