Schedules & runs
Put a plan on a schedule, or trigger it on demand. Both paths obey the same policy — there is no unattended shortcut — and both leave the same record: a run, with a state, a trigger, and a full execution log.
Scheduling
The blueprint panel’s Frequency section offers Manual, Hourly (every hour on the hour), Daily, Weekdays (Monday–Friday), Weekly (with a day-of-week picker), or Monthly (the first of each month), with a time where the cadence needs one. Anything the picker can’t express can be set as a standard five-field cron expression through the planning conversation; the panel then shows it as a custom schedule.
Schedules run in an IANA timezone — the one you set, or your saved preference, or your browser’s — so an 8:00 schedule stays 8:00 through daylight-saving changes. An Enabled toggle pauses the schedule without losing it; a paused agent shows a Paused badge and can still be run manually.
Three things can start a run, and each run records which one did:
| Trigger | Started by |
|---|---|
| Scheduled | The cron schedule firing. |
| Manual | Run Now on the agent’s detail page or in planning. |
| Webhook | An external service calling the agent’s webhook trigger. |
The run lifecycle
| State | Meaning |
|---|---|
| Ready / Dispatched | Queued and being picked up. |
| Running | The plan is executing. |
| Awaiting approval | Paused on a human — an approval, an input request, or a service reconnect. |
| Succeeded | The plan completed. |
| Failed | A step or the run errored; the terminal reason says what. |
| Skipped | Another run for this agent was still active when the slot fired. |
| Abandoned | The run stalled and the scheduler closed it — after 6 hours without progress, or 7 days waiting on a human. |
| Dispatch failed | The run could not start — usually an identity or session problem, named in the terminal reason. |
Two behaviors here are guarantees. Only one run per agent is active at a time — an overlapping slot becomes a skip, never a second concurrent run, and a run waiting on approval still counts as active, so an agent cannot pile up half-approved work. And missed slots catch up: if the platform was down over a scheduled time, missed runs are made up in order when it returns, up to a capped number.
A run that produces no output for 15 minutes is aborted and failed. A run pausing for approval is fine — that’s a state, not silence — but a plan step that genuinely takes longer than 15 minutes of silent work will be cut off.
Monitoring runs
Three views, in increasing depth:
- The dashboard shows your most recent Task Agent runs at a glance.
- Task Agents lists every agent with its latest run state and next run time. Filter by status (active, drafts, scheduled, on-demand); with team access, switch the owner filter from Mine to Everyone.
- The agent’s detail page holds the full Run History — status, trigger, start time, duration per run — and each run opens into Run Details and the Execution Log: every step’s input, result, and error, any pauses, and the final deliverable or failure report.
There are no failure notifications today — no email or chat ping when a run fails or pauses. Make the run list part of your rhythm for agents that matter, and see Get help when a run needs diagnosing.
A run that pauses for approval resumes when the decision is made from the run page — approving from there both releases the call and drives the run forward. See the approval queue for who can act.