Schedule recurring Agent work

Create a named recurring responsibility that starts a fresh bounded Run for one exact Agent revision at an elapsed interval or a daily, weekly, or monthly wall-clock time.

Prerequisites

• Full control access.
• An active Agent returned by Crewhelm.
• A bounded recurring instruction.
• Any exact Brief revisions the recurring responsibility needs.
• A trigger: interval, or calendar time with an IANA time zone.
• Capacity in the Agent’s eight shared recurring-start slots for Schedules and Event Triggers.

Authority and custody

A Schedule freezes one Agent revision, instruction, optional exact Brief revisions, trigger, output contract, and schedule limits. It supplies an occurrence, not authority. Every occurrence still passes normal Run admission and Tool gate checks.

Crewhelm owns alarm timing, one deterministic admission identity per occurrence, duplicate suppression, history, and recovery. A late alarm advances to the next future occurrence instead of replaying a backlog.

Create a Schedule

1. Call crewhelm_inspect_automations with operation.kind: "list_schedules" and the returned Agent object to review existing responsibilities and available capacity.
2. Call crewhelm_change_automations with operation.kind: "create_schedule", that Agent, and the schedule definition.
3. For an interval, choose from 60 seconds through 7 days. The first Run occurs one interval after creation.
4. For calendar timing, provide HH:mm, an IANA time zone, and the required daily, weekly, or monthly selector. A monthly day that does not exist in a month is skipped.
5. Omit the output contract for Markdown, or freeze one bounded JSON contract for every scheduled Run.
6. Add only the returned Brief objects needed for this recurring responsibility. Crewhelm keeps their exact immutable revisions internally. Crewhelm validates them without requiring the MCP client to read and resend their content.
7. Keep the returned Schedule object unchanged.

Verify the Schedule

1. Call crewhelm_inspect_automations with operation.kind: "inspect_schedule" and the returned Schedule object.
2. Confirm its status, frozen Agent revision, trigger, and next dispatch time.
3. After an occurrence, inspect its most recent scheduled Run.

Pause or update

• Pause one exact Schedule through crewhelm_change_automations with operation.kind: "pause_schedule" and the returned Schedule object.
• Update a paused Schedule with operation.kind: "update_schedule", that Schedule object, and a new definition to reuse its bounded slot.
• Inspect the Schedule before every lifecycle mutation and pass back the newly returned object.

An Agent revision change makes the bound revision stale. Crewhelm pauses or skips the Schedule instead of silently retargeting it.

Recover safely

• If a Schedule is busy, inspect it and retry the lifecycle change only after the pending admission settles. Do not infer whether its Run started.
• On a revision conflict, inspect the Schedule again before deciding to update.
• Pause the Schedule or update its current configuration without the Brief to release the recurring reference. Deleting the Brief still fails closed while any admitted Run or Workflow retains the exact revision; inspect that resource and follow its retention lifecycle.
• Inspect the most recent occurrence and Run before manually replacing delayed work. Late alarms do not replay missed intervals.
• Pausing prevents future occurrences; it cannot undo an external effect from a Run already dispatched.

Next action

Use an Event Trigger instead when a supported connected-app event, not time, should start the work.