Scheduled work and monitors
Ask Claude to repeat useful work, keep a run history, and surface jobs that stop reporting.
Some tasks are valuable because they happen again: a morning brief, a source check, a weekly report, or a data refresh. Podbay lets Claude turn that request into recurring work that uses the same project, tools, and context as your normal session.
What to ask for
Tell Claude the outcome, timing, timezone, and reporting threshold:
Every weekday at 08:30 Europe/London, check our configured sources and update the research brief. Tell me what changed and flag only items that could affect this week's launch.
Every Friday afternoon, review open project risks, test the current build, and leave a short report in the repository. Do not publish or message anyone.
Check this page every two hours. Alert me when the price or availability changes, not when the surrounding layout changes.
Claude can create the schedule, verify it, and explain when the next run will happen. You do not need to translate the request into a cron expression or a Podbay command.
What makes recurring work useful
Include:
- when: local time, timezone, days, or interval;
- where to look: the sources, inbox, database, or project files involved;
- what matters: a threshold or definition of meaningful change;
- where the result belongs: the Claude session, a project file, database, or private dashboard;
- what Claude may do: whether it should only report, prepare a draft, or make a local change;
- what still needs approval: publishing, sending, purchasing, or other external actions.
The more often a job runs, the more important it is to define what should be ignored. “Tell me only when this changes materially” is usually better than receiving the same status every hour.
What Podbay keeps track of
Podbay records when a run starts. When Claude reports completion, the durable history also records whether the run succeeded or failed. If a run stays open during the same pod boot, Podbay raises one stalled-run warning instead of quietly treating it as successful.
A restart can interrupt work before Claude records the result. That run remains visible in the log, but Podbay does not currently raise the stalled warning after the pod comes back. For important jobs, ask Claude to make the work safe to repeat and to verify the expected output on the next run.
Some environments add a more opinionated monitor workflow: save a baseline, compare new results, apply thresholds, avoid duplicate alerts, observe a cooldown, and report recovery. Those prepared workflows depend on the selected environment; recurring agent work and its run history are common pod capabilities.
When the pod is off
A suspended pod cannot do work. A missed daily job can catch up after the pod resumes on the same local day, but it cannot produce an on-time result while the machine is stopped. Keep the pod running when timing matters.
A schedule is not blanket permission
Recurring work follows the same boundaries as an interactive request. Scheduling “check our leads” does not silently authorize Claude to email them, change an account, spend money, or publish a result.