The execution contract: when your task runs and when it doesn't
After this, you'll understand the two conditions required for a Cowork task to fire, what the help center documents about a missed run, and how scheduled tasks spend from your usage quota.
Before you start
Complete The absent author first. The execution contract explains why the absent author mental model from the previous lesson matters at the system level, not just the prompt level.
The idea
A Cowork task only fires when two things are true at the scheduled time: your computer is awake and Claude Desktop is open. Miss either one and that scheduled time passes with nothing produced. So the schedule you design has to fit when your machine is actually on, not the hour you wish the result would land.
Here is the before and after: You set a news briefing for 6 AM, picture it waiting when you wake, but your laptop is closed at 6 AM, so nothing fires. The help center says a skipped task "runs automatically once your computer wakes up or you open the desktop app again," with a notification, so you get one fresh run on wake, not the four briefings four skipped mornings might suggest. The fix is to schedule for when the machine is reliably on, then design the prompt so a missed run costs you nothing you were counting on.
Now try it: Tell Claude "I open my laptop around [time] and close it around [time]. I want to run [your task] on a schedule. Which frequency fires reliably given my hours, and what should I avoid?" Let it map your real on-hours to a safe choice before you touch the scheduler.
- Scheduled time arrivesThe clock hits the time you set for the task.
- Computer awake?Your machine has to be on, not asleep or shut down.
- Claude Desktop open?The Claude desktop app has to be running, not just installed.
- Task runsBoth conditions met, so Claude does the work and uses a little of your usage quota.
- Run skippedOne condition failed, so nothing is produced at that time.
- Runs once on next wakeThe help center promises one make-up run when you wake the machine or reopen the app, not one for every missed day.
Two conditions, both true at the scheduled time: computer awake and Claude Desktop open. Miss either and the run is skipped, so build the schedule around your machine's real on-hours, not your reading hours.
Try it (10 min)
Watch out for
- Scheduling tasks for times when your laptop is reliably closed. The help center says a skipped task runs once your computer wakes, not at the hour you picked, so the result can arrive much later than you pictured. Choose a frequency that lands inside the hours your machine is actually on.
- Designing a schedule around catch-up runs. The help center promises one run on wake, not that every miss replays and not a queued backlog. Treat a missed run as a run you did not get rather than something you can count on recovering.
- Ignoring quota for the first few weeks, then hitting limits during a busy period. If you are running heavy multi-connector tasks daily, check your usage once a week until you know your baseline.
Paste this into Claude
Help me audit my computer habits against the Cowork execution contract. I would rather click answers than type them, so present each question as multiple choice. (The feature that gives you clickable multiple-choice options is called AskUserQuestion. Use it.) Ask me these three, one at a time: 1. What time do I typically open my laptop in the morning, and what time do I close it at night? 2. Do I leave my laptop open and plugged in overnight, or do I close it? 3. How many scheduled tasks am I planning to run, and how heavy are they (a simple file read vs. pulling from several connected apps)? Based on my answers, tell me: - Which time windows are reliable for scheduling tasks - Which frequencies (hourly, daily, weekdays, weekly) match my actual computer habits - Whether my planned task volume could put pressure on my quota, and what plan that implies - Any gaps between what I want to automate and what my hardware habits will allow
What a good response looks like
Based on your answers: **Reliable scheduling windows:** Mid-morning to early evening on weekdays. You open your laptop after breakfast and close it at night, so tasks set to fire inside that window have a high chance of running on time. **Frequencies that match your habits:** Daily (one run per day during your open window) and Weekdays both fit. Hourly would only fire while the laptop is open. A Weekly run on Friday afternoon is reliable. **Quota assessment:** Two daily tasks, one connecting to Gmail and one reading a local folder. A standard Pro plan handles this comfortably. If you add multi-connector tasks later, watch your usage before adding more. **Gap identified:** A pre-dawn news briefing would not run because your laptop is closed then. Either leave the laptop open overnight, or accept that the help center says it runs once on wake (not once per missed day).
What good looks like
- You can state the two conditions required for a scheduled task to run without looking them up
- You can state what the help center actually promises about a missed run, and what it does not promise
- You have identified at least one time window when your computer is reliably on and Claude is open
- You know whether your planned task volume requires attention to quota
When this breaks
- Breaks when the schedule is designed around when you want the output, not when your computer is on. A pre-dawn briefing is only reliable if your computer is awake and Claude is open at that hour. Design the schedule around your machine's habits, not your reading habits.
- Breaks when you assume a missed run is recoverable on a fixed rule. The help center documents one run on wake, not a per-day replay, and a reported stability issue (issue #47899, Windows-specific) can stop even that one when a stuck run blocks the queue. Keep your task count small, and check that recent runs actually produced output rather than assuming they did.
AI can help with this
Open Claude in the exact surface named by the lesson, then run the lesson prompt with one real recurring task. Confirm the source access, schedule condition, output location, and failure behavior before you trust the result.
You can now
You can complete the lesson outcome inside the correct Claude scheduling surface.
- You can state the two conditions required for a scheduled task to run without looking them up.
- You can state what the help center actually promises about a missed run, and what it does not promise.
- You have identified at least one time window when your computer is reliably on and Claude is open.
- You know whether your planned task volume requires attention to quota.
Key takeaways
Two conditions must both be true at the scheduled time: computer awake and Claude Desktop open. Missing either skips the run, and the help center only promises one run on wake, so schedule for when your machine is actually on.
- The execution contract has two conditions: computer awake AND Claude Desktop open. One missing means the task is skipped, not delayed.
- The help center documents one behavior for a missed run: it runs once your computer wakes or you reopen the app, with a notification. It does not promise a per-day replay or a queued backlog, so never design a schedule that depends on a catch-up.
- A catch-up run that does fire still reflects the missed scheduled time, so anchor your prompt to a clear time window in case it lands later than planned.
- Quota is how much Claude usage your plan allows before it pauses you. Scheduled runs spend from the same pool as hands-on work, so heavy multi-connector daily tasks can drain a Pro plan; Anthropic lists Max (with higher current usage than Pro) as the next tier up for that load.
Go deeper