L2Lesson 4Free

The task anatomy: what every field in the scheduler actually does

After this, you'll understand what each scheduler field does, know that the schedule has two parts (a frequency and a Time field that prefills 9:00 AM), and be able to use Run Now to test a task before committing to a schedule.

Before you start

Complete The execution contract first. The execution contract explains when tasks run; this lesson explains what they run.

The idea

The scheduler shows more fields than you will fill in, and the ones people skip decide whether the task still works in three weeks. Name, prompt, and frequency jump out; the form also carries a Description, a run Time, a Model, and a Working folder. The skipped fields each do one job a name like "Daily task" cannot.

A scheduled Claude routine for The task anatomy: what every field in the scheduler actually does starts with a loose prompt, an uncertain clock, and unclear output delivery.

Here is the before and after: Someone names a task "Daily task," leaves the 9:00 AM time on its prefill without a thought, leaves the working folder blank, and skips the description. Three weeks later they have five tasks all called some version of "Daily," one fires at 9:00 AM and reads a calendar that already moved on, and no note anywhere says what any of them was for. The fix is the fields they skipped, each carrying weight the name cannot.

When it fires has two parts, not one. Frequency sets how often; the Time field sets the clock moment, and for Daily, Weekdays, and Weekly it prefills to 9:00 AM whether or not that suits your task. Weekly adds a day-of-week picker beside it. Runs carry a built-in delay of a few minutes for server load, so treat the time as "about then," not to the second. Set it on purpose; choosing it well is its own skill, and Module 1 is built around it.

The other skipped fields each do one job. The Description is your own note Claude never reads, so use it for what the task does and when you last touched the prompt. The Working folder is the only field that decides whether output lands in a file instead of the sidebar. The Model default is fine for most tasks.

Now try it: Describe a task to Claude and ask "Fill in every scheduler field for this, recommend a frequency and a run time based on when my computer is on, and write me a Run Now test checklist." Then run it by hand with Run Now, which fires immediately and ignores the schedule: write the prompt, run it, read the output, adjust, save only once it is right. If Run Now does not produce the result you want, the scheduled run will not either.

Every field in the scheduler, top to bottomSeven fields, and the ones people skip are the ones that keep a task working weeks later. Only the Working folder decides whether output lands in a file.

NameJust the label you see in the sidebar. It does not decide where output goes.
DescriptionYour own note for what the task does and when you last touched it. Claude never reads it.
PromptThe plain-language instructions Claude actually receives and runs.
FrequencyHow often it repeats: hourly, daily, weekdays, weekly, or manually.
TimeThe clock moment it fires. Prefills 9:00 AM, so set it on purpose. Weekly adds a day picker.
ModelWhich Claude model runs the task. The default is fine for most things.
Working folderThe only field that saves output to a file. Leave it blank and the result stays in the sidebar.

When it fires has two parts, frequency and a Time that prefills to 9:00 AM, so set the time on purpose. Fill the Description, point the Working folder where output should land, and let Run Now prove the task before you trust the clock.

Try it (15 min)

Watch out for

Using a generic name like 'Daily task' when you have multiple tasks. You will not be able to tell them apart in the sidebar when you are diagnosing a problem at 9 AM. Names like 'Morning briefing,' 'Friday project status,' and 'Weekly competitor check' are immediately identifiable.
Leaving the working folder blank when your task is supposed to write output to a file. Claude cannot write to disk without a working folder set. If you want a file, configure the folder. If you do not want a file, the sidebar session is fine.
Skipping Run Now and saving directly to a schedule. Your first scheduled run is not the moment to discover that an app connector is switched off or that the output format is wrong. Test with Run Now first, every time.

Paste this into Claude

I want to build my first scheduled task and set every field correctly. Help me plan it, and ask me one question at a time with clickable multiple-choice options so I am not staring at a wall of text. (The feature that shows those clickable options is called AskUserQuestion. Use it.)

Ask me these, one at a time:
1. What is the task? (What should Claude do each time it runs?)
2. How often should it run (hourly, daily, weekdays, weekly, or manual)?
3. What clock time should it fire, and if weekly, which day? (The Time field prefills 9:00 AM, which may not suit this task.)
4. Do I want Claude to write output to a file on my disk, or is the sidebar session enough?

After I answer, give me:
- A filled-in version of every field for my task (name, description, prompt, frequency, time, day if weekly, model recommendation, working folder recommendation)
- One specific Run Now test I should do before saving the schedule, and what a good result looks like
- One thing in my prompt that I should watch for on the first live run

What a good response looks like

Here is every field for your morning briefing task:

**Name:** Morning briefing (Gmail plus Calendar)
**Description:** Daily synthesis of overnight emails and today's calendar. Prompt updated 2026-05-01.
**Prompt:** It is [current day], [current date] at [current time]. Summarize my Gmail inbox for emails received in the last 14 hours that require a reply or decision. Then list my calendar events for today. Format: first an 'Emails needing action' section with up to 5 bullets (sender, what they need, urgency: today or this week), then a 'Today's schedule' section with event times and titles. If inbox is clear, write 'Inbox clear.' If calendar is empty, write 'No events today.' Total output: under 300 words.
**Frequency:** Weekdays
**Time:** 7:40 AM, set on purpose rather than left on the prefilled 9:00 AM. I read it before my 8:00 standup, and the few-minutes server delay still lands it in time.
**Model:** Default
**Working folder:** Not needed; sidebar output is sufficient.

**Run Now test:** Click Run Now right now and check that: (1) the email section shows real emails from your inbox, not placeholder text, (2) the calendar section shows today's actual events, (3) the output is under 300 words. If the email section comes back empty but you know you have recent emails, the Gmail connector is not switched on. Do not go hunting through settings files. Tell Claude: "Connect my Gmail in the Cowork desktop app and confirm it is enabled." Then run the test again.

**Watch for on first live run:** The time context. Check whether 'last 14 hours' captured the right emails on the first automated run. If a catch-up run fired late, it may have reflected a different window than expected.

What good looks like

You have set every field for your task, not only the three most people stop at
Your task name describes the output well enough to identify it in a sidebar with five other tasks
Your description includes when the prompt was last updated
You set the run time on purpose rather than accepting the time the picker prefilled for you
You have run Run Now at least once and confirmed the output matches what you expected before saving the schedule

When this breaks

Breaks when a connector looks linked in the Cowork web interface but is not actually enabled in the desktop app that runs your task. The task runs, Claude acts as if all is well, and the output comes back empty for every app that is not switched on. The fix is no-code: tell Claude 'Connect my Gmail in the Cowork desktop app and confirm it is enabled,' then re-run. Run Now surfaces this before your first live run.
Breaks when the description field is left blank and you have multiple tasks. When one starts producing wrong output, the blank description gives you no information about when the prompt was last reviewed or what the task was supposed to do. One sentence in the description saves diagnostic time.

AI can help with this

Create a tiny manual task in Claude Cowork first. Fill in task name, prompt, frequency, model if needed, and working folder if the task reads local files. Use Run Now before you trust any scheduled run.

The routine becomes a reliable scheduled run with explicit inputs, timing, fallback, and a reviewed output path marked by the golden dot.

You can now

You can complete the lesson outcome inside the correct Claude scheduling surface.

You have set every field for your task, not only the three most people stop at.
You can verify that your task name describes the output well enough to identify it in a sidebar with five other tasks.
You can verify that your description includes when the prompt was last updated.
You set the run time on purpose rather than accepting the time the picker prefilled for you.

Key takeaways

Name and description are for you, not Claude. The Working folder decides whether output lands in a file. The schedule is two parts, a frequency and a Time that prefills 9:00 AM, so set the time on purpose. Run Now proves a task works before you commit it to a schedule.

The help center names these create-task fields: Task name, Description, the Prompt, Frequency, Model (optional), and Working folder (optional). The fields people skip (Description, Working folder, a deliberate run time) are the ones that prevent silent failures three weeks later.
Name is only the sidebar label, and Description is your own note for diagnosis, never read by Claude during a run. Where files land is decided by the Working folder, not the Name. Both deserve more than one word.
Working folder is the single field that controls output: set it and the task writes a file there, leave it blank and the result stays in the Cowork sidebar with nothing saved to disk.
Frequency options, per the help center, are hourly, daily, weekdays, weekly, and manually. Manual never fires on its own, so it is the home for tasks you want on demand, not on autopilot.
When it fires has two parts: a frequency and a Time field that prefills 9:00 AM. Weekly adds a day-of-week picker. The prefill is a default, not a recommendation, so set the time on purpose. (Choosing it well is its own skill; Module 1 is built around it.)
Run Now fires the task immediately for testing. Use it before every schedule save. If Run Now does not produce the output you want, neither will the scheduled run.

Go deeper

Was this helpful?

Up nextTiming the run: cadence and clock together

The two access surfaces: a connector and a folder Quota math for multi-connector tasks

← Back to Claude Routines

I want to build my first scheduled task and set every field correctly. Help me plan it, and ask me one question at a time with clickable multiple-choice options so I am not staring at a wall of text. (The feature that shows those clickable options is called AskUserQuestion. Use it.) Ask me these, one at a time: 1. What is the task? (What should Claude do each time it runs?) 2. How often should it run (hourly, daily, weekdays, weekly, or manual)? 3. What clock time should it fire, and if weekly, which day? (The Time field prefills 9:00 AM, which may not suit this task.) 4. Do I want Claude to write output to a file on my disk, or is the sidebar session enough? After I answer, give me: - A filled-in version of every field for my task (name, description, prompt, frequency, time, day if weekly, model recommendation, working folder recommendation) - One specific Run Now test I should do before saving the schedule, and what a good result looks like - One thing in my prompt that I should watch for on the first live run

Here is every field for your morning briefing task: **Name:** Morning briefing (Gmail plus Calendar) **Description:** Daily synthesis of overnight emails and today's calendar. Prompt updated 2026-05-01. **Prompt:** It is [current day], [current date] at [current time]. Summarize my Gmail inbox for emails received in the last 14 hours that require a reply or decision. Then list my calendar events for today. Format: first an 'Emails needing action' section with up to 5 bullets (sender, what they need, urgency: today or this week), then a 'Today's schedule' section with event times and titles. If inbox is clear, write 'Inbox clear.' If calendar is empty, write 'No events today.' Total output: under 300 words. **Frequency:** Weekdays **Time:** 7:40 AM, set on purpose rather than left on the prefilled 9:00 AM. I read it before my 8:00 standup, and the few-minutes server delay still lands it in time. **Model:** Default **Working folder:** Not needed; sidebar output is sufficient. **Run Now test:** Click Run Now right now and check that: (1) the email section shows real emails from your inbox, not placeholder text, (2) the calendar section shows today's actual events, (3) the output is under 300 words. If the email section comes back empty but you know you have recent emails, the Gmail connector is not switched on. Do not go hunting through settings files. Tell Claude: "Connect my Gmail in the Cowork desktop app and confirm it is enabled." Then run the test again. **Watch for on first live run:** The time context. Check whether 'last 14 hours' captured the right emails on the first automated run. If a catch-up run fired late, it may have reflected a different window than expected.