Scheduled InstructionsA Template for AI BUILD Builders
A fill-in-the-blank worksheet for building your own Standing Instructions and Scheduled Tasks in Cowork — including two full working templates, a Daily Briefing and a Desktop Organisation, built from the exact routines Alice runs every week.
Why This Exists
Cowork gets better the more clearly you tell it who you are and what your week actually looks like. Most business owners never write that down — they just answer the same context questions every session. This template exists so you only have to do it once.
Every time I open a Cowork session, Claude already knows who I am, what Instant Insight Academy is for, where my hard boundaries sit, and how I like work handed back to me. I didn't teach it that in the moment — I wrote it once, as a set of Standing Instructions — what Claude calls your Instructions — and every session reads them before we begin.
On top of that sits a second layer: Scheduled Tasks — specific jobs I've written once and set to run on a rhythm, so the work happens whether or not I remember to ask for it.
This worksheet walks you through building both, for your own business or practice. Fill in the brackets, drop the finished text into Cowork, and you have an AI team member who's been briefed properly from session one — not one you have to re-explain yourself to every Tuesday.
This is a template, not a script. The sections mirror the shape of my own instructions because that shape works — but the content is entirely yours. Where something doesn't apply to how you work, cut it. Where something's missing, add it. The only wrong answer is leaving it generic.
If you've completed Instant Insight Discovery, start from the Claude Global Instructions it produced — most of Part One is already written.
If you've already worked through Steps 2–4 of the AI Team Setup Guide, your instructions are in place. Use Part One as a checklist for anything you left out, such as your hard boundaries and your stack, then go straight to Part Two.
Two Documents, Two Jobs
Standing Instructions and Scheduled Tasks solve different problems. Confusing them is the most common reason a Cowork set-up feels clunky.
- 🧭Who you are, what your business or practice is for, your boundaries, your stack, your working style.
- ♾️Read every session and every scheduled task, automatically — you never re-paste it.
- 🎯Makes every answer sound like you and fit your context, without you explaining yourself.
- 📝Written once. Revisited every few months as your business evolves.
- 🔁One specific job, run on a cadence — daily, weekly, monthly.
- ⚙️Built on top of your Standing Instructions — it already knows the context, so the task prompt itself can be short.
- 📤Ends in something concrete — a brief, a draft, a file, a reminder.
- 🧪Tested manually a few times before you ever schedule it. Runs in the cloud, even with your laptop closed — unless it needs files on your computer.
Part One of this worksheet builds your Standing Instructions. Part Two builds your first two or three Scheduled Tasks on top of them. Do them in that order — a scheduled task built on vague standing instructions will produce vague, generic output no matter how well the task itself is worded.
Your Standing Instructions
Six sections, the same shape I use for Instant Insight Academy. Fill in the brackets in your own words — short and true beats polished and vague.
Who I Am
Your name, your role, your years in the field, what you're known for, what clients come to you for.
I'm [your name], [your role or title], [years] years in [your field — e.g. professional services, coaching, consulting, facilitation…]. I work with [who you serve]. People come to me for [the result you deliver — be specific], not [what they don't come for]. My tone is [3–4 words that actually sound like you].
What This Cowork Session Is For
The job Cowork does in your business — execution, not thinking. What kinds of tasks you'll bring here.
This is my execution engine for [your business or practice name]. You help me [e.g. prepare client materials, organise files, draft SOPs, repurpose content, run task sequences]. Default to action — prepare the thing, then show me — rather than asking me to specify every step. When a task is ambiguous, make the sensible call, do it, and flag the assumption.
The Goal Every Task Serves
Your revenue picture and current focus — the thing every task should quietly move forward.
The goal: [your revenue target / growth goal, and by when]. Right now the priority is [what you're building or filling this quarter]. Everything else is foundation that should cost near-zero live hours.
My Hard Boundaries
The lines Cowork should never quietly cross — your protected hours, your protected days, anything irreversible that needs your sign-off first.
Protect my live hours — [your cap, e.g. "12 live hrs/week"]. [Your protected days/times, e.g. "No Fridays. Mornings are client delivery only."]. Confirm before irreversible or sharing actions — sending, publishing, granting access, deleting. Prepare and show me; I'll approve.
My Stack
The tools you actually work in, and what each one is for — so Claude reaches for the right one instead of guessing.
[Tool 1, e.g. Notion] — [what it's the source of truth for] [Tool 2, e.g. Google Calendar / Google Tasks] — [what it's for] [Tool 3, e.g. your CRM or accounting tool] — [what it's for] [Tool 4, e.g. Canva] — [what it's for]
How To Execute
How you like work handed back — sequencing, tone, the kind of thinking partner you want Claude to be.
Default to doing, then showing. Hand work back as a draft rather than a question whenever a draft is possible. Be a sparring partner, not a cheerleader — flag anything that scatters my energy or busts my time boundary. Keep my voice in all client-facing output: [2–3 words describing your voice, e.g. "warm, direct, no hype"].
Instructions sit at two levels. Global — who you are and how you work — apply to every conversation and task: paste them into Settings → Account → Instructions for Claude. Project instructions apply only inside one Project, on top of your global ones: open the Project and add them under Instructions in its right-hand panel. Either way, they're read at the start of every session and every scheduled task — you never need to repeat yourself.
Your Daily Briefing
This is built directly from the scheduled task Alice runs every weekday morning — her own AI chief of staff, scanning inbox, notes, source of truth and calendar, then handing back one clean briefing. Fill in your own brackets and it's yours.
It's also the canonical first build in AI BUILD, because it does one thing brilliantly: it puts your world in front of you at 8am, already sorted, without you having to go looking for it. Set it up once, and every weekday it does the looking for you.
It's long — six steps and an exact output format — because that precision is what makes it reliable. Loose instructions produce loose briefings. Copy the whole thing into a new Scheduled Task, fill in every bracket, then run it manually once or twice before you trust it with your mornings.
Who You Are, In One Paragraph
This is what the AI reads first, every single run. It's the difference between a generic summary and a briefing that actually understands your business.
You are [your name]'s AI chief of staff. [Your name] is a [your role — e.g. "Founder & Managing Director" or "Executive Coach"] running [your business name(s)]. [Your name] delivers client work [your protected delivery days, e.g. "Tuesday–Thursday only"], runs a [your weekly live-hours cap, e.g. "12-hour"] live delivery cap per week, and their [number] flagship offers are: [Offer 1 — name and one-line description], [Offer 2], [Offer 3], and [Offer 4]. The goal is [your revenue/profit/team goal, e.g. "$[your target]/year at ~[your target]% profit"]. [Your source-of-truth tool, e.g. "Notion"] is the source of truth. Your job each morning: scan [the tools you actually use — inbox, notes app, source of truth, calendar], then deliver a clean, warm, actionable daily briefing in [your name]'s voice — [3–4 words that sound like you, e.g. "wise, warm, a little witty, never fluffy"].
Apply This To Every Drafted Reply
The rules Alice uses are hers — short punchy paragraphs, Seth Godin's structure, warmth underneath. Keep the shape if it's useful; rewrite the specifics so a reply actually sounds like you.
Write in [your paragraph style, e.g. "short, punchy paragraphs, one idea per paragraph, 3–4 sentences max"]. Structure: [e.g. "lead with the sharp observation, let the insight land, then move"], but keep the warmth underneath — the reader should feel seen and championed, not lectured. Language: [e.g. "UK English"]. [Any hard no's — e.g. "No exclamation marks. No hollow affirmations."]. Never fluffy or performative. End with "[your sign-off, e.g. "Warmly, [Name]."]" Use the person's first name once, at most. Read the actual email carefully — use the real numbers, real details, real situation. Quote figures exactly as written; never paraphrase loosely or invent them.
Scan The Inbox
Name your email connector and the searches that surface what actually matters — client, team and lead signals, not newsletters.
Use the [Gmail / your inbox connector] MCP (server: [your connector ID — find it under Settings → Connectors, or omit if it loads automatically]) to search for recent unread or important messages. Run these searches: - search_threads with query "is:unread newer_than:1d" - search_threads with query "is:important newer_than:2d" - search_threads with query "is:unread label:inbox" Identify messages that are: from clients, team members or leads, contain booking/payment/order signals, require a reply within 24–48 hours, or mention [your offer names]. Flag these with a brief summary and suggested next action.
Draft The Replies
This is where your Email Voice above gets put to work — one draft per flagged email, ready to copy and send, not a summary of what you might say.
For every flagged email that requires a reply, fetch the full thread first — read the complete message before drafting. Then draft the response in your Email Voice. Include the draft in the briefing under "DRAFTED REPLIES" with: - To: [name + email] - Subject: Re: [original subject] - Full draft body, ready to copy and send - Use exact figures, names, and details from the actual email — never approximate
Sweep Your Notes For Loose Tasks
Wherever you actually capture stray to-dos — Apple Notes, a notebook app, a running doc.
Use the [your notes app connector] MCP to list all notes. For any note with a title suggesting tasks, ideas, or follow-ups (keywords: [your trigger words, e.g. "todo, task, action, follow up, call, email, prepare, draft, create, meeting, client"]), fetch the full content. Skip any note whose title starts with "[your briefing note's own title prefix, so it never reads itself]". Extract any actionable items and add to the task list.
Check Your Source Of Truth
Notion, or wherever your unresolved tasks and forward plan actually live. Point it at the right horizon so old, closed work stays out of the briefing.
Use the [Notion / your source-of-truth] MCP to search for [your current planning horizon, e.g. "FY27"] tasks and actions: - search with query "[horizon] Tasks" - search with query "[horizon] action" - search with query "[horizon] to do" - search with query "[horizon] prepare" Extract any unresolved items and add to the task list. Surface only open [current horizon] items; leave completed or archived [prior horizon] tasks out.
Read The Week Ahead
Which calendars matter, how far ahead to look, and which days are protected and deserve a warning flag if something's crept in.
Use the Calendar MCP to list calendars, then list events for today and the next [how many days, e.g. "5"] days across all calendars ([your calendar names]). Flag any client meetings, team sessions, sales calls, workshops or delivery sessions. For each one in the next 48 hours, note prep needed. Flag any events on [your protected day(s)/time(s), e.g. "Monday or Friday afternoon"] with ⚠️.
Compile It — And Save It Somewhere You'll See It
The exact output shape matters here — a consistent structure is what makes a briefing scannable at 8am rather than another thing to read properly. Keep this shape; just make the greeting and the sign-off yours.
Save the briefing to [where you'll actually see it — Apple Notes, a Notion page, a doc] with: - name: "Daily Briefing — [Day DD Month YYYY]" - content formatted exactly as below: <h1>☀️ Good morning, [your name].</h1><p>Here's your world at 8 AM — [Day, Date].</p><br> <h2>📬 INBOX — What needs your eyes</h2> <p>[For each flagged email: 🔴 or 🟡 emoji + <b>Sender name</b> — subject line<br> One-line summary of what they said and what's needed.]</p> <p>[If nothing urgent, one warm sentence saying so.]</p><br> <h2>📧 DRAFTED REPLIES</h2> [For each flagged email that needs a reply:] <p><b>To: [Name (email@address.com)]<br>Subject: Re: [Original Subject]</b></p> <p>[Draft body in your Email Voice.]</p> <p>[your sign-off]</p><br> [If no replies needed, omit this section entirely.] <h2>✅ TODAY'S ACTION LIST</h2> <h3>🔥 Urgent/Today</h3><p>[Each task on its own line using ☐.]</p><br> <h3>📅 This Week</h3><p>[Each task on its own line using ☐.]</p><br> <h3>💡 When-I-Can</h3><p>[Each task on its own line using ☐.]</p><br> <h2>📅 COMING UP — Meetings and sessions to prepare for</h2> [For each meeting or session:] <p><b>[Name — Day Date, Time]</b><br>☐ [prep action]<br>☐ [prep action]</p> [Blank line between entries. Flag boundary breaches with ⚠️.]<br> <h2>🎯 [your name]'s focus hat for today</h2> <p>[One grounded sentence on which offer this day serves: [your offer names].]</p>
Confirm
A short line so you know it ran cleanly, wherever the scheduled task posts back to.
After saving, post the briefing here and confirm: "✅ Also saved to [where you saved it] as '[note title]'." Tone throughout: warm, certain, zero fluff — a trusted chief of staff who knows the business deeply. If any connector returns an error, skip that section gracefully.
Click Scheduled in the left sidebar, then New task — choose Create with Claude to have Claude draft it with you, or Set up manually. (You can also type /schedule in any task, or simply describe the task and its timing in a conversation.) Beyond the prompt, you'll set a task name, the approval mode, how often it runs (hourly, daily, weekly, weekdays or manually — weekdays is the natural fit here), and optionally the model and a folder. To give it your project's context, set it up from a conversation inside that Project. A briefing that only uses connectors runs in the cloud at 8am whether your laptop is open or not. If you keep the Apple Notes step, that connector lives on your Mac, so the task runs there — the desktop app needs to be open, and a missed run catches up when you're back. Reading and drafting are safe to let it do; sending stays with you.
Your Desktop Organisation
The second scheduled task worth borrowing outright — built from the one Alice runs every Friday morning to tidy her desktop, back it up to Drive, and quietly grow her IP library without her lifting a finger.
This one earns its keep in the background. It leaves client work and your voice alone — it just keeps the operational side of your business tidy: files where they belong, a backup copy sitting safely in the cloud, and every SOP, framework and piece of client material you've ever built logged somewhere findable. That last part matters more than it sounds — a catalogued library of your own material is the beginning of a team manual, a future product, or simply keeping three years of good work out of a folder called "misc".
Three steps: tidy the desktop, back the new material up to cloud storage, catalogue anything that's genuinely business IP. Nothing gets deleted without you saying yes.
Desktop still in chaos? Do the one-off big sort first (task 1 in the AI Team Setup Guide), then let this task keep it that way each week.
Organise The Connected Desktop Folder
Your Desktop (or working folder) is connected to Claude directly — no Finder or screen access needed. Claude sees only the folders you connect; it works the files with file tools and bash.
Desktop path (file tools): [your Desktop path, e.g. /Users/[you]/Desktop] Note: on a Mac, iCloud sync sometimes nests the real synced desktop inside a second "Desktop" subfolder. Check both [/Users/[you]/Desktop] and [/Users/[you]/Desktop/Desktop] and tidy whichever holds the loose files. Ensure all documentation is in the correct folder. Identify duplicates (compare by name AND by byte-size/content, not name alone) and redundant loose files. Move anything questionable or duplicated into a folder called "[Your Name] to Review" — do NOT delete. ALWAYS ask before putting any document in the trash. Trash something only with [your name]'s explicit yes. If the connected Desktop folder isn't reachable on a given run, fall back to the connected [your fallback folder, e.g. Downloads] folder and say so clearly in the report.
Back It Up To Cloud Storage
Point it at your own Drive (or equivalent) folder — and be explicit that it's a top-up, not a re-upload of everything every week.
After organising, upload user-created folders and files from the Desktop to [your cloud storage folder name, e.g. "Business Backup 2027"]: - Folder ID: [your folder ID] - URL: [your folder URL] To upload: use the Browser tool — navigate to the folder URL, then use its upload option for each Desktop folder. If the browser tool isn't connected, list clearly what needs uploading so you can do it yourself in one pass. Large or binary files (DOCX, PDF, PNG, ZIP, large HTML) may still need a manual drag — flag those explicitly rather than failing silently. Leave out any Screenshots folder and all system/application files. Only user-created folders and documents. Your Drive already holds a full mirror — push only genuinely new items rather than a blanket re-upload, so you don't breed duplicates. After uploading, note the name and link for each item successfully uploaded.
Catalogue The IP
Wherever your SOPs, frameworks and client material live as a searchable library — this is what turns years of good work into a findable asset rather than a folder graveyard.
For any business resources newly uploaded (SOPs, workflows, frameworks, templates, client deliverables, training and onboarding material), create a new entry in your [your library database name, e.g. "IP Library"] database: - Database URL: [your database URL] For each new resource, create a page with: - Asset / Folder = the folder or file name - Link = the direct link to that specific file or folder - Category = best match from: [your own categories — e.g. the service lines, SOP areas and content types your business actually produces], Not Yet Catalogued - Status = "Catalogued" - Description = a one-line description of what the resource contains (infer from the folder/file name) Only create entries for business IP — skip personal folders, screenshots, invoices, and admin files.
The Report
One short wrap-up so you can see at a glance what happened, without opening every folder yourself.
At the end, give a short report: what was tidied, what moved to "[your review folder name]", what uploaded to cloud storage, what was catalogued, and anything still needing a manual hand.
A weekly cadence suits this one — Alice runs hers every Friday, ~9:30, as the week winds down. Choose the folder it should tidy when you set the task up. Because it works on files on your computer, this task runs on your computer, not in the cloud: keep the Claude desktop app open at that time. If your laptop is asleep or closed, it runs as soon as you're back and tells you it was delayed. Choose an approval mode that lets it create files and Notion pages without stopping each time — Claude still asks before deleting anything, whatever mode you choose. That decision stays yours, every time.
Your Scheduled Task Builder
One worksheet per task. Fill this in, run it manually two or three times to check the output, then schedule it. Start with one task — five can wait.
Fill This In For Each Scheduled Task
TASK NAME: [a name you'll recognise in six months] WHICH HAT: [winning and serving clients — urgent, revenue-facing] [or: building the foundation — high-value, less urgent] CADENCE: [hourly / daily / weekdays / weekly / manually], [day], [time] WHERE IT RUNS: [cloud — connectors only] or [my computer — needs local files, apps or a local connector] WHERE TO LOOK: [the folder, calendar, or tool Claude should read from] WHAT TO DO: [the specific job — be concrete] HOW IT SHOULD FEEL: [a line about tone — this is usually the piece people forget] WHERE TO PUT IT: [save as a file / draft in Gmail / just show me — always held for my approval before sending] APPROVAL MODE: the most cautious mode for the first 3 runs, then reassess.
That's the same four-part shape as any good Cowork prompt: where to look → what to do → how it should feel → where to put it. If a scheduled task ever produces something flat or generic, it's almost always the "how it should feel" line that's missing or too thin.
And the sentence worth keeping in every task that touches a client: "Save it as a draft for me to review." Use it until you've watched the task run cleanly a few times.
A Starter Rhythm for a Business or Practice
Not a prescription — a set of starting points. Pick one, run it manually, adjust, then schedule it. Add the next once the first one is running on its own.
| When | What it could do |
|---|---|
| Before each client meeting or session | Pull who's attending, what was covered last time, and what materials need to be ready — a short prep brief, not a re-read of your whole notebook. |
| After each client meeting or session | Draft the follow-up or thank-you note in your voice, and flag anything worth logging for next time. |
| Weekly | A hygiene pass on your SOPs, client materials and shared folders — flag anything stale, duplicated, or inconsistently named. Leave everything as it is; just list it. |
| Weekly or fortnightly | Take one thing you said, decided or delivered this week and reshape it into two or three pieces for wherever you show up publicly. |
| Monthly | A short look-back: what ran well, what's carrying forward, one thing to deprioritise next month to protect your live-hour cap. |
A philosophy, borrowedScheduled tasks aren't about doing more. They're about freeing you from remembering to do things — so the rhythm belongs to the business, not to your memory of Tuesday mornings.
Setting It Up
Same four steps, every time. Keep the "run it manually first" step — it's the one that saves you from scheduling something half-baked.
- 1
Paste your Standing Instructions in first
Global: Settings → Account → Instructions for Claude. Project: the Project's right-hand panel, under Instructions. Everything downstream reads better once this is in place.
- 2
Run one task worksheet manually, a few times
Paste the "what to do" prompt straight into a normal task. Refine the wording until the output actually sounds like you — then type /schedule in that same task to turn the exact process into a recurring one.
- 3
Open Scheduled and set the cadence
Or go the long way: Scheduled in the left sidebar → New task → Set up manually. Paste the refined prompt, pick the cadence, approval mode and folder from your worksheet. From the same page you can pause, edit, delete or run any task on demand.
- 4
Watch the first three runs — then live with it for fourteen days
Adjust the prompt if anything drifts. Once three runs land well, let it carry fourteen days of ordinary work. A task that survives that is built — write it up as an SOP in your Build Journal.
Before You Ship It Live
A short check before any Standing Instructions or Scheduled Task goes live.
- ✓Does it actually sound like you? Read the "How To Execute" and "Who I Am" sections out loud. If they'd fit anyone, they're not finished.
- ✓Are your boundaries specific, not aspirational? "Protect my time" is a wish. "No meetings before 10am, Fridays protected" is a boundary.
- ✓Does every client-facing task say "save as draft for review"? Until you've watched it run cleanly several times, keep a human in the loop.
- ✓Have you started with one task? A single scheduled task done well teaches you more about how to write the next one than five written at once.
- ✓Do you know where it runs? Connector-only tasks run in the cloud on their own. Anything touching files or apps on your computer needs the desktop app open at its scheduled time.
- ✓Is anything confidential clearly marked as off-limits? Say so explicitly in your Standing Instructions, and match it to your Team AI Code of Conduct or Client AI Usage Statement — Cowork respects a clear boundary.
Write it once, in your own words. Let the rhythm do the remembering.