Your work, done with you.
Cowork is the agentic part of Claude. It lives in Claude Desktop, reads your files, connects to your apps, and runs multi-step work on your behalf. This practice turns seven real-life workflows into reusable assets, plus three setup prompts and twenty copy-paste recipes — covering personal admin, decisions, and the specific roles you actually play. All running on your week by Monday.
- Set up Cowork with the three connectors that match your stack
- Save and rerun three repeatable skills (meeting prep, weekly review, contract review)
- Run 23 paste-and-run prompts — three setup, twenty recipes — spanning personal admin, decisions, and specific roles (founder, PM, sales, EM, creator, manager)
- Apply the reversibility discipline so Cowork drafts but never sends
- Choose between Cowork and Claude Code by the shape of the work
What Cowork is.
Cowork = model + your filesystem + your connectors. It reads from your apps (Google Drive, Notion, Slack, Calendar, Gmail, GitHub, Linear) and takes multi-step action.
Three things that separate Cowork from chat:
- It reads your data. Connect Google Drive once; from then on, Cowork can read any document you point it at without you copy-pasting.
- It runs multi-step. “Find the renewal contract, extract the auto-renew clause, draft an email to legal” is one prompt, not three.
- It is reversible by default. Drafts before sends. Reads before writes. Preview before execute.
Setup & connectors.
Install Claude Desktop. Authenticate the connectors you actually need. Skip the rest.
- Download Claude Desktop from claude.ai/download.
- Sign in with the same account as the web app.
- Open Settings → Connectors. Authenticate the ones that match your stack.
- Start with three: Google Calendar, Gmail or Outlook, and either Notion or Google Drive. That covers ~80% of the use cases below.
Recommended connector trio by role
| If you are | Add these three first |
|---|---|
| Manager / PM | Calendar · Notion · Slack |
| Founder / consultant | Calendar · Gmail · Drive |
| Analyst | Drive · Notion · GitHub |
| Sales / CS | Calendar · Salesforce/HubSpot · Gmail |
| Engineering manager | Linear/Jira · GitHub · Slack |
The connector triage.
Before you authenticate every connector Cowork offers, this prompt picks the three that actually unlock the most for your specific stack and role — so the first hour is the highest-leverage hour.
Setup: which connectors do I authenticate first?
I'm setting up Claude Cowork for the first time. Help me decide which connectors to authenticate first so I don't waste the first hour turning everything on. MY ROLE: [e.g. Founder of a 3-person consultancy / Product manager at a B2B SaaS / Solo creator] MY STACK (list everything I use weekly): - Email: [Gmail / Outlook / other] - Calendar: [Google / Outlook / other] - Documents: [Google Drive / OneDrive / Dropbox / Notion / other] - Tasks: [Notion / Linear / Jira / Asana / Todoist / other] - Comms: [Slack / Teams / Discord / none] - Code: [GitHub / GitLab / none] - CRM: [HubSpot / Salesforce / Pipedrive / none] - Finance: [QuickBooks / Xero / Zoho / none] MY TYPICAL WORKWEEK: [3-5 sentences on what your week looks like — meetings, writing, deciding, shipping, etc.] THE RECIPES I WANT TO RUN FIRST: [Pick 3 from the recipe library, e.g. "meeting prep, solo retro, inbox reset"] GIVE ME: ## 1. Top 3 connectors to authenticate today (~15 min) - Connector name - Which recipes it unlocks - Why this one before others ## 2. Tier 2 connectors for this week (~15 min more) - Connector name - Trigger: which recipe makes me reach for it ## 3. Skip for now - Connectors I shouldn't bother with yet - Criteria: would I use it more than once a week? ## 4. Scope per connector - Read-only or read-write? - Default recommendation: read-only until I've run the workflow 10 times and trust it. ## 5. The 30-day check-in question - One question to ask myself in 30 days to decide which connectors to add or drop. REVERSIBILITY: Recommend only. Never authenticate on my behalf. I click through OAuth.
Pro tip. The 30-day check-in is the most valuable part. Most people over-authenticate on day one and never re-evaluate. The list of connectors you’re actively using in month two is the real list.
The first-touch test.
Before trusting Cowork with anything that matters, run this safe smoke test. It verifies every connector you authenticated, performs one small read-only action per source, and confirms the reversibility contract.
Setup: verify Cowork is wired up and safe
This is my first real test of Cowork. Do something useful but small so I can see how it works without risk. TEST PROCEDURE: ## 1. Verify the connectors I authenticated For each connector I have on, do ONE read-only action and confirm success: - Calendar → "I can see N events in the next 7 days." - Gmail → "I can see N unread messages in your inbox." - Drive → "I can see N files modified this week in your /Documents folder." - Notion → "I can see your N most recent pages." - Slack → "I can see the last 5 channels you read." - GitHub → "I can see your last N commits." - (and any others I've authenticated) If any fails, tell me which connector to re-authenticate. Do not try to fix it yourself. ## 2. Read one Downloads file Pick the most recently modified file in my Downloads folder. - Tell me its name and what it appears to be. - Summarize the content in 3 sentences. - Suggest which folder it should probably live in. - DO NOT MOVE IT. ## 3. Preview my next 3 calendar events - List date, time, title, attendees for each. - For each: do I have a related document or recent email thread for this person? - DO NOT modify the events. ## 4. Confirm the reversibility rule Read this back to me word-for-word so we both agree: "I am running read-only by default. I do not send, post, publish, delete, or move anything unless you explicitly ask. I show you a draft before any action. If I'm not sure whether an action is reversible, I stop and ask." If you can do steps 1-4 without errors, you are cleared to run real recipes. REVERSIBILITY: Read-only. No moves, sends, deletes, or modifications anywhere.
Pro tip. Save this as a starred chat named “Cowork smoke test”. When something seems off in another chat, re-run this one first to isolate whether it’s a connector issue or a prompt issue.
The skill mental model.
A skill is a named, repeatable workflow. The way to think about Cowork is “a small set of personal skills,” not “a chat I have over and over.”
Three properties every good Cowork skill has:
- One job. “Prep me for tomorrow’s meetings” is a skill. “Help me work” is not.
- Specific inputs. A calendar event, a folder, a file, a Slack channel — not “think about my company.”
- A definite output. A one-page brief. A bulleted action list. A draft email. Something with a shape.
The use cases that follow are all built to this template. After the practice, the third column of each — the output shape — is what you save.
Meeting prep.
Walk into your next meeting knowing what was last discussed, what is still open, what to ask, and the one outcome that would make the meeting worth it.
Skill: prep my next meeting
2. Identify the person: name, role, company, when we last spoke.
3. Pull every past note and email that mentions them or their company.
4. Extract: decisions made, commitments on both sides, open threads, anything they said that hinted at deeper concern.
5. Draft a one-page brief in the exact format below.
Last time: date + what we decided
Open threads: unresolved, who owes what, by when
What they care about: stated goals, fears, KPIs
Talking points: 3 worth surfacing
Questions to ask: 5 sharp, tied to open threads
Objections to expect: 2 likely pushbacks + how I’d answer
The ask: the one outcome from this meeting
Next step: what good looks like after the call
Prep me for my next meeting. Build a one-page brief I can read in two minutes before I walk in. THE MEETING: [Default: my next calendar event with a non-team attendee — or name it here] SOURCES TO READ: - Calendar (the event + attendees) - Gmail (the last thread with the attendees) - Notion / Granola / Drive (past meeting notes that mention them or their company) - CRM history if connected DO THIS: 1. Identify the person: name, role, company, when we last spoke. 2. Pull every past note and email that mentions them or their company. 3. Extract decisions made, commitments on both sides, open threads, and anything that hinted at a deeper concern. 4. Produce the brief in the exact format below. PRODUCE A ONE-PAGE BRIEF: **Who**: name, role, company, how we know each other **Last time**: date + what we decided **Open threads**: unresolved, who owes what, by when **What they care about**: stated goals, fears, KPIs **Talking points**: 3 worth surfacing **Questions to ask**: 5 sharp, tied to open threads **Objections to expect**: 2 likely pushbacks + how I'd answer **The ask**: the one outcome from this meeting **Next step**: what good looks like after the call REVERSIBILITY: Draft only. Read-only on calendar, email, and notes. Never send, never modify the event or any contact record. MY CONTEXT: [My role, the relationship history, what I want out of this meeting]
Save this prompt as a saved chat in Cowork. Reuse before every external meeting. After 30 days you have prepped 30 conversations using the same shape — and you can compare what worked.
Weekly review.
Friday afternoon. Cowork reads the week out of your tools and produces the one-page status you would have written if you had two clear hours.
Skill: my Friday review
2. Cluster by theme — 3 to 5 themes max.
3. For each theme: what moved, what stalled, what I learned.
4. Surface anything I committed to without it appearing in any system (risk).
5. Produce the 5-section status note.
Moved forward: 3 things in flight.
Stuck: 1 to 3 with the actual blocker.
Promises I made: who I owe what, by when.
One thing for next week: the one decision or action.
It's Friday. Read my week out of my tools and write the one-page status note I'd write if I had two clear hours. WEEK: [Default: Monday of this week through now] SOURCES TO READ: - Notion / Drive (pages I wrote or edited this week) - Calendar (every meeting) - Gmail (sent items — what I committed to) - Linear / Jira (closed tickets) - GitHub (merged PRs) DO THIS: 1. Read everything I wrote, said, decided, or shipped between Monday and now. 2. Cluster it into 3-5 themes max. 3. For each theme, note what moved, what stalled, what I learned. 4. Surface anything I committed to that doesn't appear in any system (a risk). 5. Produce the 5-section note below. PRODUCE: **Shipped**: 3 specific things, one line on impact each. **Moved forward**: 3 things in flight. **Stuck**: 1-3 with the actual blocker. **Promises I made**: who I owe what, by when. **One thing for next week**: the single most important decision or action (exactly one). REVERSIBILITY: Read-only. Produce the note as a draft. Never post it anywhere — I paste it where it goes. MY CONTEXT: [My role, who reads this status, what I'm trying to make visible]
Contract review.
A vendor sends a Master Services Agreement. Cowork checks it against your checklist and produces a one-page review with the three things you actually need to push back on.
Skill: review this MSA
2. Compare against our template. Flag any deviation.
3. Run the red-flag checklist: indemnification cap, auto-renewal, payment terms, IP ownership, data residency, termination for convenience, audit rights, limitation of liability multiplier.
4. Quote the exact clause for each flag.
5. Suggest the three negotiation moves with highest leverage-to-relationship cost ratio.
Red flags found: table of clause | severity | what to negotiate.
Negotiation positions: 3 specific edits with rationale.
Out of scope: things to defer to legal.
Review the agreement I'm dragging into this chat. Produce a one-page review with the three things I should actually push back on. INPUTS: - The contract (PDF or Word, attached to this chat) - Our standard template, if I have one: [Drive link or "none"] - Agreement type: [MSA / SOW / NDA / Lease / Vendor terms / Other] DO THIS: 1. Read every clause. 2. If I gave you our template, compare against it and flag every deviation. 3. Run this red-flag checklist and quote the exact clause for each hit: indemnification cap, auto-renewal, payment terms, IP ownership, data residency, termination for convenience, audit rights, limitation-of-liability multiplier. 4. Pick the three negotiation moves with the highest leverage-to-relationship-cost ratio. PRODUCE: **Summary**: what kind of agreement, what the vendor is asking for. **Red flags found**: a table of Clause | Severity | What to negotiate (quote the exact clause text). **Negotiation positions**: 3 specific edits with rationale. **Out of scope**: things to defer to a lawyer. REVERSIBILITY: Produce a memo for me and legal. Never reply to the vendor, never sign, never mark the contract complete. Sending is mine. MY CONTEXT: [My company, the relationship with this vendor, what I can and can't walk away from]
Customer triage.
A backlog of customer messages. Cowork sorts them, drafts the easy replies, and surfaces the ones that need a human.
Skill: triage my inbox
2. Classify each: REPLY, FYI, ACTION, ESCALATE, SKIP.
3. For each REPLY where the FAQ has an answer, draft the response (do not send).
4. For ESCALATE, write a one-line summary of why.
5. Produce the queue.
Triage my customer messages. Sort them, draft the easy replies, and surface the ones that need me. SOURCES: - Inbox (Gmail / Outlook) — every unread or unassigned customer message from the last [24 hours] - Help desk (Zendesk / Intercom) if connected - My standard answers: [FAQ doc in Drive, or "none yet"] DO THIS: 1. Pull every unread or unassigned customer message in scope. 2. Classify each as exactly one of: REPLY, FYI, ACTION, ESCALATE, SKIP. 3. For each REPLY the FAQ can answer, draft the response (do not send), in my voice, under 5 sentences. 4. For each ESCALATE, write a one-line reason it needs a human. 5. Produce the queue. PRODUCE A RANKED QUEUE: - ESCALATE on top (with the one-line reason). - ACTION next, with deadlines. - REPLY drafts ready to review and send with one gesture. - SKIP reasoned but collapsed. REVERSIBILITY: Drafts only. Never send. Suggest labels but never archive, delete, or mark as read. MY CONTEXT: [My product, my tone, senders I must never keep waiting]
Sales prep.
A discovery call with a prospect tomorrow. Cowork builds the one-pager that turns “tell me about yourselves” into a conversation that earns the next meeting.
Skill: prep this prospect
2. Profile the company: industry, size, recent news (10-K if public, blog if not).
3. Identify the most likely pain points based on their stack and role.
4. Map our top three differentiators to those pains.
5. Draft 5 discovery questions and 3 stories from our customer base that would resonate.
Company: 3-line profile + 2 recent events.
Hypothesized pains: 3 with confidence.
Our hooks: 3 differentiators tied to those pains.
Discovery questions: 5 specific.
Customer stories to have ready: 3 short.
Prep me for a discovery call. Build the one-pager that turns "tell me about yourselves" into a conversation that earns the next meeting. THE PROSPECT: - Name + title: [NAME, TITLE] - Company: [COMPANY] - The call: [Default: my next calendar event with them] - What we sell: [ONE LINE on our product] - Our top 3 differentiators: [LIST, or "infer from our website"] SOURCES: - Their company website (fetch it) - Their LinkedIn / recent posts if public - Recent company news (10-K if public, blog if not) - CRM history if connected DO THIS: 1. Profile the person: title, tenure, recent posts or talks. 2. Profile the company: industry, size, 2 recent events. 3. Hypothesize their 3 most likely pain points from their stack and role. 4. Map our differentiators to those pains. 5. Draft 5 discovery questions and pick 3 customer stories that would resonate. PRODUCE: **Person**: 2-line profile. **Company**: 3-line profile + 2 recent events. **Hypothesized pains**: 3, each with a confidence level. **Our hooks**: 3 differentiators, each tied to a named pain. **Discovery questions**: 5 specific. **Customer stories to have ready**: 3 short. REVERSIBILITY: Read-only. Never log CRM activity, never message anyone from my account. MY CONTEXT: [My role, deal stage, what a good outcome from this call looks like]
Financial close.
Month-end. Cowork reads the bank export and the GL, reconciles, and writes the variance memo that explains the bigger swings.
Skill: monthly close memo
2. Reconcile against GL entries. Flag mismatches with magnitudes.
3. Compute month-over-month variance per category.
4. For variances over 20% or $5k, propose an explanation from the underlying transactions.
5. Produce the memo.
MoM movers: top 5 variances with proposed explanations.
Reconciliation flags: bank-to-GL mismatches to investigate.
Open questions for the accountant: 2 or 3.
Watchlist for next month: 2 items.
It's month-end. Read the bank export and the GL, reconcile them, and write the variance memo that explains the big swings. PERIOD: [Month Year] INPUTS (attached or pointed to): - Bank transactions (CSV export) - Accounting export (QBO / Xero / Zoho) - Prior month's memo, for comparison - The chart of accounts DO THIS: 1. Group bank transactions by category. 2. Reconcile against GL entries; flag every mismatch with its magnitude. 3. Compute month-over-month variance per category. 4. For any variance over 20% or $5k, propose an explanation from the underlying transactions. 5. Produce the memo. PRODUCE: **Snapshot**: revenue, COGS, opex, net. **MoM movers**: top 5 variances, each with a proposed explanation. **Reconciliation flags**: bank-to-GL mismatches to investigate. **Open questions for the accountant**: 2-3. **Watchlist for next month**: 2 items. REVERSIBILITY: Read-only on all financial data. Never post journal entries. Output a markdown memo I approve before circulating. MY CONTEXT: [Entity type, what the prior month looked like, anything unusual this month]
Content production.
You wrote rough notes. Cowork turns them into the LinkedIn post, the blog draft, and the email — in your voice, ready to ship.
Skill: turn this into content
2. Study the voice in the samples — sentence length, vocabulary, what you include and what you leave out.
3. Produce three drafts in three formats, in your voice:
— LinkedIn (250–400 words, one hook, one story, one ask).
— Blog (~700 words, with H2s).
— Email to your list (~300 words, with a subject worth opening).
4. For each, propose two variations on the opening line so you can pick.
Turn my rough notes into content — the LinkedIn post, the blog draft, and the email — in my voice, ready to ship. INPUTS: - My rough notes: [paste them here, or point at a file] - 3 samples of my past content (for voice): [paste or link] - Target audience (optional): [WHO THIS IS FOR] DO THIS: 1. Read the notes and extract the one claim worth defending. 2. Study the voice in my samples — sentence length, vocabulary, what I include and what I leave out. 3. Produce three drafts in three formats, in my voice: - LinkedIn (250-400 words: one hook, one story, one ask) - Blog (~700 words, with H2 subheads) - Email to my list (~300 words, with a subject worth opening) 4. For each draft, propose two variations on the opening line so I can pick. OUTPUT: The three drafts side by side. Each with its opening-line A/B. I pick, edit, ship. REVERSIBILITY: Drafts only. Never post to LinkedIn, never publish the blog, never send the email. My fingers do the send. MY CONTEXT: [My niche, the one belief that runs through my writing, words I never use]
Twenty recipes. Pick your three.
Day 04 is the paste-and-run library. You wired Cowork up on Day 01 (the connector triage and first-touch test); the me.md primer below grounds it in who you are. Then twenty recipes, split into two halves — ten for personal admin and decisions, ten built for the roles you actually play (founder, PM, sales, EM, creator, hiring, traveling). Don’t read all twenty top to bottom. Use the router below to grab the three that match your week.
Your 3-recipe path by role
Don’t read all twenty. Find the row closest to your job, run those three this week, then come back for more.
| If you are a… | Start with these three |
|---|---|
| Consultant | R6 · Hand-off doc · R2 · Topic brief · R17 · Negotiation prep |
| Product manager | R14 · Feedback themes · R5 · Week reset · R8 · Solo retro |
| Founder | R11 · Investor update · R1 · Investment memo · R7 · Receipt sort |
| Analyst | R2 · Topic brief · R1 · Investment memo · R4 · Smart buy |
| Sales / CS | R13 · Call write-up · R16 · Renewal risk · R5 · Week reset |
| Engineering manager | R15 · Standup digest · R20 · 1:1 prep · R8 · Solo retro |
| Creator | R19 · Newsletter draft · R2 · Topic brief · R8 · Solo retro |
| Everyone, week one | R9 · Inbox reset · R3 · Subscription sweep · R10 · Desktop reset |
How to use a recipe
- Pick one matching a task on your list today.
- Copy the prompt (button in the prompt block).
- Replace
[BRACKETS]with your specifics. - Paste into Claude Desktop (Cowork).
- Approve permissions as Cowork asks (read-only by default).
- Save the prompt as a starred chat if you’ll reuse it.
The me.md primer.
Every future Cowork chat starts with you re-explaining who you are, what you do, and how you write. This prompt builds a one-page me.md document Cowork can reference in every recipe — so the first paragraph is never wasted again.
Setup: build my reusable context document
me.md, it has all six headings, the “writing voice” section sounds like you (and names words you avoid), and nothing was written to Drive — you saved it yourself.Build me a personal context document I can save as me.md and reference in every future Cowork chat. INPUTS — I'll point you at: - My LinkedIn export or About page - My company's About / Mission page - My 3 most recent customer-facing emails (from Sent) - My last 2 weekly reviews or status notes (if any) - One project doc I'm genuinely proud of ALSO USE: - My calendar (last 30 days) as behavioral evidence - My Sent items (last 30 days) for voice PRODUCE me.md WITH THESE SECTIONS: ## Role - Title and what I actually do (not the job description) - Who pays me / who I report to - What I'm accountable for ## Recurring work - 3-5 projects or workstreams that are always on ## People I work with most - For each: relationship, what we work on together, communication style ## My writing voice - 4-6 sentences in my actual style (extracted from sent items) - Vocabulary I use, vocabulary I avoid - Length preference (short emails vs long, formal vs casual) ## Tools I trust - My current stack - Tools I'm migrating to or away from - Tools I will NOT use (and why) ## How I want Cowork to behave - Always draft, never send - Read-only on financial data - Always summarize before doing - Always ask before chained destructive actions - Anything else specific to me OUTPUT: A markdown document ready for me to save as me.md. REVERSIBILITY: Read-only on all sources. Drafts me.md for my review. Never saves to Drive directly — I copy the final markdown myself. MY CONTEXT: [Add anything else you want Cowork to know about me]
Pro tip. The “vocabulary I avoid” line is the secret weapon. Cowork will stop using words you hate (“synergy”, “leverage”, “at the end of the day”) in every future draft.
The investment memo.
Before any investment over $1,000, the 10 minutes Cowork spends on this can save thousands. Public stock, private company, real estate, crypto — same shape, same depth.
Recipe: research an investment before I commit
I'm considering an investment and need thorough due diligence. Analyze this opportunity. INVESTMENT: [COMPANY NAME / STOCK TICKER / ASSET] INVESTMENT TYPE: [Public stock / Private company / Real estate / Crypto / Other] AMOUNT CONSIDERING: [AMOUNT] TIME HORIZON: [Short-term < 1 year / Medium 1-5 years / Long-term 5+ years] --- CONDUCT THIS ANALYSIS: ## 1. OVERVIEW - What does this company/asset do? - Business model in one paragraph - Key products/services + target market ## 2. FINANCIAL ANALYSIS (Public companies) - Revenue and growth trends (last 5 years) - Profitability metrics (gross/net margins) - Balance sheet health (debt, cash position) - Cash flow analysis - Valuation metrics (P/E, P/S, PEG vs industry) (Private / other) - Funding history and last round terms - Revenue estimates if available - Burn rate if known ## 3. COMPETITIVE POSITION - Top 3 competitors - Market share + the moat (or lack of one) - Where they lose ## 4. MANAGEMENT & LEADERSHIP - Key executives and their track records - Insider ownership % - Recent insider buying/selling - Any red flags (lawsuits, sudden departures) ## 5. GROWTH CATALYSTS - Specific events that could move price up - Product launches, expansions, regulatory wins - Each with rough timeline ## 6. RISKS - Industry / company-specific / regulatory - For each: probability (H/M/L) and impact (H/M/L) ## 7. RECENT NEWS & SENTIMENT (last 90 days) - Significant news items - Analyst ratings and price targets - Notable social/forum sentiment ## 8. COMPARISON - vs 2-3 alternatives in the same space - Why this over competitors (or not) ## 9. VALUATION - Fair / over / undervalued - What needs to be true for the price to be justified - Bear / base / bull price targets ## 10. RECOMMENDATION - Buy / hold / avoid - If buy: at what price, with what position size - Three things to monitor monthly --- MY INVESTMENT STYLE: [Conservative / Moderate / Aggressive] MY PORTFOLIO CONTEXT: [What else I own, for diversification] SOURCING & RELIABILITY (do not skip): - Cite a source for every hard number (revenue, margin, price target, insider %). - Rate each source High / Medium / Low reliability (filing or earnings call = High; blog or forum = Low). - If you cannot verify a figure, say "unverified" — never invent a number to fill a section. - List what you could NOT confirm so I check it before I commit a dollar. Be direct. Tell me what you actually think, not just neutral analysis.
Pro tip. Run this once per quarter on positions you already hold, not just before you buy. The “monitor” section in your last memo is the diff worth checking.
Trust, then verify. A model can confidently state a wrong P/E or a price target that doesn’t exist. That’s why this recipe forces a reliability rating and a “couldn’t confirm” list — the same discipline as Recipe 02. Before you move real money, open the primary source (the 10-Q, the earnings call) for any number the decision turns on. Treat unsourced figures as a question, not a fact.
The topic brief.
Forty-seven Google tabs replaced by a structured report. Use it for competitor research, market analysis, learning a new domain — anything where you need to come up to speed fast.
Recipe: brief me on this topic
I need comprehensive research on a topic. Create a detailed report I can reference. TOPIC: [YOUR TOPIC] DEPTH: [Quick overview / Standard 3-5 pages / Deep dive 10+ pages] STRUCTURE THE REPORT AS: ## Executive Summary - 3-5 bullet findings - Bottom-line conclusion - Recommended action if applicable ## Background & Context - What this topic is about - Why it matters now - Key players / stakeholders ## Current State - What's happening today - Recent developments (last 12 months) - Key statistics + data ## Key Findings - Finding 1 [with evidence] - Finding 2 [with evidence] - Finding 3 [with evidence] - ... (as many as warranted) ## Different Perspectives - Perspective A: who believes this and why - Perspective B: who believes this and why - Areas of consensus - Areas of debate ## Implications - For [ME / MY INDUSTRY / MY DECISION] - Risks to consider - Opportunities ## Recommended Actions - What I should do based on this - Prioritized next steps - Timeline if applicable ## Sources - List every source used - Reliability rating per source (High/Med/Low + why) - Note any information gaps ## Questions for Further Research - What you couldn't find - What would require deeper investigation SPECIFIC ANGLES I CARE ABOUT: [Add the questions or angles you want covered] MY CONTEXT: [Why I'm researching this — helps tailor findings] Save the final report as a markdown document I can reference and share.
Pro tip. The “reliability rating” line is the single most important line. Without it, you can’t tell a TechCrunch rumor from a peer-reviewed paper.
The subscription sweep.
Most people discover $100–300/month in forgotten subscriptions on their first audit. Cowork finds them all and produces the cancel list — you do the clicks.
Recipe: audit and shrink my subscriptions
I want a complete audit of every recurring subscription I'm paying for — and a plan to cancel what I'm not using. SOURCES TO SCAN: - My Gmail inbox: search "subscription", "renewal", "auto-renew", "your invoice", "monthly billing", "thanks for your purchase" - 12 months of credit-card or bank statements (CSV attached) - Apple ID / Google account subscription pages (screenshots if shared) - 1Password / Bitwarden export if I provide it FOR EACH SUBSCRIPTION: - Service name + category (Streaming / SaaS / News / Cloud / Other) - Cost: monthly equivalent + annual total - Billing cadence (monthly / annual / lifetime) - Next renewal date - Last time I touched it (from emails or activity) - Cancellation URL or step-by-step instructions - "Likely forgotten?" flag if no activity in 60+ days PRODUCE: ## 1. The full list Table: Service | Monthly | Annual | Renews | Last used | Cancel link ## 2. Total damage - Monthly total - Annual total - Annualized cost of "forgotten" ones (the headline number) ## 3. Recommended cancellations - Tier 1 — cancel today (no activity in 90 days) - Tier 2 — cancel after this billing cycle (using but not enough) - Tier 3 — keep (used weekly, essential) ## 4. Drafted cancellation emails For services that require email cancellation (no self-serve), draft the email — do not send. ## 5. The savings number "If you cancel Tier 1 + Tier 2: $X/month, $Y/year." REVERSIBILITY: Never cancel anything yourself. Never click cancellation links. Produce the report. I do the clicks. MY CONTEXT: [Anything specific — e.g. "I run a one-person business, flag anything tax-deductible"]
Pro tip. Run this on the same date every quarter (the “quarterly subscription review”). Add it to your calendar. The cancel-now Tier 1 keeps shrinking once you institutionalize the habit.
The smart buy.
Stop checking 15 sites manually. Cowork surveys retail, marketplaces, refurbished, and education discounts, then gives you the all-in cost with shipping and tax.
Recipe: find the best price on this
Find me the best deal on a product or service I'm about to buy. WHAT I WANT TO BUY: [PRODUCT NAME, MODEL, OR EXACT SERVICE] CATEGORY: [Software / Hardware / Travel / Services / Other] QUANTITY: [N] TIME PRESSURE: [Need today / This week / Flexible] SHIP-TO: [CITY, COUNTRY] DO THIS: ## 1. Identify the canonical item - Exact product or plan (SKU if applicable) - Manufacturer / publisher price - Standard features at that price ## 2. Survey real prices today At least 8 places (whichever apply): - Manufacturer site - Amazon - 3 specialty retailers - 2 marketplaces (eBay, FB Marketplace for hardware) - Refurbished / open-box (Apple Refurb, BackMarket, Woot) - Education / non-profit / startup discount programs I qualify for For SaaS, also check: - Annual vs monthly pricing - Coupon code sites (RetailMeNot, Honey, dealhack) - Bundle deals - Lifetime deals on AppSumo / StackSocial ## 3. Total cost of ownership Not just sticker. Include: - Shipping - Tax (estimate at my zip) - Recurring fees - Switching cost from my current solution ## 4. The table Source | Price | Shipping | Tax | All-in | Condition | Notes ## 5. The recommendation "Cheapest: X at $Y. Best value: Z at $W because [reason]. Avoid: A because [reason]." Plus: should I wait? (upcoming sales, price history if available) ## 6. Direct purchase links For the top 2 options, give me direct links. REVERSIBILITY: Never purchase. Never enter payment info. Produce the comparison. I click buy. MY CONTEXT: [Anything that affects the choice — e.g. "I'm in Singapore, willing to pay 20% more for local warranty"]
Pro tip. Save the resulting table as a chat in Cowork before purchase. If the item drops 15% in price within 30 days, many retailers honor a price-match retroactively — you’ll have receipts.
The week reset.
Cowork reads your calendar, finds the anti-patterns (back-to-backs, no agenda, single-attendee meetings), and drafts the decline messages and reschedule proposals.
Recipe: fix my week before it starts
Look at my calendar and help me work better next week. LOOK AT: - All events for [DATE RANGE — default: the next 7 days] - Pattern of the last 30 days for baseline DO THIS ANALYSIS: ## 1. The week's shape - Total meeting hours per day - Number of meetings per day - Longest unbroken focus blocks - Days with > 5 meetings (red zones) ## 2. Anti-patterns to fix - Back-to-backs > 3 in a row (no buffer) - Meetings without an agenda in the description - Recurring meetings I'm on but haven't actually attended 3+ times - Single-attendee meetings (could be a doc?) - Meetings before 9am or after 6pm I didn't explicitly accept ## 3. Focus time to protect Suggest 2–3 deep-work sessions per week. Based on: - Open gaps that already exist - My historical productive hours (if I share) - Meetings I could legitimately decline or shorten ## 4. Decline candidates List 3–5 meetings where my presence is optional. For each: organizer, why I might not need to be there, draft polite decline message. ## 5. Reschedule candidates List meetings that would work better at a different time. For each: current slot, suggested slot, draft message proposing the change. ## 6. New meetings I should hold Based on my open threads and commitments, what 1:1s or check-ins should I schedule? Draft the invite description for each. OUTPUT FORMAT: Markdown report. Drafts in fenced blocks I can copy. REVERSIBILITY: Never decline, reschedule, or send anything. Never modify the calendar. Draft messages — I send them. MY CONTEXT: [Role, boss, team dynamics that should inform tradeoffs]
Pro tip. The “recurring I haven’t attended 3+ times” flag is gold. These are calendar zombies. Decline once, with grace, and reclaim hours per month.
The hand-off doc.
Turn any process in your head into step-by-step documentation you can hand off — to a teammate, a virtual assistant, or future you who’s forgotten.
Recipe: write the SOP for this process
Turn a process I do regularly into a step-by-step Standard Operating Procedure I can hand off.
THE PROCESS: [BRIEF NAME — e.g. "Monthly invoice run", "Onboarding a new client", "Posting a blog"]
WHO WILL USE IT: [Future me / Junior teammate / Virtual assistant / New hire]
HOW I'LL DESCRIBE IT TO YOU: [Pick one]
- Voice memo transcript I'll paste
- Screen recording transcript
- Bullet list of what I do
- Walk-through chat where you ask me questions
---
PRODUCE THIS SOP:
## Overview
- Purpose
- When to run (trigger + frequency)
- Approximate time to complete
- Required access / tools
## Prerequisites
- Accounts and logins needed
- Files or templates required
- Permissions
- Skills assumed
## The procedure
Numbered steps. Each step has:
- Action verb at the start ("Open", "Click", "Paste", "Send")
- Exact location (URL, app, menu path)
- Expected result
- Screenshot placeholder: [SCREENSHOT: what to capture]
- Common error + fix (if any)
## Quality checks
- How to verify the output is correct
- Self-check questions before marking complete
## Handoff section
- Who to escalate to if something breaks
- Where to log completion
- Where the artifact ends up
## Edge cases
- What's different on the first run
- International customers / weekends / month-end / etc.
## Revision log
- Date of this version
- What I'd test changing next quarter
INTERVIEW MODE:
Ask me 5–8 sharp questions to fill gaps in my description.
Then write the SOP.
Then ask 2 more questions to fill any holes in the draft.
OUTPUT:
Markdown SOP saved as a document I can drop into Notion or Drive.
REVERSIBILITY:
You write the SOP. Never automate or execute the process. The SOP is for humans.
MY CONTEXT:
[Industry, team size, anything that affects tone or detail]
Pro tip. The “Interview mode” instruction is the lever — without it Cowork guesses. With it, the SOP becomes audibly your process, not a generic template.
The receipt sort.
Tax-season saver. Reconstructs expenses from receipts, statements, and download folders, then categorizes, flags deductions, and outputs a spreadsheet your accountant can import.
Recipe: prep my expenses for tax season
Help me reconstruct my business or personal expenses for tax season (or quarterly review). PERIOD: [DATE RANGE — e.g. "2025 calendar year" or "Q1 2026"] PURPOSE: [Tax filing / Reimbursement / Budget review / Audit prep] SOURCES I'LL PROVIDE: - Email inbox (search receipts: "receipt", "order confirmation", "your invoice", "thank you for your purchase") - Credit-card / bank CSV exports (attached) - Downloads folder receipts (PDFs and images) - Recurring subscriptions list --- DO THIS: ## 1. Receipt extraction For every charge over [$X], find the matching receipt. For each: - Date - Vendor - Amount + currency (and converted USD if foreign) - Payment method - Description - Category ## 2. Categorization - Software / SaaS (deductible for business) - Hardware / equipment (capex, possibly depreciable) - Travel (airfare, hotels, ground, meals — rules apply) - Meals & entertainment (typically 50% deductible) - Professional services (legal, accounting, consultants) - Marketing / advertising - Office supplies - Education / training (deductible if business-related) - Health insurance (if self-employed) - Personal (not deductible — keep separate) ## 3. Tax-deductible flags For each item, flag: - DEDUCTIBLE (clearly business) - LIKELY DEDUCTIBLE (defensible with supporting note) - AMBIGUOUS (ask me a follow-up) - NOT DEDUCTIBLE (personal) ## 4. Missing receipts List charges on statements with no receipt found. Sorted by amount desc. Each: vendor, date, amount, "where to look" hint (vendor portal, app, etc.) ## 5. Spreadsheet Output a CSV ready to import into QuickBooks / Xero / Wave / Zoho Books. Columns: Date | Vendor | Amount | Currency | Category | Tax flag | Receipt link | Notes ## 6. Summary memo - Total spend - Spend by category - YoY change (if I share prior data) - Top 5 vendors - Anomalies worth investigating REVERSIBILITY: Read-only. Never categorize anything in QBO directly. Never delete receipts. Produce the spreadsheet — I import it. MY CONTEXT: [Self-employed? S-corp? Schedule C? Country? Industry-specific deductions?]
Pro tip. The AMBIGUOUS flag is the most valuable column — it’s where the conversation with your accountant actually pays off, and where AI shouldn’t pretend to know.
The solo retro.
Different from the Friday status note in Unit 05 — this is for you, not your stakeholders. Wins, losses, lessons, energy audit, and the one thing for next week.
Recipe: my personal weekly review
Generate my personal weekly review. This is for me, not for my boss. WEEK ENDING: [DATE — default: today] SOURCES: - Calendar (all events, including ones I declined or didn't attend) - Gmail sent items (what I committed to) - Notion / Drive (what I wrote or worked on) - Linear / Jira / GitHub (what I shipped or moved) - Slack DMs and channels I read (key threads) - A note from me (optional): [OPTIONAL NOTE] --- PRODUCE THIS REVIEW: ## Wins 3–5 things I actually accomplished. Specific, one-line outcome each. No "worked on X" — only "delivered X" or "decided X". ## Losses or stalls 1–3 things that didn't move. Honest cause: my time? Their dependency? Unclear scope? ## What I learned 2–3 lessons. Specific to a moment that happened this week. Each tagged with a context where I'd reuse it. ## Promises I made Every commitment I made this week — who I owe what, by when, current status. ## Surprises Things I didn't expect at the start of the week that ended up mattering. (These are usually signals.) ## Anti-portfolio Things I'm consciously NOT doing — to prove I made a choice, not just forgot. ## Energy audit Times I felt the best (sources of energy). Times I felt drained (sources of friction). Pattern worth changing next week? ## Next week — the one thing The single most important outcome for next week. Not three. One. ## The 4-week theme spotter If you have my last 4 reviews, what theme is emerging? What are my own actions telling me about what I actually care about right now? OUTPUT: Markdown document. Save as a dated file in [Notion page / Drive folder] so I can run this every Friday and compare. REVERSIBILITY: Read-only. Never modify Notion pages or Drive docs. Just produce the markdown. MY CONTEXT: [Role, what's on my mind, what I'm consciously trying to improve]
Pro tip. Run the “4-week theme spotter” only after week 4. The signal in the noise needs at least three preceding reviews to show up.
The inbox reset.
3,847 unread emails? Cowork triages everything into seven buckets, drafts the easy replies, lists the newsletters worth killing, and turns the rest into a prioritized work queue.
Recipe: get me to inbox zero today
Help me get to inbox zero without losing anything important. INBOX SCOPE: - All unread mail in my primary inbox - Last 30 days of read-but-not-archived (the "I'll get to it" pile) - Spam folder for false positives I should rescue --- DO THIS: ## 1. Triage every message into one of seven buckets: - REPLY-NOW — Important, quick to handle, < 2 minutes - REPLY-LATER — Important, needs thought, > 5 minutes - WAITING — I'm waiting on someone else, just track - TASK — Becomes a to-do, not a reply - READ — Newsletter or article, save and read later - ARCHIVE — Already handled, no action - DELETE — Spam, expired promotions ## 2. REPLY-NOW emails Draft the reply. Do not send. Keep under 5 sentences unless tone requires more. Use my style (pull from sent items as voice samples). ## 3. REPLY-LATER emails - Summarize the thread (2 lines) - Suggest the shape of my reply (decision needed / info needed / soft no) - Estimate time to draft a full reply - Sort by deadline + relationship importance ## 4. WAITING - Pull out the commitment + person + when promised - Flag any that have waited > 5 days for a nudge - Draft the nudge ## 5. TASK - Extract the action - Suggest the right list (calendar block, task manager, project file) - Draft the task description ## 6. READ - Categorize the newsletter / source - Note which I haven't engaged with in 30+ days - List unsubscribe candidates with unsubscribe URLs ## 7. The queue Prioritized work queue: - "Spend 10 min: process REPLY-NOW (count: N, all drafted)" - "Spend 30 min: review REPLY-LATER (count: N, ranked by deadline)" - "Schedule: TASKs go to [my system]" - "One-click: unsubscribe from [N] sources" ## 8. Trends - Top 5 senders by volume - Top 5 threads I keep avoiding (be honest) - Patterns in what's overwhelming me REVERSIBILITY: Drafts only. Never send. Never archive or delete. Never mark as read. Never unsubscribe — give me the URLs and I'll click. MY CONTEXT: [Job, communication style, things to never miss, sensitive senders]
Pro tip. The “threads I keep avoiding” line is the most useful diagnostic. The pattern of avoidance usually points to a single hard conversation you’ve been postponing.
The desktop reset.
Your Desktop, Downloads, and Documents folders organized into a sane tree. Duplicates flagged. Stale files surfaced. Sensitive files protected. The plan is a script you review first.
Recipe: organize my desktop chaos
cleanup.sh with a --dry-run mode, sensitive files (tax / passport / contract) flagged separately and never auto-moved, and running bash cleanup.sh --dry-run just prints the moves without touching a single file.Organize my Desktop, Downloads, and Documents folders. Find duplicates. Recommend deletions. FOLDERS TO SCAN: [Default: Desktop + Downloads + Documents] DEPTH: [Just these / Include subfolders / Up to N levels] OS: [macOS / Windows / Linux] --- DO THIS: ## 1. Inventory For each folder: - Total files and total size - File-type breakdown (docs, images, screenshots, archives, code, video, other) - Oldest file and newest file - "Stale" files: not opened or modified in 6+ months ## 2. Proposed structure Suggest a folder tree: - 01_Active (last 30 days) - 02_Reference (older but I might need it) - 03_Archive (older than 1 year, rarely touched) - 04_To_Sort (need a human pass) Plus topical subfolders if patterns emerge (e.g. "01_Active/Pradhya", "01_Active/Personal"). ## 3. The move plan For every file in scope: - Stay in place - Move to [proposed path] - Delete - Merge with [existing file] Don't actually move anything — produce the plan. ## 4. Duplicates - Exact duplicates (same hash) — keep newest, list deletable copies with full paths - Near-duplicates (same filename, different folders) — flag for review - Screenshot floods — "Screen Shot 2026-01-15 at 14.23.png" × 47 — bulk action proposal ## 5. The clutter list - Files I downloaded but never opened - Installers I no longer need (old DMGs, ZIPs) - Browser-generated junk (10 versions of "Untitled.pdf") - Cache / temp files safe to remove - Empty folders ## 6. Sensitive file warning Files that might contain sensitive data (filenames with "tax", "passport", "id-", "contract", "salary", "medical") — flag separately, never propose to move or delete without my explicit yes. ## 7. The shell script Generate `cleanup.sh` (or `cleanup.bat`) with: - Every move and delete as an explicit command - A `--dry-run` mode that just prints what it would do - A header comment explaining what it does - A `--backup` mode that copies to a temp folder before deleting REVERSIBILITY: You produce the plan and the script. You do NOT move, rename, or delete anything yourself. I review the script. I run `--dry-run` first. I run for real only when I'm sure. MY CONTEXT: [Cloud-synced folders (Drive, Dropbox, iCloud)? Anything in scope I should treat as sacred (don't touch)?]
Pro tip. The sensitive-file warning is non-negotiable. Even with the dry-run safety net, a typo in the script can ruin your day. Read the script before running. Always.
The investor update.
Monthly investor email used to take a Saturday morning. Cowork drafts it from what actually happened — tickets closed, deals won, calendar evidence — in the voice of your previous updates.
Recipe: draft my monthly investor update
Draft my monthly investor update from what actually happened this month. MONTH: [Month Year] SOURCES TO READ: - Calendar (every event this month) - Linear / Jira (closed tickets, shipped features) - GitHub (merged PRs, releases tagged this month) - KPI dashboard or spreadsheet (if shared) - Sent emails (commitments I made) - Slack #wins or #shipped channel - My last 2-3 investor updates for tone + format reference PRODUCE THIS UPDATE (600-800 words): ## TL;DR (3 bullets) - The single most important thing this month - The biggest miss or risk - The one ask for investors ## Metrics - MRR / ARR + change vs last month and goal - New customers + churn - Net revenue retention - Cash + months of runway - 1-3 north-star metrics specific to my business ## Wins - 3-5 specific shipped things - Each: what shipped + early signal of impact ## Lowlights & risks - 1-3 things that didn't go well - Honest about why - What we're changing ## Hires & team - New hires this month - Open roles (with the profile) - Departures (if any) ## Asks (the most important section) - Intros: specific people or companies - Customer leads: profile of customers we want - Hiring: roles where investors might know talent - Feedback: 1-2 strategic questions ## Next month - Top 3 priorities - Big milestones to watch for USE MY VOICE: Pull from my last 2 investor updates as voice samples. Match the sentence length, the directness, the in-jokes if any. REVERSIBILITY: Draft only. Never send to investors. I review, edit, send through my normal channel. MY CONTEXT: [Stage, sector, last round, any sensitivities to know about]
Pro tip. The “Asks” section is the single highest-leverage paragraph. Most updates bury it or leave it generic. Specific asks (named companies, named profiles) get investors to actually move.
The interview prep.
You have an interview tomorrow. You need to be sharp on the company, the interviewer, and the role. Cowork researches all three and gives you the brief, the questions to ask, and the stories to have ready.
Recipe: prep me for tomorrow’s interview
Prep me for an interview tomorrow. THE INTERVIEW: - Company: [COMPANY NAME] - Role: [ROLE TITLE] - Interviewer(s): [NAME · LinkedIn URL] - Format: [Behavioral / Technical / Case / Panel / Other] - My target start: [DATE or "ASAP"] SOURCES: - Their website (Careers + About + Blog) - The interviewer's LinkedIn + recent Twitter/posts if public - Their last 4 funding rounds or earnings calls (if public co) - Glassdoor / Levels.fyi for comp + culture signals - The job description (paste below or attach) - My resume (attach) PRODUCE: ## 1. The company brief (1 page) - What they do, in one sentence - Business model + key revenue lines - Stage + funding + key investors - Recent strategic moves (last 12 months) - The 3 things they likely care about right now ## 2. The interviewer brief - Tenure at company - Background (where they were before) - What they post or write about - 3 things to mention that show I've done my homework ## 3. The role decoded - The actual job behind the job description - 3 likely day-1 priorities - 3 things that could go wrong in the first 90 days ## 4. Questions to ask (10) - 3 about the role - 3 about the team - 2 about the company strategy - 2 about the interviewer's journey ## 5. Likely questions they'll ask - 5 behavioral they will ask (with STAR-ready bullet I'd use) - 3 hard ones I should rehearse - The one curveball they might throw ## 6. My stories ready - 3 stories from my career that fit this role - For each: situation, action, result, what I learned - Tag each with which interview question it answers ## 7. Red flags to watch for - 3 things to listen for that would change my interest REVERSIBILITY: Research and prep only. Never apply, never send messages to anyone at the company. MY CONTEXT: [Resume highlights, what excites me about this role, any deal-breakers]
Pro tip. The “red flags to watch for” line saves people from accepting jobs they’d regret. Three specific signals you commit to listening for, before adrenaline kicks in.
The post-call write-up.
You just got off a 45-minute call. You have four more today. The CRM entry would never get written. Cowork turns your notes (or the transcript) into a structured summary, drafted follow-up, and task list.
Recipe: turn this call into a complete write-up
Turn my call notes into a complete write-up: CRM entry, follow-up email, and task list. THE CALL: - Type: [Discovery / Demo / Follow-up / Renewal / Internal] - Customer or person: [NAME, COMPANY] - Date: [DATE] - Duration: [N minutes] INPUTS: - My typed notes (paste below or attach) - Transcript from Granola / Otter / Read.ai (if I have one) - Calendar event with attendees - The prior thread in this conversation (Gmail) NOTES: [Paste your notes here] PRODUCE: ## 1. The CRM-ready summary (5 sections) - **Who was on the call:** Names + roles - **What they said:** Their stated needs, pains, current situation - **What I said:** What I committed to, what I offered, what I demoed - **Decisions made:** Anything we agreed on - **Open questions:** Things to follow up on ## 2. The follow-up email (drafted) - Subject line (clear, action-oriented) - 4-6 sentences max - Recap the one thing we agreed - The single ask - Calendar link or specific next step ## 3. The task list Table: Task | Owner | Deadline - Tasks for me (immediate next 24h) - Tasks for the customer (if any) - Internal handoffs (sales engineer, legal, etc.) ## 4. The deal/relationship signal - Sentiment (warm / neutral / cool) - Risks observed - Champion vs blocker on their side - Probability change (up / down / flat) ## 5. What I learned (for me) - One thing I'd do differently next call - A question I should have asked REVERSIBILITY: Drafts everything. Never logs to the CRM directly. Never sends the follow-up. I approve and send. MY CONTEXT: [My role, what stage of the deal, any sensitivity]
Pro tip. Section 5 — “what I learned” — is the compounding return. Twenty post-call write-ups later, you have a personal sales-skills journal with patterns you can’t see in real-time.
The feedback themes.
Hundreds of pieces of user feedback across support, Slack, calls, NPS. Cowork clusters them, ranks them by frequency and trend, and tells you what users are actually saying — not what you wish they were.
Recipe: cluster user feedback into themes
I have hundreds of pieces of user feedback across channels. Cluster them and tell me what users are actually saying. SOURCES (last 90 days unless otherwise noted): - Support tickets (Zendesk / Intercom / Help Scout) - Slack channels where customers post (if connected) - Customer call transcripts (Granola, Gong) - NPS / CSAT survey responses (CSV attached) - App store / G2 reviews (if applicable) - Twitter / LinkedIn mentions (if I provide them) PRODUCE: ## 1. The themes (5-8 clusters) For each theme: - Name (short, evocative) - Frequency (count + % of total) - Trend (rising / steady / declining over the period) - 3-4 verbatim quotes (with source) - Sentiment (frustrated / confused / delighted / pragmatic) ## 2. The hidden themes - Things mentioned often as side-comments, not the main complaint - "Polite" feedback that's actually serious - Quotes from power users that read different from new users ## 3. The contradictions - Where users disagree with each other - Where users disagree with our public roadmap ## 4. Suggested PM actions - For each top theme: ship / investigate / accept / deflect - Effort estimate for ship-able fixes - Risk of doing nothing on each theme ## 5. Customers worth interviewing - 3-5 people who said something specific and articulate - For each: what to ask + why their case is interesting ## 6. The 3-slide engineering pitch - One theme per slide - Verbatim quote - Suggested move + estimate OUTPUT FORMAT: Markdown report I can drop into Notion. REVERSIBILITY: Read-only across all sources. Never replies to customers. Never edits tickets. MY CONTEXT: [Product, target user, current roadmap so themes can be tagged "already planned"]
Pro tip. Run this with the same prompt every quarter, but save the previous run. The diff between quarters — what’s a new theme, what dropped off — is the actual signal.
The standup digest.
Daily standup loses 15 minutes to “what did you ship yesterday” updates that everyone reads off Linear. Cowork generates the digest from the data, surfaces real blockers, and lets standup be conversation.
Recipe: generate today’s standup digest
Generate the standup digest for my team. What did each person ship yesterday and what's blocked. THE TEAM: - Names: [LIST OF TEAM MEMBERS] - Where they work: GitHub orgs, Linear teams, Slack channels SOURCES TO READ (last 24 hours): - GitHub: merged PRs, opened PRs, review comments per person - Linear / Jira: tickets moved (any column change), comments added - Slack: each person's posts in #team-eng and other relevant channels - Calendar: meetings each person had PRODUCE: ## 1. Per-person update (5 bullets max) For each team member: - ✅ Shipped: PR titles + impact - ⏳ In progress: ticket title + state - ⏸ Stuck: blocker + who they need - 📅 Today: from their calendar / assigned tickets - 🔔 Worth surfacing: anything notable ## 2. Team-level signals - Total PRs shipped vs rolling 7-day average - Bottlenecks (PRs waiting for review > 24h) - Risks (people working solo on critical paths) ## 3. The blocker list (most important) Table: Blocker | Blocked person | Who can unblock - Anything blocked > 48h gets flagged ## 4. Questions for me to ask in standup - 3 questions surfaced from the data - Each tied to a specific person/blocker ## 5. Things I can skip - Routine work that doesn't need air time - Status that's clear from the data OUTPUT FORMAT: Markdown. Concise. Standup is 15 minutes total. REVERSIBILITY: Read-only. Never posts to Slack, never updates tickets, never reassigns work. MY CONTEXT: [Team size, sprint cycle, what I'm trying to make more visible]
Pro tip. Run this in async-standup mode: post the digest in #standup at 9am, ask the team to reply with corrections + their actual blockers. Saves the live meeting for the 1–2 things that actually need conversation.
The renewal risk.
You have 30 renewals this quarter. Which 3–5 are most likely to churn, and what’s the one move per account that would change the trajectory? Cowork audits and ranks before the calls.
Recipe: audit my renewal pipeline for risk
Of my renewals this quarter, which 3-5 are at highest risk of churning? CONTEXT: - Renewal quarter: [QUARTER, e.g. Q3 2026] - Total accounts up for renewal: [N, or "look it up in CRM"] DATA SOURCES: - CRM (Salesforce, HubSpot): account size, last touch, owner, history - Support tickets per account (last 90 days) - Product usage data if available (logins, feature adoption, MAU/DAU) - Champion / sponsor status (still employed?) - Last NPS response per account PRODUCE: ## 1. The full ranked list Table: Account | ARR | Renewal date | Risk score (1-10) | Top 2 risk signals | Owner ## 2. The top 5 at-risk accounts (deep dive) For each: - Why they're at risk (specific signals, not vibes) - What we've done about it so far - What we should do in the next 2 weeks - Who else internally should know - The unlock: the one move that would change the trajectory ## 3. The drafted outreach For each top-5 account, draft: - Subject line - A 4-sentence email to the champion - A 3-bullet talking-points doc for the CSM's call ## 4. The internal nudges - Slack messages I should send to coworkers about specific accounts ## 5. The pattern - Across at-risk accounts, what's common? - Is the risk a product issue, a CSM staffing issue, or a segment issue? REVERSIBILITY: Drafts everything. Never sends, never logs to CRM, never tags accounts. MY CONTEXT: [My role, what counts as a save, our renewal motion]
Pro tip. Section 5 — the pattern across at-risk accounts — is the part that goes to product and exec. Individual saves are CSM work. Pattern detection is leverage.
The negotiation prep.
Salary, vendor renewal, lease, service contract. Cowork builds the playbook: their likely moves, your counters, scripts for three turns, your non-negotiables, and the closing line.
Recipe: prep me to negotiate this
Prep me to negotiate. Give me a playbook with their likely moves, my counters, and scripts. THE NEGOTIATION: - Type: [Salary / Vendor renewal / Office lease / Service contract / Other] - Other party: [NAME / COMPANY] - What I want: [SPECIFIC OUTCOME] - What I'll walk away over: [BATNA] - Time horizon: [When this needs to happen] DATA TO PULL: - Their public pricing (if applicable) - Market comp (Levels.fyi, BuiltIn, industry reports — research this) - My prior agreement or current contract - Their incentives (quarterly close, fiscal year end, etc.) - Anything they've said publicly PRODUCE: ## 1. The opening - What I should ask for first (the anchor) - The supporting evidence (3 points) - The tone (warm / firm / collaborative) ## 2. Their likely moves - 3 opening positions they might take - For each: why they'd open there + my counter ## 3. The script for 3 turns - Turn 1 (their open): What they say / what I say - Turn 2 (their counter): What they say / what I say - Turn 3 (the deal): The closer move ## 4. Their pressure points (leverage) - 3 specific reasons they want this deal to close - How to surface them without being mercenary - What I should NOT bring up (because it weakens me) ## 5. My non-negotiables - Walk-away conditions - Items I'll never give up - The clear "no" sentence ## 6. The trade matrix - Things I value > they value (give these up last) - Things they value > I value (concede these for big asks) - Things equal-value (use as currency) ## 7. The closing - The phrase that makes "yes" easy for them - What to do if they say "let me think about it" - The follow-up email if I have to walk REVERSIBILITY: Plans only. Never sends messages to the other party. I rehearse the script myself. MY CONTEXT: [Background, relationship history, what I really want vs what I'll accept]
Pro tip. Section 5 — the explicit walk-away conditions — is the part you write before you’re emotionally invested. Locking it in cold lets you walk warm.
The travel planner.
Multi-city trip. Calendar conflicts. Hotels. Flights. Visa requirements. Cowork plans the whole thing and surfaces what you’ll forget — passport expiry, international SIM, the meeting you should dial into.
Recipe: plan this trip end-to-end
Plan a trip with all the details handled. THE TRIP: - Cities + dates: [CITY 1 from-to · CITY 2 from-to · ...] - Travel reason: [Business / Personal / Conference / Family] - Travelers: [N adults · N kids · ages] - Budget: [TOTAL or per-day] - Constraints: [Dietary, mobility, can't-miss meetings, etc.] SOURCES: - My calendar (catch conflicts during travel dates) - My email (existing flight/hotel confirmations to incorporate) - Passport expiry date (I'll share) - Frequent flyer / hotel program tier (I'll share) PRODUCE: ## 1. Calendar conflict check - Every event during travel dates that needs handling - Suggested action per conflict (decline / reschedule / dial in / delegate) - Drafted decline messages where needed ## 2. The day-by-day itinerary - Morning / midday / evening per day - Travel time between things factored in - Buffer for jet lag on day 1 - One must-do + one easy backup per day ## 3. Booking checklist - Flights: routes, suggested airlines (with my loyalty tier), departure windows - Hotels: 3 options per city in budget, with one stretch pick - Local transit: best mode per city (Uber / metro / car / walking) - Reservations needing > 7 days notice (restaurants, museums) ## 4. Pre-trip checklist - Visa requirements per country (if international) - Passport expiry check (> 6 months from return date) - Vaccinations / health requirements - Travel insurance suggestion - International data plan / SIM - Currency / payment cards that work abroad ## 5. The packing list - Climate-aware basics - Business need vs personal - Easy-to-forget stuff (adapters, chargers, melatonin) ## 6. Emergency contacts - Embassy info per country - Insurance hotline - Family contact card formatted to print ## 7. The 24-hour-before run - Things to do the day before flying - Things to set up (out-of-office, mail hold) REVERSIBILITY: Plans only. Never books flights or hotels. Never confirms reservations. MY CONTEXT: [Loyalty programs, dietary, fitness routine I want to maintain, anything else]
Pro tip. The passport-expiry check has saved more trips than any other line in this prompt. Many countries require 6 months validity beyond return date, and people regularly find out at the gate.
The newsletter draft.
Your newsletter is due Thursday. Your week happened. Cowork turns one into the other — in your voice, pulled from your sent items and past issues, with subject line variants ready.
Recipe: turn this week into a newsletter
Draft this week's newsletter from my week. In my voice. Ready for me to edit and send. NEWSLETTER CONTEXT: - Name: [NEWSLETTER NAME] - Audience: [WHO READS IT] - Frequency: [Weekly / Bi-weekly] - Length: [SHORT 300w / MEDIUM 500-700w / LONG 1000w+] - Tone: [Conversational / Analytical / Punchy / Reflective] SOURCES: - This week's Notion notes / Drive docs - Sent items (anything I wrote in long form this week) - Calendar (what I attended, who I talked to) - Drafts folder (anything I started but didn't publish) - My last 5 newsletters (for voice + format reference) PRODUCE: ## 1. The hook (opening line) - 3 variations - Each should make a regular reader stop scrolling ## 2. The main piece - One clear claim or story - 3 supporting moves (specific examples, data, or counterpoints) - A satisfying landing ## 3. The bonus section (if relevant) - 1-3 short items: links, things I'm reading, a question to readers ## 4. The CTA - One specific ask (no more than one) - Reply to / share with / book a call / etc. ## 5. The signoff - In my voice (pulled from samples) ## 6. The metadata - Subject line (3 variants, A/B testable) - Preview text - Suggested send time based on past open rates if I share data ## 7. The pieces that didn't make it - Stories from the week that didn't fit - Save these in a "future newsletters" doc REVERSIBILITY: Draft only. Never sends, never schedules. I copy the final version and send myself. MY CONTEXT: [My niche, the one thing readers value most, anything off-limits this week]
Pro tip. Section 7 — the cutting-room floor — is gold. After 12 newsletters you have a backlog of half-baked ideas to pull from on weeks when nothing happened worth writing about.
The 1:1 prep.
You have a 1:1 tomorrow with someone on your team. The last one was two weeks ago. What did they ship? What’s stuck? What should you ask without surveilling them? Cowork builds the brief.
Recipe: prep tomorrow’s 1:1
Prep me for tomorrow's 1:1 with [TEAM MEMBER]. THE PERSON: - Name: [NAME] - Role: [ROLE] - Last 1:1: [DATE] - Cadence: [Weekly / Bi-weekly / Monthly] SOURCES: - Their work in the last 1:1 cycle: - GitHub (PRs they shipped, code review they did) - Linear / Jira (tickets they closed, moved, opened) - Notion / Drive (docs they wrote) - Slack (channels and DMs they were active in) - Our last 3 sets of 1:1 notes (Notion or Drive) - Their goals doc / career conversation notes if I have them PRODUCE: ## 1. What they shipped - Specific output since last 1:1 (with impact lines) - Quality observations (PR review patterns, code quality) - Pace observations (faster / steadier / slower than usual) ## 2. What they're working on - Current focus + how it's progressing - Anything stuck ## 3. What's changed about them - Tone / morale signals from Slack - New skills demonstrated - Patterns worth surfacing ## 4. Follow-through on last 1:1 - Things they said they'd do, status of each - Things I said I'd do (most important to track) ## 5. Three things to discuss tomorrow - Specific topics, not generic categories - For each: why now, what good looks like coming out of it ## 6. Three questions to ask - Open questions, not yes/no - Questions that surface what they aren't saying ## 7. What I shouldn't bring up - Things that would feel like surveillance - Things they should be allowed to raise first ## 8. After-meeting tasks - Things I'll likely commit to - Notes I should take during the meeting (not just after) REVERSIBILITY: Read-only. Never DMs the person before the 1:1. Never modifies their tickets or goals. MY CONTEXT: [How long they've been on the team, what we're working on changing, any sensitivities]
Pro tip. Section 7 — what NOT to bring up — is the discipline that separates a good manager from a creepy one. Surveillance has tells. So does avoidance. Bring up neither in the wrong way.
Reversibility — the Cowork rule.
All seven use cases above share the same safety pattern. Cowork reads, summarizes, and drafts. Sending, posting, paying, deleting — that’s you.
The pattern is durable, not just for this practice. The agentic future is built on this:
- Draft, don’t send. Email, Slack, LinkedIn — always.
- Copy, don’t move. Files in Drive.
- Append, don’t overwrite. Notion pages.
- Read-only by default. Until you have run the same workflow ten times and trust it, the connector is read-only.
Where Cowork ends and Code begins.
Cowork is for you. Claude Code is for your team and your repos. The hand-off is where most professionals get stuck. The rule is simple.
| Reach for | When the work is |
|---|---|
| Cowork | Personal, ad-hoc, one-off. Lives in your inbox, your calendar, your files. |
| Claude Code | Repeatable, codified, shared. Lives in a repo. Has tests. Other people use it. |
| A Cowork skill | Repeatable and personal. Same workflow 10x. Lives in a saved chat or starred prompt. |
| A Code plugin | Repeatable, codified, team-wide. Lives in a versioned plugin manifest other devs install. |
The most common upgrade path: a Cowork workflow you run weekly for two months becomes a Code plugin your team uses for years. Practice 02 (Claude Code) is where you take it next.
- Name the job. One sentence, one outcome (“Turn my Monday board-meeting notes into action items with owners”). If it needs “and,” it’s two recipes.
- Fill the template below — the same shape every recipe on this page uses: When to use, What you get, You know it worked when, Reversibility, then the prompt with a sectioned output spec and
[BRACKETS]for what changes each run. - Run it once in Claude Desktop on a real input from your week (a real file, a real thread — not a hypothetical).
- Save it as a starred chat named for the job.
# Recipe: [THE JOB, AS A VERB PHRASE] WHEN TO USE: [the trigger — what on my week makes me reach for this] WHAT I GET: [the artifact this produces, in one line] YOU KNOW IT WORKED WHEN: [one observable check — a section count, a file, a number] REVERSIBILITY: [what it must never do — e.g. draft, never send] --- THE PROMPT: [One sentence stating the job.] INPUTS — I'll point you at: - [source 1] - [source 2] DO THIS: 1. [step] 2. [step] PRODUCE (use these exact sections): ## [Section 1] ## [Section 2] ## [Section 3] REVERSIBILITY: [Read-only / draft-only rule, stated plainly.] MY CONTEXT: [Anything Cowork should know to tailor the output.]
Verify: your recipe has all five header lines (When to use, What I get, You know it worked when, Reversibility, The prompt) and at least one ## Section in its output spec; you ran it once and the result matched your “you know it worked when” check. If it did, you don’t have a chat anymore — you have a repeatable skill. That’s the upgrade path to Practice 02.