What Cowork Actually Is (and Is Not)

Before you install anything, you need to understand what you are installing. Claude Cowork is not another chatbot. It is not a fancier version of ChatGPT. It is, without exaggeration, the closest thing available to an AI employee who actually does things instead of just talking about things.

Regular Claude Chat lives in a browser tab and gives you text responses you then have to copy, paste, reformat, and wrangle into something useful. Cowork skips all of that. You point it at a folder on your computer, describe what you need done, and walk away. Come back to finished files: Word documents, Excel spreadsheets with working formulas, presentations, organized folders. Saved directly to your hard drive.

Three Modes of Claude

What Cowork Can Do That Chat Cannot

• Read, create, edit, move, and delete files on your computer.
• Build real Word documents, Excel spreadsheets with formulas, PowerPoint decks.
• Connect to Gmail, Google Calendar, Slack, Drive, and 8,000+ apps.
• Control your Chrome browser to interact with any website.
• Run tasks automatically on a schedule (daily, weekly, custom).
• Be triggered remotely from your phone (Dispatch).
• Remember context within Projects across sessions.

The 5 Levels of Cowork Mastery

Every Cowork user moves through these levels. The goal of The CoWork Accelerator is to take you from Level 0 (never used it) to Level 3 (skills installed, real work flowing), and give you the roadmap to reach Level 5 (full automation).

The 7 Core Capabilities

Before you start building, understand the full scope of what Cowork can do. These seven capabilities are the building blocks of everything.

Rank these by relevance to your business. Which 3 capabilities would solve your biggest current pain points? Write them down. You will prioritize your setup around these.

Requirements + Pre-Flight Checklist

System Requirements

Subscription Plans

Pre-Flight Checklist

• ☐Paid Claude account (Pro or higher) is active. Verify at claude.ai/settings/billing.
• ☐Claude Desktop app downloaded from claude.ai/download (not the browser version).
• ☐A plain-text editor is available for creating .md files.
• ☐Chrome browser is installed (optional now, required for Step 17).
• ☐You have a Google account (for Gmail, Drive, Calendar connectors later).
• ☐You understand: Chat = conversation. Cowork = delegation. Code = development.

Download + Install Claude Desktop

1. Go to claude.ai/download in your web browser.
2. Click the download button for your operating system (macOS or Windows).
3. Open the downloaded file and follow the installation prompts.
4. Launch the Claude Desktop application from your Applications folder (Mac) or Start Menu (Windows).
5. Sign in with the same Anthropic account tied to your paid subscription.
6. Look at the top of the window. You will see tabs: Chat, Cowork, and possibly Code.
7. Click the Cowork tab to switch modes.
8. If this is your first time, the app will download a Linux virtual machine (approximately 2 GB). Wait for it to finish.
9. Once the download completes, you should see the Cowork interface: an empty state asking you to select a folder or start chatting.

Lock Down Your Privacy

Before you type a single prompt, before you connect a single folder, before you do anything at all, lock down your privacy settings. Cowork has direct access to files on your computer. If you will be feeding it proprietary business data (competitor intel, customer lists, pricing, client info, internal strategy), this step is non-negotiable.

Configure Privacy Settings

1. In the Claude Desktop app, click the Settings icon (gear icon, or File > Settings).
2. Navigate to Settings > Privacy.
3. Toggle OFF any options related to using your data for model improvement or training.
4. Confirm the setting saved by navigating away and back.

Additional Safety Settings

• Never point Cowork at your entire Documents or Desktop folder. Cowork has real read/write access to whatever folder you share. Keep it contained to one dedicated project folder.
• Plan to start with approval-required permissions. When you connect tools like Gmail later (Step 16), set destructive actions like "Send Email" to "Needs Approval" so Claude asks before doing anything irreversible.
• Add this rule to your instructions (we will do this in Step 8): "Never overwrite existing files without asking me first. Never delete files without asking me first."
• Create a Safety SOP. Define three tiers of actions: Allowed (reading files, drafting content), Review-Required (sending emails, posting to Slack), Blocked (deleting anything, publishing without approval).

Choose Your Model + Use Extended Thinking

Cowork lets you choose which Claude model powers your tasks. This choice directly affects output quality AND how fast you burn through your usage allocation. Claude comes as three models: Opus, Sonnet, and Haiku. Each is purpose-built for a different kind of work.

The Three Models (Opus, Sonnet, Haiku)

The Model Selection Rule

• Opus + Extended Thinking for intelligence work. Strategy, frameworks, offer design, positioning, launch engineering, copy that sells.
• Sonnet for simple work. File ops, routine content, basic research, everyday documents.
• Haiku for speed-critical batch work. Classification, quick lookups, high-volume repetitive workflows.

Extended Thinking (and Adaptive Thinking)

Extended Thinking gives Claude permission to reason longer before responding. Instead of answering instantly, it works through the problem step-by-step, explores tradeoffs, and checks its own reasoning before committing to an answer. In practice, this is like asking a senior strategist to actually think about your problem for 30 seconds instead of firing back a reflex response.

This is the single biggest quality lever most people leave unused. Extended Thinking on complex tasks is the difference between generic output and genuinely considered output.

How Extended Thinking Works in Chat vs Cowork

When Extended Thinking Earns Its Keep

Turn it on (in Chat) or lean into Opus (in Cowork) for:

• Complex reasoning and multi-step analysis.
• Strategic planning and framework design.
• Writing that has to persuade, sell, or teach.
• Debugging logic problems or tracing unclear root causes.
• Math, pricing models, and financial analysis.
• Any high-stakes decision where accuracy matters more than speed.

When to Skip It

• Simple factual questions and quick lookups.
• Routine formatting or file operations.
• Casual brainstorming where you want speed.
• One-off creative prompts that do not need layered reasoning.

How to Switch Models

1. Open Cowork.
2. Click the model dropdown at the top of the interface.
3. Select Opus, Sonnet, or Haiku.
4. The selection applies to the current session. Switch anytime.

Understanding Usage

• Usage resets on a rolling 5-hour window. If you hit your limit, wait. It resets automatically.
• Cowork uses significantly more tokens than Chat because it is doing real work, not just conversation.
• Opus with deeper thinking burns capacity faster than Sonnet. That is the tradeoff for higher-quality output.
• Check your usage: Settings > Usage. You will see a percentage for each model.
• Use Chat for simple questions. Reserve Cowork for tasks that genuinely need file access and multi-step execution.
• Simplify prompts to reduce token usage. Overly verbose prompts waste tokens. Be specific but concise.

Program Your Two Systems (Personal Preferences + Global Instructions)

Most people do not realize they are programming two separate systems when they use Claude. These systems do not share settings. They do not sync. Configuring one does nothing for the other. Understanding the distinction is the difference between Claude that feels like a generic chatbot and Claude that operates like a trained team member.

The Core Framework: Why Two Settings Exist

The Token Economy: Why This Matters Mechanically

In claude.ai chat, Personal Preferences loads at the start of every conversation. The cost is low. You can afford to be more expansive here. Write instructions that teach Claude how to challenge you, match your standards, and think at your level.

In Cowork, Global Instructions loads PLUS CLAUDE.md PLUS Brief files PLUS any active Skills PLUS your prompt. A single Cowork session can load 5,000 to 20,000+ tokens of context before Claude starts thinking about your task. Every unnecessary word costs capacity on every single session.

Part 1: Personal Preferences (Claude Chat)

Where to Find It: claude.ai > Settings > Profile > "What preferences should Claude consider in responses?"

This template has 8 sections. Each one programs a different dimension of how Claude interacts with you in every chat conversation on web and mobile.

Personal Preferences Template (Copy and Customize)

# IDENTITY
              I am [Your Name], a [your role] specializing in [your domain].
              I help [target audience] achieve [specific outcome] through [your method/framework].
              [One sentence about your experience level or credibility marker.]

              # COMMUNICATION PROTOCOL
              - Tone: [bold and direct / warm but precise / analytical and structured]
              - Default to [paragraphs / frameworks / numbered steps] unless I ask otherwise
              - Lead with the answer, then explain the reasoning. Never bury the point.
              - Use [American/British] English
              - Never use em-dashes. Use periods, colons, or parentheses instead.
              - Keep sentences under [15/20/25] words when possible
              - When I ask a simple question, give a concise answer. Do not over-elaborate.
              - When I ask for strategy, show me the system architecture, not a list of tips.

              # MY TERMINOLOGY
              - I use "[your term]" not "[generic term]"
              - I say "[your phrase]" not "[common phrase]"
              - I refer to "[your framework name]" not "[generic description]"
              - I avoid these words entirely: [list 3-5 banned words/phrases]
              - My core philosophy is "[your motto/principle]"

              # EXPERTISE CALIBRATION
              - Treat me as an expert in: [list 2-4 domains]. Do not explain fundamentals.
              - I am actively learning: [list 1-2 areas]. Provide more depth here.
              - Assume I understand: [marketing funnels / code / financial modeling / etc.]
              - Do not assume I understand: [legal nuance / advanced statistics / etc.]

              # DECISION FRAMEWORK
              - I prioritize [profitability / speed / simplicity / scalability] over [growth / perfection / complexity / features]
              - When presenting options, give me [one clear recommendation / 2-3 ranked options with tradeoffs]
              - Always include: the implementation path, not just the idea
              - Filter every suggestion through: "Can this be systematized and taught to others?"
              - I value [frameworks that compound over time] over [one-time tactics]

              # THINKING PARTNERSHIP
              - Challenge my assumptions with data and logic. Do not agree just to be agreeable.
              - If you see a flaw in my reasoning, say so directly. Explain why.
              - Offer counterarguments when I present a strong opinion. Steelman the opposing view.
              - When I am wrong, tell me clearly and show me what is right.
              - Ask clarifying questions before making assumptions on ambiguous requests.
              - Do not flatter or praise my ideas. Evaluate them honestly.

              # OUTPUT QUALITY BAR
              - Every recommendation must include concrete implementation steps
              - Use real numbers (%, $, timeframes) not vague claims like "significant improvement"
              - When referencing a framework, explain how it applies to my specific situation
              - Break complex ideas into memorable, visual metaphors I can teach to others
              - If the answer requires nuance, provide it. Do not oversimplify to sound confident.

              # NEVER DO THIS
              - Never start with "Great question!" or similar filler
              - Never use phrases: "move the needle," "at the end of the day," "game-changer," "effortless," "leverage," "synergy"
              - Never hedge with "it depends" without then committing to a specific recommendation
              - Never apologize unnecessarily or pad responses with qualifiers
              - Never provide generic advice that could apply to any business
              - Never repeat my question back to me before answering
              - Never use more than 2 bullet points when a paragraph would be clearer
              - Never guess on facts. If you are not confident in a claim, say so and verify with web search before answering. I would rather wait 10 seconds for a verified answer than get a plausible-sounding wrong one.

Part 2: Global Instructions (Claude Cowork)

Where to Find It: Claude Desktop > Settings > Cowork > Global Instructions > Edit

This template has 10 sections. Sections 1 through 3 are a COMPRESSED version of your identity from Personal Preferences (lean, no teaching, just facts). Sections 4 through 10 are entirely new: they program how Claude operates as an employee.

Global Instructions Template (Copy and Customize)

# IDENTITY
              I am [Your Name], [your role] at [business name].
              I help [audience] achieve [outcome]

              # COMMUNICATION STYLE
              - Direct tone. Action-oriented. No filler.
              - No em-dashes. Use periods, colons, or parentheses.
              - Avoid: [list 3-5 banned words]
              - Default to frameworks and systems, not isolated tactics.
              - Include implementation steps in every recommendation.

              # BUSINESS CONTEXT
              - Primary offer: [describe your main product/service]
              - Audience: [one sentence describing your ICA]
              - Content platforms: [YouTube, email, Instagram, LinkedIn, etc.]
              - Email platform: [Kartra, ConvertKit, ActiveCampaign, etc.]
              - Website/funnels: [GHL, WordPress, ClickFunnels, etc.]
              - Payment processor: [Stripe, PayPal, etc.]
              - Other tools: [Slack, Google Workspace, Notion, etc.]

              # SESSION START PROTOCOL
              - Before starting any task: read CLAUDE.md at the folder root.
              - For content tasks: also read voice.md and icp.md from the-brief/.
              - For marketing tasks: also read offer-stack.md and proof-points.md.
              - For research tasks: also read competitive-positioning.md.
              - If context files are missing, ask me before proceeding with assumptions.

              # PLANNING PROTOCOL
              - For tasks requiring 3+ steps: show a numbered plan before executing.
              - Wait for my approval before beginning execution on complex tasks.
              - For simple tasks (single file, clear instructions): execute directly.
              - When breaking work into phases, name each phase and its deliverable.
              - If a task will take significant tokens, warn me and suggest simplification.

              # FILE OPERATIONS
              - Save all final outputs to the outputs/ folder.
              - Name files: YYYY-MM-DD-descriptive-name.[ext]
              - Default to .docx unless I specify otherwise.
              - One deliverable per file. Never combine multiple deliverables.
              - When creating drafts, prefix with DRAFT- until approved.
              - Never overwrite existing files without asking me first.
              - NEVER delete files. If something needs to be removed, move it to _to-be-deleted/ at the workspace root.
              - Never rename existing files without telling me.

              # APPROVAL GATES
              Allowed (do without asking):
              - Reading any file in the project folder
              - Drafting content and documents
              - Web research and analysis
              - Creating new files in outputs/
              - Suggesting improvements to existing work

              Review-required (show me first, wait for approval):
              - Sending emails via Gmail
              - Posting to Slack or social media
              - Modifying any existing file I created
              - Bulk operations on multiple files
              - Any action that cannot be undone

              Blocked (never do):
              - Deleting files or folders. Move them to _to-be-deleted/ instead. I review and empty that folder monthly.
              - Publishing content externally
              - Accessing folders outside my project directory
              - Sharing files with external parties
              - Making purchases or financial transactions

              # QUALITY ASSURANCE
              Before delivering any output:
              1. Verify tone matches my voice guidelines (check voice.md if available).
              2. Confirm all claims are specific and supported, not vague.
              3. Check that the deliverable matches the requested format and structure.
              4. Verify file is saved to the correct location with correct naming.
              5. Confirm no guardrail violations (check guardrails.md if available).
              6. Flag any sections where you are uncertain or made assumptions.

              # ERROR HANDLING
              - If a task is ambiguous: ask one focused clarifying question before proceeding.
              - If a file or context is missing: tell me what is missing and ask how to proceed.
              - If you are unsure about a claim or recommendation: flag it with "[NEEDS VERIFICATION]".
              - If a task exceeds your capability: say so directly and suggest an alternative.
              - NEVER guess on facts. If you are not confident, run a web search and verify before answering. I would rather wait 10 seconds for a verified answer than get a plausible-sounding wrong one.

              # TOKEN EFFICIENCY
              - Do not repeat back the contents of files you have read. Summarize in one sentence if needed.
              - Do not add preambles like "Sure, I'd be happy to help with that."
              - Do not explain your reasoning unless I ask for it.
              - Keep plans under 10 lines.
              - When delivering files, state the filename and location. Do not describe the entire contents.

How to Verify Both Systems

Personal Preferences Verification

1. Open a new claude.ai chat (web or mobile).
2. Type: "Based on what you know about me, describe who I am, how I prefer to communicate, and what I would consider a low-quality response."
3. Claude should nail your identity, tone preferences, and anti-patterns.
4. If it misses anything, your Personal Preferences have a gap. Fix it.

Global Instructions Verification

1. Open a new Cowork session in the Claude Desktop app.
2. Type: "Before I give you a task, tell me: (1) Who am I? (2) What are the rules for how you handle files? (3) What requires my approval before you act? (4) What should you read before starting a content task?"
3. Claude should accurately state your identity, file protocols, approval gates, and session start protocol.
4. If it misses anything, your Global Instructions have a gap. Go back and re-save.

Cross-System Verification

1. Open claude.ai chat and ask Claude to write a short LinkedIn post in your voice.
2. Open Cowork and give the same task.
3. Both outputs should match your tone and terminology.
4. If one is on-brand and the other is generic, the generic one has a configuration gap.

When to Update

Build Your Folder Architecture

This is the most important structural decision in your entire setup. In Cowork, each folder you connect is a self-contained brain. Different folder equals different Claude brain. Skills, context files, and documents in one folder are invisible from another folder. Get this architecture right and the system compounds. Get it wrong and you will rebuild the same context six times across six sessions.

The Cowork Method uses one of two patterns depending on how you work. Pick the one that matches your business reality today. You can always upgrade later.

Pattern A: Single Workspace (Starter Architecture)

Use this if you run one business focus and do not need separate project memory. Simple. Fast. Good for starting. The entire workspace is one folder with one CLAUDE.md.

Pattern B: The Command Center (Architect Architecture, Recommended)

Use this if you run multiple projects, clients, ventures, or content streams under one roof. This is the architecture Travis runs. It is designed to scale without collapsing. CLAUDE.md becomes a thin boot loader that points to a living master index called the Command Center. Each project gets its own sub-folder and its own Project Index file. Claude reads the Command Center at the start of every session and knows exactly where to go.

Step 1: Create Your Workspace Folder

1. Open your file explorer (Finder on Mac, File Explorer on Windows).
2. Navigate to the location where you want your workspace to live. Most people use Documents. If you want cloud-based redundancy, put it inside Dropbox, iCloud Drive, Google Drive, or OneDrive (see optional Section 9B).
3. Create a new folder with a specific, descriptive name. Good names: "Cowork Method," "Content Business," "Client Work." Bad names: "Desktop," "Documents," "Stuff."

Step 2: Build the Internal Structure (Pattern B: Command Center)

Step 3: Connect the Folder to Cowork

1. In Cowork, click the folder icon or "Select Folder" button.
2. Navigate to the workspace folder you just created.
3. Click "Always Allow" to give Cowork persistent access.

The Command Center File (Your Living Master Index)

The Command Center is the file that makes everything work. Open it and you see the current state of your business at a glance. Claude reads it at the start of every session. You update it at the end of every session. This is how you beat the "Claude forgets everything" problem.

The Command Center has five blocks, in this order:

Keyword Triggers: Scan and Full Scan

Two text commands built into the Command Center. No installation. Just type them at the start of a session.

The Project Index Pattern (One Per Project)

Every project folder gets its own index file at the root named 0) PROJECT-INDEX.md. Required. Not optional. This is where the project's memory lives. When Claude works inside a project folder, it reads this file first.

The Project Index has seven blocks:

• Project Summary. What this project is, in one to three sentences. Plain English.
• Canonical Files. Table of the handful of files that matter most in this project. Location and description of each. Claude references these first.
• Folder Structure. Built out as the project develops. Shows Claude where things live.
• Open Decisions. Decisions you are still chewing on. Claude picks these up and helps you close them.
• Open To-Dos. What is pending. Simple checkbox list.
• Session Log. Table with Date and Summary columns. Claude writes a row at the end of every session inside this project. This is how project memory compounds.
• Notes. Everything else. Decisions, context, random thoughts, links.

The 4 Folder Mistakes That Break Everything

Use Cowork Projects for Persistent Memory

On top of your folder architecture, Cowork now includes Projects. A Project is a dedicated workspace that wraps your folder with four things: persistent memory, custom instructions, scheduled tasks, and local context (files, links, and folders). Think of a Project as the folder plus a running brain that remembers what you did last time.

Standalone Cowork sessions forget everything when they end. CLAUDE.md and Brief files give Claude static context, but they do not record what happened yesterday. Projects solve that. Memory is scoped to each individual project, so what Claude learns while building your Practitioner Accelerator launch stays with the Practitioner Accelerator project. It does not bleed into your Heal At Scale content project or your client work.

What a Project Includes

How to Create a Project

1. Open Cowork. Click the Projects area in the sidebar.
2. Choose how to start: fresh setup, import from an existing Claude chat project, or point at an existing folder on your computer.
3. Name the project. Attach your working folder.
4. Add custom instructions (tone, rules, constraints specific to this project).
5. Add any scheduled tasks you want running for this project.
6. Start working. Claude remembers what happens across sessions.

Optional: Add Cloud-Sync for Extra Redundancy

Every file the Cowork Method produces lives inside one folder on your computer. If you are serious about that folder, you want it protected by at least one backup system. Most operators run one of two setups: daily computer-wide backup (Time Machine, Backblaze), or per-folder cloud-sync (Dropbox, iCloud Drive, Google Drive, OneDrive). Either one works. Running both is belt-and-suspenders territory and gives you the highest resilience.

The advantage of cloud-sync on top of a daily backup: your workspace lives on every machine you log into. Walk to a different laptop, open your cloud-sync folder, and everything is there. No manual restore. No "wait for the backup to finish." Same files, same state, instantly.

Pick Your Cloud-Sync Service (If You Want One)

The Setup Sequence

1. Install your cloud-sync app. Download the desktop client for whichever service you picked. Sign in. Let it install its root folder on your hard drive.
2. Move your Cowork workspace folder inside the cloud-sync root. Example: Dropbox/Cowork Method or iCloud Drive/Cowork Method. If your workspace folder already exists elsewhere, drag it in.
3. Confirm sync is active. On Mac, look for the cloud or checkmark icon next to the folder in Finder. On Windows, check the sync status icon in the system tray. If you see "syncing" or "up to date," you are good.
4. Re-select the folder in Cowork if the path changed. Every file Claude writes will land in the synced folder and get backed up automatically.

Write Your CLAUDE.md File

If you only do one thing from this entire guide, do this. Everything else (skills, connectors, automation) works dramatically better when Claude has project-specific context. The CLAUDE.md file provides that context automatically.

This is a plain text file at the root of your project folder. Claude reads it automatically at the beginning of every Cowork session before you type your first message. Think of it as a department briefing that tells the employee: "Here is what THIS project needs. Here is what is different from the global defaults."

How CLAUDE.md Relates to the Other Layers

What CLAUDE.md IS For: The 6 Project-Specific Sections

The 4 Mistakes That Cripple Your CLAUDE.md

• Mistake 1: Skipping it entirely. Every session in this folder starts from zero. Output feels generic because Claude knows nothing about this specific project.
• Mistake 2: Duplicating Global Instructions. You already told Claude who you are, your tone, your tech stack, and your file rules in Global Instructions. Repeating it in CLAUDE.md burns tokens twice and creates potential conflicts if you update one but not the other.
• Mistake 3: Making it too long. 200+ lines of rules, context, and edge cases. Important info gets buried. Sweet spot: 25 to 40 lines. Put deep reference material in the-brief/ folder, not here.
• Mistake 4: Never updating it. Your goals shift. Your projects evolve. A CLAUDE.md from 3 months ago is giving Claude outdated priorities. Review every 2-3 weeks.

Template (Copy and Customize)

# CLAUDE.md

## Project Purpose
This folder is for [what this project produces]. It supports [which part of my business].
[One sentence about what makes this project different from my other work.]

## Audience for This Project
This project serves [specific audience for THIS work].
They are [their level / situation / what they need].
[One sentence about what success looks like for this audience.]

## Project-Specific Rules
- [Rule that applies ONLY to this folder, not globally]
- [Any override to global defaults: different tone, different format, different audience level]
- [Specific constraints: word counts, format requirements, platform-specific rules]

## Context Files
- For content and copy tasks: read the-brief/voice.md
- For audience targeting: read the-brief/icp.md
- For offers and pricing: read the-brief/offer-stack.md
- For competitive context: read the-brief/competitive-positioning.md
- For guardrails: read the-brief/guardrails.md

## Current Goals (Update Every 2-3 Weeks)
- [Goal 1: what you are working toward in this project right now]
- [Goal 2: next milestone or deliverable]
- [Goal 3: what success looks like this month]

## Session Start Behavior
When I open a session in this folder:
1. Read this file and any context files relevant to my first task.
2. Check logs/ for the most recent daily log.
3. Summarize what is open and what was last completed.
4. Ask if priorities have changed before starting work.

How to Test It

1. Save your CLAUDE.md file at the root of your project folder (not inside any subfolder).
2. Start a brand new session in that project folder.
3. Ask Claude: "What is this project for and who does it serve?"
4. Claude should answer with project-specific details, not your generic business description. If it gives your global identity instead of project context, your CLAUDE.md is duplicating instead of specifying.
5. Ask Claude to write something for this project. Check: Does it reflect the project-specific audience and rules, or does it default to your global tone?

When to Update

• Your project goals change (new phase, new milestone, new direction).
• You add new context files to the-brief/ that Claude should know about.
• You discover project-specific rules or constraints Claude keeps missing.
• The audience for this project shifts or narrows.

Plan to review every 2-3 weeks. The Current Goals section will change most frequently. Set a recurring reminder.

Command Center Companion (For Pattern B Architects)

If you chose Pattern B in Section 9, your CLAUDE.md is thin. It does not carry the full project brain. Instead, it tells Claude to read the Command Center file, which carries the brain. This keeps the boot loader small (saves tokens) and moves the living intelligence into a file you can see and edit at a glance.

The Thin Boot Loader (Pattern B CLAUDE.md)

# CLAUDE.md

Read `0) COMMAND-CENTER.md` at the root of this workspace before doing anything else.
It contains the Architect's Foundation, Active Build Board, project registry,
operating standards, keyword triggers, and current goals.

When I am working inside a specific project subfolder, also read that project's
`0) PROJECT-INDEX.md` before touching any project files.

All operating instructions live in those two files. This boot loader only exists
to tell you where to look first.

The COMMAND-CENTER.md Template

Save this as 0) COMMAND-CENTER.md at the root of your workspace folder. Replace the placeholders with your own content. Claude will read it at the start of every session.

# [YOUR NAME] Cowork Command Center

## How This Works
At the start of each session, Claude reads this file. If I type a keyword trigger, Claude runs
the matching protocol. Otherwise, default to a light entry: scan the Active Build Board,
surface what is live, and ask what I want to work on.

If I am working in a specific project, Claude also reads that project's
`0) PROJECT-INDEX.md` before touching project files.

**Keyword Triggers**
- **Scan.** Light orient. Read this Command Center (focus on Active Build Board), then ask
  what I want to work on. Fast path.
- **Full Scan.** Comprehensive sweep. Read this Command Center + every project's Project
  Index + my calendar + any unfiled inputs. Summarize current status across all projects.
  Flag anything stale, mismatched, or overdue.

**End-of-Session Discipline.** Before closing any working session, Claude updates the
Active Build Board: promote new flags, clear resolved items, roll the date forward.

**Living Document.** If something comes up that should be reflected here (a new project,
a new rule, a new rhythm), Claude proactively prompts: "This seems worth adding to the
Command Center. Want me to update it?" Then updates after I confirm.

---

## The Architect's Foundation

This is my north star. Every plan, decision, and priority gets filtered through this
lens. Claude's job is to keep my actions aligned with these commitments. If something
drifts, call it out.

### Vision
[What the world looks like if I succeed beyond my expectations. Long-horizon. Aspirational.]

### Mission
[My concrete 1-3 year goal. Specific. Measurable.]

### Values
[My 3-5 non-negotiables. The principles I will not compromise, even under pressure.]

### How to Apply
- Before helping plan any initiative, check it against the Vision.
- Before helping evaluate any opportunity, check it against the Mission.
- Before helping design any system or commitment, check it against the Values.
- If I am getting lost in tactics or busywork, pull me back to this level. That is the job.

---

## Active Build Board

Rolling dashboard. Claude reads this on "Scan" and updates it at the end of every
working session.

**Today: [DATE]**
- [To be filled in]

**Tomorrow: [DATE]**
- [To be filled in]

**Open Flags**
- [To be filled in]

---

## Active Projects Registry

| # | Project | Folder | Index File | Status |
|---|---------|--------|------------|--------|
| 1 | [Project name] | [Folder name] | 0) PROJECT-INDEX.md | [Active / Paused / Shipped] |

Add a row when a new project starts. Archive a row when a project ships.

---

## Operating Standards

### General Principles
- Simple scales. Complex fails.
- Leave the workspace better than you found it.
- Measure twice, cut once. If you are unsure, confirm before proceeding.
- Start simple. Add complexity only when needed.
- One change at a time. Explain your reasoning.

### AI-Specific Constraints
- NEVER guess on facts. If you are not confident, say so and look it up using web search
  before answering. I would rather wait 10 seconds for a verified answer than get a
  plausible-sounding wrong one.
- When in doubt, ask. Do not assume.
- Working output beats perfect documentation.

### Safety Boundaries
- NEVER delete files. If something needs to be removed, move it to `_to-be-deleted/`
  at the workspace root.
- NEVER overwrite existing files without asking first.
- Backups matter. If I ever disable my backup system (daily backup or cloud-sync), remind me and help me re-enable it.

### Session Hygiene
- Clean up scripts and artifacts that are no longer necessary.
- Close connections. Leave the environment ready for the next session.

---

## Recurring Tasks + Key Dates

- Weekly: [e.g., Friday Weekly Audit]
- Monthly: [e.g., Empty _to-be-deleted/ folder]
- Quarterly: [e.g., Verify cloud-sync is still active]

---

## Open To-Dos (Workspace-Level)

- [ ] Fill in the Architect's Foundation (when ready)
- [ ] Verify cloud-sync is running (revisit quarterly)

The PROJECT-INDEX.md Template

Save this as 0) PROJECT-INDEX.md at the root of every project subfolder. One per project. Required.

# [PROJECT NAME] Project Index

## Project Summary
[What this project is, in 1-3 sentences. Plain English.]

## Canonical Files
| File | Location | Description |
|------|----------|-------------|
| | | |

## Folder Structure
[Built out as the project develops.]

## Open Decisions
- [ ]

## Open To-Dos
- [ ]

## Session Log
| Date | Summary |
|------|--------|
| | |

## Notes
[Decisions, context, links, random thoughts.]

Create Your Context Files (The Brief)

Your CLAUDE.md tells Claude who you are. Context files tell Claude everything else it needs to produce specific, on-brand, on-target output. These live in the the-brief/ subfolder.

Context Files Overview

How to Build Your Brief

Use this prompt to generate starter templates for all 7 files at once:

Create a folder called the-brief inside my project folder. Generate these as separate markdown files with fill-in templates and helpful placeholder text:
1. icp.md (demographics, psychographics, pain points, desires, objections, language)
2. voice.md (tone words, sentence style, signature phrases, avoid list, 3 writing examples)
3. offer-stack.md (products, prices, transformations, value ladder)
4. audience-language.md (verbatim phrases from customers)
5. proof-points.md (testimonials, case studies, credentials)
6. competitive-positioning.md (competitors, differentiators, unique mechanism)
7. guardrails.md (topics to avoid, compliance, legal disclaimers)
Make each file easy to fill in. Include section headers and instructions for what to write in each section.

The Four Layers of Context (How They Stack)

You now have four places where you have given Claude information. Most people only configure one or two, then wonder why output quality is inconsistent. Understanding how all four layers work together prevents confusion, eliminates duplication, and makes every token count.

The Four Context Layers

The Decision Tree: Where Does This Go?

The 7 Mistakes That Will Undermine Your Setup

The Rule of Thumb

• ☐If it applies to every chat conversation: Personal Preferences.
• ☐If it applies to every Cowork session: Global Instructions.
• ☐If it applies to one project folder: CLAUDE.md.
• ☐If it is deep reference material for specific tasks: a context file in the-brief/.

Outcome-Based Prompting + Voice

This is the single most important concept in this entire guide. The way you give Cowork tasks determines 90% of your output quality. Stop writing prompts. Start writing task briefs.

Why Most People Prompt Wrong (and Get Diminished Results)

Here is the truth most people miss: the way most people prompt AI is completely wrong. Not a little wrong. Completely.

Most people think dictating a lot of HOW the AI should do the thing (the method, the steps, the process) is what makes a good prompt. That is actually the worst thing you can do with AI. When you micromanage the method, you are capping the AI's ability to optimize for the actual result you want.

Outcome-Based Prompting flips the script. You tell AI the ultimate outcome you want. What does the finished result look like? What does it accomplish? Who is it for? Then you add the guardrails: the constraints, the voice rules, the formatting requirements, the things to avoid. That is the architecture.

Think of it this way: "I want to sit down at my desk, look at this screen for an hour a day, and handle everything that needs to be handled with zero friction." That is an outcome. You do not care if the solution is built in React or HTML. You do not care about the technical method. You care about the result. When you communicate that clearly to AI, it can actually optimize for the best path to get there.

This changes everything about how you interact with Cowork. You are not typing step-by-step instructions into a chatbot. You are describing the finished state of the world you want to see, and letting an AI operator figure out the most efficient path to deliver it.

The Wrong Way: Step-by-Step Micromanagement

"First, open the file. Then read line 1. Then copy it. Then open a new document. Then paste it. Then format it as bold."

This is how people talk to chatbots. You are doing the thinking. Claude is just executing keystrokes.

The Right Way: Outcome-Based Delegation

"Create a competitive analysis report comparing my product to Competitor A and Competitor B. Research both online. Analyze their pricing, features, positioning, and weaknesses. Create a professional Word document with an executive summary, comparison table, and strategic recommendations. Save it to outputs/."

Tell Cowork WHAT you want. Describe the finished product. Let it figure out the HOW.

The 4 Elements of a Good Task Brief

Agentic Task Brief Template

Objective: [business outcome you need]
              Inputs: [files/folders/URLs Claude should read first]
              Deliverables: [exact artifacts + formats + count]
              Quality bar: [voice, proof, guardrails, formatting standards]
              Constraints: [length, tone, forbidden claims, what NOT to do]
              Review gate: Propose your plan + filenames first. Wait for my approval before executing.
              Output: Save to /outputs using YYYY-MM-DD-type-description.ext

The "Before You Start" Technique

For important tasks, add this to your prompt:

Before you start creating anything, tell me:
              (1) Your understanding of what I am asking for
              (2) Your planned approach
              (3) Any assumptions you are making
              (4) Questions you have before proceeding

This prevents wasted work and ensures alignment. Use it for any task that would take more than 15 minutes to redo.

Supercharge Outcome-Based Prompting with Voice (Wispr Flow)

Here is something most people do not realize: you think faster than you type. Outcome-Based Prompting is about describing the result you want in natural language. You know what is even more natural than typing? Speaking.

Wispr Flow is a voice-to-text tool that captures your natural speech and turns it into polished, professional text instantly. Instead of typing a detailed task brief into Cowork, you speak it. Flow transcribes it, cleans it up, and drops it right into the Cowork input field. The result: you are 4x faster than typing, and your prompts are actually better because you naturally describe outcomes when you speak (instead of getting bogged down in step-by-step typing).

Setting Up Wispr Flow (15 Minutes)

1. ☐Go to travishouston.com/wispr and download Wispr Flow for free.
2. ☐Install the app on your Mac or Windows machine. Follow the setup prompts.
3. ☐Grant microphone access when prompted.
4. ☐Open Cowork. Click into the message input field.
5. ☐Activate Wispr Flow (default hotkey or click the menu bar icon).
6. ☐Speak your task brief naturally. Describe the outcome you want, the context, the quality bar, and where to save it.
7. ☐Wispr Flow transcribes and polishes your speech into clean text in the Cowork input field.
8. ☐Review the text (quick scan), then hit Enter to send it to Cowork.

When to Use Voice vs. Typing in Cowork

The pattern: Speak the big-picture outcome. Type the precise details. This combination is the fastest, most effective way to work with Cowork, and it reinforces Outcome-Based Prompting naturally because your voice defaults to describing results, not dictating steps.

Download Wispr Flow free: https://travishouston.com/wispr
              Setup time: 15 minutes
              Works with: Claude Cowork, any text input field on your computer
              Best for: Task briefs, context descriptions, feedback, brainstorming

Run Your First Real Tasks

Task 0: The Verification Test (5 minutes)

I just set up Cowork and want to verify everything is working. Please:
(1) Confirm you can access my working folder and list all files/folders you see.
(2) Read my CLAUDE.md and summarize who I am in one paragraph.
(3) Create a test file called cowork-test.md in my outputs/ folder with a summary of your capabilities.
(4) Tell me what connectors, skills, and tools are available in this session.
(5) Confirm the writing style matches my Global Instructions preferences.

What Success Looks Like

• ☐Claude lists your folder structure correctly.
• ☐Claude accurately summarizes who you are from your CLAUDE.md.
• ☐A new file appears in your outputs/ folder.
• ☐The writing style matches your tone preferences.

Tasks 1-3: Real Work (Not Dummy Projects)

• Task 1: File Operation. Organize a folder, extract data from a document, or compile research into a structured deliverable.
• Task 2: Professional Document. Create a report, proposal, email series, content piece, or sales sheet.
• Task 3: Content Multiplication. Take one piece of existing content and have Cowork produce 10-15 platform-specific variations (blog, LinkedIn, Twitter, email, Instagram captions).

Install Your First Skills

Skills are instruction files that trigger entire workflows with one command. Instead of typing a detailed prompt every time, you write the instructions once and save them. Type the name. Claude follows the entire workflow automatically. This is where Claude goes from answering questions to actually doing work.

How to Install a Skill

No restart needed. No configuration. Upload the file and start using it.

When to Build a Skill (vs not)

Build: You do the same task more than twice a week. The task has specific steps you want followed every time. You find yourself typing the same long prompt repeatedly. You want consistent output format. You want someone else to be able to do the task with one command.

Don't: It is a one-time task. The task changes significantly every time. It is simple enough that a quick prompt handles it fine.

The test: "Am I going to explain this to Claude the same way again next week?" If yes, build a skill.

10 Essential Skills to Consider

Connect Your Tools (Connectors)

Connectors let Claude access your actual business tools directly. No copying. No pasting. Direct access.

How to Connect

1. Go to Customize > Connectors (or Manage Connectors).
2. Browse available connectors.
3. Click Connect next to the tool you want.
4. Follow the OAuth sign-in flow (log into the tool, grant permission).
5. Test with a simple task.

Recommended Connection Order

Permission Safety Settings

Plugins + Browser Automation

What Plugins Are

Plugins bundle skills, connectors, sub-agents, and slash commands into complete toolkits. Instead of setting up five separate pieces, you install one plugin and get everything pre-integrated from the first conversation.

The Cowork plugin marketplace has become the fastest-growing part of the ecosystem. The public marketplace carries plugins for sales, marketing, finance, legal, HR, engineering, design, operations, data analysis, and more. Team and Enterprise plans can also stand up private marketplaces stocked with org-specific plugins that auto-install for new team members.

How to Install a Plugin

1. Open Claude Desktop and switch to the Cowork tab.
2. Click Customize in the left sidebar. This is where plugins, skills, and connectors live.
3. Click Browse plugins to see the marketplace.
4. Click Install on the plugin you want. Installed plugins save locally to your machine.
5. To use a skill from an installed plugin inside a session, type "/" or click the "+" button to see available skills.

Plugin Categories

• Personal Productivity: Task management, scheduling, inbox workflows.
• Sales: Lead research, email sequences, deal analysis.
• Marketing: Content creation, campaign planning, multi-channel repurposing.
• Data Analysis: Report generation and visualization.
• Operations: Capacity planning, compliance tracking, change requests.
• Legal, HR, Finance: Document generation, contract review, compliance workflows.
• Engineering and Design: Code review, design ops, spec generation.

Building Your Own Plugin

1. Go to Customize > Plugins > Browse Plugins. Find and install Plugin Create.
2. Open a new session and tell Plugin Create what you want to build.
3. Plugin Create walks you through the setup: slash commands, connectors, customization options, sub-agents.
4. Review the generated files. Make edits to the SKILL.md files.
5. Tell Cowork to install the plugin. To distribute across a team, publish it to a private marketplace repo (private or internal GitHub repos only; public repos are not allowed for organization marketplaces).

Computer Use + Browser Automation

Claude can now control your computer directly from inside Cowork. This is the final fallback in the priority chain: native connector > MCP server > browser automation > computer use (clicking, typing, opening apps like you would). Use it for tools that have no connector, no MCP, and no Chrome extension. Claude takes screenshots to navigate and asks permission before touching each app.

Enabling Computer Use in Cowork

1. Update Claude Desktop to the latest version.
2. Open Settings > General.
3. Toggle on Computer Use.
4. Open Cowork. Start assigning tasks. Claude will prompt for permission each time it needs to interact with a new app.

Browser Automation (Claude in Chrome)

For browser-specific tasks (web scraping, form filling, navigating web apps that do not have connectors) you can also use the Claude for Chrome extension. This is lighter-weight than full Computer Use for browser-only workflows.

1. Install Chrome if you have not already.
2. Go to the Chrome Web Store. Search "Claude for Chrome." Install the extension.
3. In Cowork Settings, enable browser-related permissions.
4. In Customize > Connectors, ensure "Claude in Chrome" is enabled.
5. Test: "Open my browser, go to YouTube, and tell me the top 3 recommended videos on my home screen."

Scheduled Tasks + Dispatch

This is where Cowork stops being a tool you use and starts being an employee that works while you sleep.

Creating a Scheduled Task

1. Open Cowork. Navigate to your project folder.
2. Run the task manually first. Make sure it works and produces the output you want.
3. Click the schedule icon (or type /schedule).
4. Configure: what to run, when, how often, where to deliver results.
5. Monitor the first 2-3 automatic runs to verify reliability.

The Morning Automation Stack (5 Staggered Tasks)

You open your outputs/ folder Monday morning and your entire workday is already planned. Content is pre-written. Follow-ups are drafted. Metrics are summarized. Priorities are clear.

6 Automated Workflow Ideas

Dispatch: Trigger Cowork From Your Phone

Dispatch lets you text a task request from your phone. By the time you sit down at your desk, the result is waiting in your outputs/ folder.

1. Enable Dispatch in Settings > Cowork > Dispatch.
2. Provide the phone number you will text from. Confirm it.
3. Set a destination for results (outputs folder, email, or Slack).
4. Test: Text "What is my calendar for today?" from your phone. Confirm the result appears.

Dispatch use cases: Meeting prep on the go. Inbox sweep. Quick research. Slack catch-up. Expense report from receipts. Any skill you have built can be triggered via Dispatch.

Your Daily Workflow Pattern

This is the most underestimated section in this entire guide. Most people skip it because it looks simple. Three commands. Start, work, stop. But this pattern is not about commands. It is about solving the single biggest limitation of AI agents: they forget everything between sessions.

The Problem This Solves

Claude does not remember what happened yesterday. Every new Cowork session starts from zero. Claude will re-read your CLAUDE.md and Brief files (those are persistent), but it has no idea what you worked on last session, what is still open, what decisions you made, what got completed, or what the next logical step is. Without a system to bridge that gap, every session feels like onboarding a new employee who has amnesia.

The daily workflow pattern creates a manual persistence layer. You tell Claude to write down what happened. Tomorrow, Claude reads that file back. Now it has continuity. Over days and weeks, those logs stack into a running operational record that makes Claude dramatically more useful, faster, and more aligned with where your actual work stands.

The Three-Phase Daily Pattern

The Morning Boot Prompt (Copy and Customize)

Add this instruction to your CLAUDE.md so Claude does this automatically whenever you start a session:

# SESSION START BEHAVIOR
              When I say "Good morning" or start a new work session:
              1. Read CLAUDE.md and any updated context files.
              2. Check for the most recent daily log in logs/.
              3. Summarize: what was completed last session, what is still open, and what the recommended focus is for today.
              4. Ask me if priorities have changed before diving in.

The End-of-Day Wrap Prompt (Copy and Customize)

This is the prompt that creates the persistence file. Use this exact format or add it to your CLAUDE.md as a standing instruction:

Before we wrap up for the day, create a daily log and save it to logs/YYYY-MM-DD-daily-log.md with this structure:

              # Daily Log: [Date]

              ## Completed Today
              - [List everything we accomplished this session with file names and locations]

              ## Still Open
              - [List every task that is in progress or was discussed but not finished]

              ## Decisions Made
              - [List any decisions, direction changes, or approvals given during the session]

              ## Tomorrow's Focus
              - [The single highest-leverage task to start with tomorrow]
              - [2-3 supporting tasks in priority order]

              ## Notes
              - [Anything Claude flagged as uncertain, needs verification, or requires my input]

Why This Pattern Compounds

Day 1, the daily log is a simple list. Day 5, Claude has a running record of your week. Day 30, Claude has a full operational history of your project: what was built, what changed direction, what decisions were made, and what patterns keep emerging. This is not journaling. This is building an operational memory that makes every future session faster and more accurate.

Folder Structure for Logs

Create a logs/ folder inside your project directory. This is where all daily logs live. Name them with the date prefix so they sort chronologically:

your-project-folder/
              CLAUDE.md
              the-brief/
              outputs/
              logs/                    <-- Daily logs go here
              2026-04-01-daily-log.md
              2026-04-02-daily-log.md
              2026-04-03-daily-log.md
              2026-04-04-daily-log.md

Common Mistakes With the Daily Pattern

Name Your Sessions (The Findability System)

Claude auto-names every chat and Cowork session with something like "Architected dual-system framework distinguishing relational versus operational Claude programming." That is a 12-word academic description that tells you nothing useful 3 weeks later when you are scrolling your sidebar looking for the session where you built your email sequence. Multiply that by 10 to 20 sessions per week, and your sidebar becomes an unsearchable wall of AI-generated word salad.

The fix takes 5 seconds: rename every session the moment you finish (or during your end-of-day wrap). Use this naming convention so every session is findable by when it happened, what type of work it was, and what was inside.

The Naming Convention

FORMAT:  MM-DD [CATEGORY] Descriptive Topic Name
              WIP:     WIP: MM-DD [CATEGORY] Descriptive Topic Name

              EXAMPLES:
              04-05 [Content] Email Sequence for Workshop Launch
              04-05 [Research] Competitor Pricing Analysis
              04-03 [Strategy] Q2 Offer Stack Redesign
              04-02 [Admin] Weekly Audit Report
              04-01 [Build] Sales Page Copy Draft v2
              WIP: 04-04 [Client] Sarah - Funnel Architecture Review

The 7 Categories

How to Rename

• In Cowork (Desktop App): Click the session title at the top of the chat window or hover over it in the left sidebar. Click the pencil icon or double-click the name. Type your new name using the convention above.
• In Claude Chat (Web): Hover over the conversation in the left sidebar. Click the three-dot menu or pencil icon. Rename using the same convention.
• When to rename: During your end-of-day wrap or the moment you finish a session. Do not wait. If you wait until Friday, you will not remember what "Synthesized multi-channel attribution framework" was about.

Advanced: The Weekly Audit (Every Friday)

The daily pattern handles session-to-session continuity. The weekly audit handles system-level health. Run this every Friday to keep your entire Cowork setup current:

Audit my Cowork setup. For each area, tell me what is working, what needs fixing, and your top 3 recommended improvements:
              1. Is my CLAUDE.md current and accurate?
              2. Are my context files (the-brief/) up to date?
              3. Which skills have I used this week? Which are unused?
              4. Are all connectors still active and authorized?
              5. Have my scheduled tasks run successfully this week?
              6. What is one workflow I do manually that should become a skill?
              7. Review my daily logs from this week: what patterns do you see in my work? Any recurring blockers?

              Save the audit report to outputs/ as weekly-audit-[date].md.

Verification: Is It Working?

1. Run the daily pattern for 3 consecutive days (morning boot, work, wrap-up).
2. On day 4, say "Good morning" and see if Claude references specific work from the previous days.
3. If Claude summarizes your open items and yesterday's completions without you explaining anything, the system is working.
4. If Claude responds generically ("How can I help you today?"), your daily logs are not being saved to the right location, or your CLAUDE.md does not include the session start behavior. Fix both.

The Post-Install Checklist + 90-Day Roadmap

The Cowork Method Accelerator is complete when every item below is confirmed. Print this page. Check every item off. Do not move to advanced features until your foundation is solid.

Post-Install Checklist

• ☐Workspace folder is protected by at least one backup system (daily computer-wide backup like Time Machine or Backblaze, OR optional cloud-sync via Dropbox, iCloud Drive, Google Drive, or OneDrive).
• ☐Claude Desktop app is installed, updated, and signed in with a paid account.
• ☐Privacy settings configured: opted out of model training.
• ☐Cowork tab is visible and functional.
• ☐You understand when to use Opus (brand intelligence work), Sonnet (simple tasks), and Haiku (speed-critical batch work). Extended Thinking confirmed as on for strategic sessions.
• ☐Global Instructions saved with your name, tone, business context, rules, and file preferences (including the "Never guess on facts" rule and the "_to-be-deleted/ instead of delete" rule).
• ☐Workspace folder exists with correct internal structure (CLAUDE.md, 0) COMMAND-CENTER.md, .claude/commands/, the-brief/, project folders, docs/, outputs/, data/, inputs/, _to-be-deleted/).
• ☐Folder connected to Cowork with "Always Allow" access.
• ☐CLAUDE.md saved at root. If Pattern B, it is a thin boot loader pointing to the Command Center.
• ☐0) COMMAND-CENTER.md saved at workspace root with Architect's Foundation, Active Build Board, Active Projects Registry, Operating Standards, and Recurring Tasks blocks.
• ☐Scan and Full Scan keyword triggers tested and working.
• ☐Each active project has its own subfolder and its own 0) PROJECT-INDEX.md with Session Log table.
• ☐At least 3 context files in the-brief/ (icp.md, voice.md, offer-stack.md minimum).
• ☐Personal Preferences configured in claude.ai > Settings > Profile with all 8 sections.
• ☐You understand the 4 layers of context (Personal Preferences, Global Instructions, CLAUDE.md, The Brief) and what belongs in each.
• ☐You can write an Outcome-Based task brief (ATB format).
• ☐Verification test task completed with output in outputs/.
• ☐3 real work tasks completed successfully.
• ☐At least 1 skill installed and tested.
• ☐At least 1 connector active (Google Calendar recommended first).
• ☐Connector permissions reviewed and set appropriately (read=allow, send=approval, delete=blocked).
• ☐End-of-session discipline active: Claude updates Active Build Board and Session Logs before closing.
• ☐Daily workflow pattern active: morning boot, work session, end-of-day wrap. logs/ folder created.
• ☐You understand how Scheduled Tasks work and have set up at least 1.
• ☐Dispatch enabled and tested from your phone.
• ☐Weekly audit scheduled for every Friday. Backup verification (daily backup or cloud-sync, whichever you use) added to quarterly rhythm.

The 90-Day Implementation Roadmap

Phase 1: Foundation (Days 1-7)

• Days 1-2: Install and Configure. Install the Claude Desktop app. Open Cowork. Create your first project folder. Set up Personal Preferences in claude.ai (8 sections). Set up Global Instructions in Cowork (10 sections). Opt out of model training. Total time: 45 minutes.
• Days 3-4: Build Your Context Layer. Write your CLAUDE.md file. Write three starter context files: icp.md, voice.md, offer-stack.md. These do not need to be perfect. They need to be real. Total time: 2 hours.
• Days 5-6: First Real Tasks. Complete three real work tasks: one file operation, one professional document, one research deliverable. Use real work, not dummy projects.
• Day 7: Benchmark. Time yourself completing a task the old way. Time yourself completing the same task with Cowork. Document the difference. This number becomes your anchor.

Phase 2: Acceleration (Days 8-30)

• Week 2: Skills. Install the 3 essential skills (Context Builder, Cowork Visualizer, Weekly Audit). Build your first custom skill from a workflow you run repeatedly.
• Week 3: Connectors. Connect Google Calendar, Gmail, and one more connector. Test each with a real task. Run your first cross-platform workflow.
• Week 4: Plugins. Install 2 official plugins that match your role. Customize them with your context files. Build one custom plugin bundling 2+ skills.

Phase 3: Optimization (Days 31-60)

• Scheduled Tasks. Set up 3 scheduled tasks (morning brief, inbox triage, weekly pulse). Enable and test Dispatch from your phone.
• Weekly Audit. Run the weekly audit skill every Friday to keep your system clean. Build compound workflows that combine skills + connectors + browser automation.

Phase 4: Scale (Days 61-90)

• Calculate ROI. Hours saved per week. Quality improvement. Revenue impact. Document your complete system (folders, skills, connectors, scheduled tasks).
• Document and Share. Create SOPs for team members. Identify the next 5 workflows to automate. Share your system with your team.

Phase Checkpoints

Troubleshooting: Quick Fixes

Every issue below has a clean fix. None of these are reasons to abandon the system. They are normal friction points every new Cowork user runs into in the first two weeks. Work through them once and they never come back. This is why Cowork is the architecture I trust to write a full book in 5 days. The fixes are small. The compounding output is massive.