Step 1
Open Cowork
Cowork lives inside Claude Desktop, which you already have on your computer. Here is how to find it.
One rule to remember: Claude Desktop needs to stay open the whole time Cowork runs a task. Think of it like leaving someone in charge of the office. They can only work while they are still in the building.
Step 2
Set Up Permissions
Cowork needs two kinds of access to do its job. You control both. Think of this like handing over keys to your office. You decide which doors Cowork can open, and you are responsible for what happens behind those doors. Take your time here. The permissions you set shape how much power Cowork has.
Permission 1: Network Access
By default, Cowork can run code and create files, but its network access is limited. You can open that up so Claude can install packages, pull data, and reach the web during tasks.
Permission 1b: Web Browsing (Chrome Extension)
If you want Claude Cowork to actually browse the web for you, there is one more step. You need to install the "Claude for Chrome" extension. This is currently in beta. It gives Claude access to your Chrome browser so it can surf the web, visit pages, and pull information in real time.
Permission 2: Connectors (Your Cowork Power-Ups)
Connectors let Cowork talk to other apps. Google Drive, Gmail, Slack, DocuSign: each one is a connector. Install the connector and Cowork can reach into that app. Skip the connector and Cowork works with local files only. Simple as that.
What Cowork Can Actually Do
Cowork runs inside a virtual machine (VM) on your computer. A VM is a protected space where code runs safely. Claude works inside that space, which keeps things contained. Here is the key thing to understand: Claude CAN make real changes to the files you give it access to. And according to Anthropic, you are responsible for every action Claude takes on your behalf. That includes content it publishes, messages it sends, files it changes, and purchases it makes.
Before Cowork permanently deletes any file, it will ask you first. You will see a popup with an Allow button. Read what it says before clicking.
During any task, you can see what Claude is doing in real time. Jump back in and redirect it whenever you want. Keep an eye out for these patterns:
- Claude accessing files you did not mention
- The task growing beyond what you asked for
- Claude asking for access to something unexpected
If you see any of those, stop the task and redirect.
Step 3
Give Cowork a Folder
Here is the step that separates great results from confusing ones.
A folder gives Cowork a clean workspace with clear edges. It knows where to look and where to put things. Inside this folder, you will add multiple files that tell Claude everything it needs to know about you, your business, your frameworks, your target audience, and how you like to work. The more context you give, the better the results.
Picture setting up a desk for a new helper. A clean, well-stocked desk with a few good notes on it means they hit the ground running. A bare desk with nothing on it means they spend their first hour guessing.
How to Create Your Cowork Folder
Cowork Brain or Claude Workspace.You Can Point Cowork at Any Folder
Cowork can work inside any folder on your computer. Each time you start a new session, you choose which folder to open. This means you can have your main Cowork Brain folder for everyday work, and still point a session at a different project folder when you need to.
This is useful when you want Claude to work directly inside a client project, a code repository, or a shared team folder. You stay flexible.
Step 4
What to Put in the Folder
Your folder does the heavy lifting of giving Claude long-term memory. Since Cowork starts fresh each session, your folder is how you carry context forward. Think of it as the briefing packet waiting on Claude's desk every morning.
Create simple text files and drop them in. Here is exactly what to include and why each one matters.
File 1: Who I Am
Create a file called ABOUT-ME.txt. Write 10 to 15 sentences about yourself. Cover your name, your business name, what you sell, who your clients are, and how you like to communicate. Keep it plain. You are writing for Claude, and Claude reads fast.
File 2: My Products and Services
Create a file called MY-OFFERS.txt. List each thing you sell, its name, what it does, who it is for, and the price. Claude uses this any time you ask it to help with proposals, content, or client work. With this file in place, Claude knows your business before you say a word.
File 3: My Current Projects
Create a file called ACTIVE-PROJECTS.txt. Write two to five sentences about each active project. Include client names, what stage it is at, and key deadlines. Update this once a week. This is the file that lets Cowork pick up right where you left off.
File 4: How I Like Things Done
Create a file called MY-PREFERENCES.txt. Write the rules Claude should follow every time. Examples: save output as a PDF, use my brand voice, add bullet points only when I ask, always use YYYY-MM-DD date format, save finished files in the Output subfolder.
This is the file that turns Claude from a capable stranger into your actual assistant.
File 5: Writing Samples (Optional but Worth It)
Drop in two or three examples of your best writing. Past emails, proposals, or posts all work well. Name them SAMPLE-1.txt, SAMPLE-2.txt, and so on. Claude uses these to match your voice in anything it writes for you.
Drop Your Existing Files In Too
If you have already been building your context library, add copies of those files to this folder. Your Business Context document, your Target Audience profile, even your Second Brain. Anything that helps Claude understand you, your business, and the people you serve belongs here.
The more context Claude has, the sharper every output becomes. These files turn a general-purpose AI into an agent that sounds like you and thinks about your audience.
Want to learn how to build a Business Context doc, Target Audience profile, and Second Brain from scratch? We teach that inside our Patreon.
Step 5
Set Global Instructions
Global instructions are rules that apply to every Cowork session across every folder. Use this for things that should always be true, no matter what you are working on.
Step 6
Driving Cowork
Your folder is loaded. Your permissions are set. Your global instructions are saved. Now it is time to actually use Cowork. There are two ways to talk to it: plain English and slash commands.
Plain English: Your Primary Tool
The main way to use Cowork is to describe what you need in your own words. Start a new task, type your request, and Claude figures out the plan. It breaks the work into steps, runs them inside the VM, and delivers the output to your folder. You can watch every step in real time and jump in to redirect whenever you want.
Example prompts
Slash Commands: Shortcuts for Structured Work
Slash commands are shortcuts you trigger by typing / in the chat box (or clicking the + button). Each one opens a structured form where you fill in the details, and Claude takes it from there.
Cowork comes with one built-in slash command:
/scheduleCreates a recurring or on-demand task. Set it to daily, weekly, weekdays only, or any frequency you want. This is where Cowork becomes a system instead of just a tool.
Every other slash command comes from plugins. When you install a plugin (sales, marketing, finance, engineering, and more are available), it adds its own slash commands to your menu. Type / to see every command available from your installed plugins. Each one launches a form so running a workflow like "generate report" or "create dashboard" feels as simple as filling out a brief.
How a Typical Session Works
/ to pick a slash command from your installed plugins.Stay Safe: What You Need to Know
Cowork is powerful. That power comes with real responsibility. Anthropic is upfront about this: Cowork is a research preview, and the chance of something going wrong is, in their words, "non-zero." Here is how to protect yourself.
The Golden Rule: You Own Every Action
Anthropic says it clearly: you are responsible for all actions Claude takes on your behalf. That covers content it publishes, messages it sends, files it edits or deletes, purchases it makes, and data it accesses. If Claude does it while logged into your accounts, it is treated the same as if you did it yourself.
This applies to both live tasks (where you are watching) and scheduled tasks (where Claude works on its own). The more autonomy you give Cowork, the more important it is to set clear boundaries.
7 Safety Rules From Anthropic
These come directly from Anthropic's official safety guide. Follow them.
1. Be selective about file access
Keep sensitive files away from Cowork. That means financial documents, login credentials, personal records, and anything regulated. Create a dedicated working folder and only share what Claude needs for the task at hand.
2. Monitor tasks, not just commands
Watch what Claude actually does during a task. Look for unexpected patterns: files you did not mention being accessed, the task growing beyond what you asked for, or Claude requesting access to something surprising. If you see any of that, stop the task.
3. Be cautious with scheduled tasks
Start with simple, low-risk tasks. Avoid scheduling anything that touches sensitive data, sends messages, or makes purchases. Review the output after every run. Pause or delete tasks you are no longer using.
4. Limit web access to trusted sources
Web content is the primary way bad actors could try to manipulate Claude through prompt injection. When Cowork browses the web, stick to sources you know and trust.
5. Be careful with plugins and custom connectors
Stick to verified extensions from the official Claude directory. Custom connectors connect to third-party services that Anthropic has NOT reviewed. Read every permissions screen before clicking approve.
6. Be mindful of cross-app data sharing
Data can flow between apps during a session. If Claude is connected to both your email and your spreadsheets, information from one can end up in the other. Think about which connectors are active before starting a task.
7. Report suspicious behavior immediately
If Claude does something unexpected or concerning, report it to Anthropic at [email protected]. They take this seriously.
What Anthropic Has Built In
Anthropic has built safety measures into Cowork to reduce risk. These are running in the background:
Model training: Claude is trained to recognize and refuse malicious instructions, even when they are hidden inside web pages or files.
Content classifiers: Automated systems scan untrusted content for prompt injection attempts before Claude processes it.
Deletion protection: Claude asks for your explicit permission before permanently deleting any file.
Supercharge Cowork With Skills
Skills are reusable instruction sets you can add to Cowork. Think of them as playbooks. Each skill teaches Claude a specific way to work, like writing in your brand voice, following a content framework, or running a research process step by step.
What Skills Do
When you add a skill, you give Claude a set of instructions it can follow every time it works on a certain type of task. Instead of explaining your process from scratch each session, the skill handles it. Claude reads the skill, follows the steps, and delivers work that matches your standards.
You can have skills for content writing, client onboarding, competitor research, social media scheduling, email sequences, and more. Each one lives as a file in your Cowork folder.
How to Add a Skill
.txt or .md file with a clear name, like SKILL-BLOG-WRITER.txt.The Skills Masterclass
Want to see exactly how to build skills that turn Cowork into a full operating system for your business? We built a masterclass that walks you through it step by step. Real examples. Real frameworks. Ready to use.
Get Instant AccessWhat You Have Built
The Big Picture
You now have a personal agent that knows who you are, what you do, and how you work. That changes everything about what is possible.
You Have an Agent. Now What?
Before this setup, Claude was a chatbot. You typed a question, it gave an answer, and the conversation ended. Now you have something different. You have an agent with memory, context, file access, internet access, and the ability to run tasks on a schedule while you sleep.
This opens up entirely new playbooks. Workflows that were only possible with expensive custom software or a full team can now run from a single folder on your computer. And anything the Claw can do? Cowork can handle it too. We will show you how to outclaw the claw.
The key is this: the quality of your folder determines the quality of your agent. The context files you built in this guide are the foundation. Every new skill, every new workflow, every new automation builds on top of what you just created.
Content Control Center
Imagine an agent that already knows your brand voice, your audience, and every platform you publish on. You drop in a single piece of content. The agent reads it, analyzes it against your style guide, and produces ready-to-publish versions for every channel you care about. Blog post, LinkedIn update, email newsletter, social threads. Formatted. On-brand. Instant.
That is the power of agents with context. One input, many outputs, all shaped by the files sitting in your folder right now. This is just one example. The playbooks are endless.
The Foundation is Set
Folder built. Instructions loaded. Connectors in place. Scheduled tasks running. You now have a complete Cowork system ready to work for you.
If Things Go Wrong
Here are quick fixes for the situations you are likely to run into.
Cowork tab is missing
Make sure you have the latest version of Claude Desktop. Update the app and restart it.
Scheduled task did not run
Check that Claude Desktop was open and your computer was awake at the scheduled time. Cowork catches up on missed tasks the next time you open the app.
Claude is giving generic results
Your folder files likely need more detail. Go back to Step 4 and add more context to ABOUT-ME.txt and MY-PREFERENCES.txt.
Connector stopped working
Go to Settings > Connectors, remove the connector, and reinstall it. This refreshes the secure connection.
What Comes Next
Ready to Outclaw the Claw?
You just built the foundation. Now it is time to put agents to work for your vision and your business.
Inside our Patreon, we are building agents and workflows with Cowork that do everything OpenClaw can do. Full control. Same power, safer path.
And once your setup is dialed in? We will show you how to go beyond what the Claw ever could.