Mastering Claude Co-Work
Course
Mastering Claude Co-Work
Module 2 of 18

Setup That Actually Sticks

Configure Co-Work once, correctly — global instructions, safety guardrails, and the Business Brain that makes every task better.

What you'll learn

Describe the purpose of global instructions and the no-delete guardrail in your own words
Identify the four components of a complete initial configuration: profile, global instructions, sandbox folder, and Business Brain
Configure global instructions that include business context, safety guardrail, and output preferences
Explain why a Business Brain document improves output quality without increasing per-task effort

Configure Once, Benefit Forever

Annotated global instructions block showing three required zones: business context, safety guardrail, output preferences

Most Co-Work setups fail the same way: people start using it immediately, run a few tasks, get mediocre results, and conclude the tool is not as good as advertised. The problem is almost never the model. The problem is the setup.

Co-Work reads global instructions before every single task it executes. That means the quality of your global instructions has a compounding effect on every task you ever run. A thirty-minute investment in getting this right will improve thousands of future tasks. That is the highest-leverage thirty minutes in this entire course.

This module produces four concrete outputs: global instructions, a sandbox folder structure, a Business Brain document, and the dynamic reference that connects them. When you finish, Co-Work should be able to answer "Tell me about my business" accurately and specifically — without you typing a single word of context.

Global Instructions: Your Standing Orders

Global instructions are persistent context that Co-Work loads before every session, every task, every scheduled run. They are not conversation starters. They are permanent operating parameters — the equivalent of the onboarding document you give every new employee, except this employee reads it every single morning.

Three things belong in global instructions, and only three things. Anything else should live in a folder-level file (covered in Module 3) or a Business Brain document (covered below).

1. Business Context

Who you are. What you do. Your primary tools and platforms. The lens Co-Work should use to interpret every task you give it. This should be three to five sentences — specific enough to be useful, brief enough that it does not dominate the context on every task.

Example: "I run a boutique UX consulting firm. My clients are mid-market SaaS companies. My primary tools are Notion for documentation, Linear for project tracking, and Google Workspace for communication. I work alone with two part-time contractors."

2. Safety Guardrail (Non-Negotiable)

This is not optional. Co-Work has filesystem access. It has email access once you connect Gmail. It can take actions that cannot be undone — deleting files, sending messages, creating calendar events. Without a clear safety guardrail in your global instructions, Co-Work will execute these actions without asking.

Add This Before Your First Real Task

Copy this exact text into your global instructions, then customize: "Never delete files without explicit written confirmation from me. Never send emails, messages, or calendar invites without my explicit approval. Always show me a preview before any action that cannot be undone. When in doubt, ask." Brooke Wright, Jack Roberts, and Paul all include a version of this — three independent practitioners who learned the same lesson.

3. Output Preferences

How you want information formatted by default. Bullet points or paragraphs? Detailed analysis or executive summary? Include sources or skip them? These preferences apply everywhere — you can override them per task, but having sensible defaults means most tasks produce useful output without adjustment.

The No-Delete Guardrail in Practice

Here is what happens without it: you ask Co-Work to "clean up my Downloads folder." Co-Work interprets "clean up" as "delete duplicates and old files." It deletes 847 files. Some of them were things you needed.

Here is what happens with it: Co-Work says "I found 847 files I could delete. Here is a summary by category. Confirm to proceed." You review. You confirm. Nothing irreversible happens without your knowledge.

The guardrail does not prevent Co-Work from doing the work. It prevents Co-Work from completing the work without your sign-off. That distinction is the entire safety model for the first three tiers of this course.

The Business Brain Document

Business Brain document structure showing labeled sections: About, ICP, Products and Services, Brand Voice, Key Processes, Team
Framework Disclosure — Business Brain

Business Brain is a practitioner-coined term by Jack Roberts. It is not an official Co-Work feature name. The underlying capability — a markdown file with business context, dynamically referenced rather than hardcoded into every prompt — is real and verified. The name is Jack's.

Your global instructions give Co-Work a brief orientation. Your Business Brain gives it a complete briefing. The document contains everything Co-Work would need to know to represent your business accurately: who your customers are, what you sell, how you communicate, what your processes look like, who is on your team.

The key insight is dynamic loading. You do not paste the Business Brain into every task prompt. You add one line to your global instructions: "If I ask any business question, check ~/CoWork/Business/business-brain.md before responding." Co-Work reads it on demand — only when it needs it. This is far more token-efficient than hardcoding business context into every prompt.

Framework Disclosure — Token as Employee Cost

Jack Roberts frames token spend this way: "Think of tokens as employee cost, not a SaaS subscription." This is a practitioner framing, not official Anthropic guidance. The practical implication — that token efficiency matters and dynamic loading is smarter than hardcoding — is sound regardless of framing.

What Goes in a Business Brain

Structure the document with these six sections as headers. Fill in whatever is real for you — partial information is better than none:

  • About [Your Business] — one paragraph overview, founding context, mission
  • Ideal Customer Profile — who you serve, their problems, what they have tried before
  • Products / Services — what you offer, pricing (if comfortable), positioning
  • Brand Voice and Tone — adjectives that describe your communication style; examples
  • Key Processes and Tools — how you work, what stack you use, recurring workflows
  • Team Members and Roles — names, roles, decision authority

The document does not need to be perfect on day one. Add to it as you use Co-Work. When a task produces output that seems off-brand, add a clarification to the Business Brain and re-run. Over months, it becomes genuinely comprehensive.

The Sandbox Folder

Folder tree: CoWork root with Inbox, Output, Skills, and Business subfolders, with business-brain.md inside Business

Before Co-Work can write files, save outputs, or organize anything, it needs a home base — a folder it has explicit permission to work within. Create this now, before you connect any integrations or run any real tasks.

Create ~/CoWork/ as your root. Inside it, create four subfolders:

  • Inbox/ — things Co-Work receives or monitors
  • Output/ — drafts, reports, generated files
  • Skills/ — skill markdown files (used in Module 6)
  • Business/ — Business Brain and other reference docs

Give Co-Work explicit permission to read and write within this folder in your global instructions. Expand folder permissions deliberately as trust increases — start narrow, not wide. This incremental trust model is important: it is much easier to expand permissions you started narrow than to recover from permissions you started broad.

Framework Disclosure — Co-Work OS

Co-Work OS is a community-created framework by Paul. It is not an official Anthropic product or feature name. It describes a systematic architecture for Co-Work: parent folder with subfolders, each with isolated context, connected via plugins, skills, and scheduled tasks. The underlying approach is sound and verified. Modules 3 through 8 build this architecture out in detail.

Use Co-Work to Write Your Global Instructions

Here is the best shortcut in this module. Before you write a single word of your global instructions, run this prompt:

"Based on what you know about me from our conversations so far, what would you say are 3–4 sentences I should have as my general system instructions?"

Co-Work will draft them based on whatever context you have shared. Review, edit, and paste into Settings → General → Global Instructions. Then ask it to draft your Business Brain structure — it will generate the template and ask what to fill in.

The Fastest Path Through This Module

Run the prompt above to get a draft of your global instructions. Let Co-Work interview you to build your Business Brain. The entire setup — global instructions, folder structure, Business Brain, dynamic reference — takes about thirty minutes with this approach. Do not skip the safety guardrail. Add it manually if Co-Work does not include it in its draft.

Build-Along — Complete Co-Work Initial Configuration

Complete all four steps. Each one builds on the next. Do not skip ahead.

  1. Global Instructions — Open Settings → General → Global Instructions. Write or paste three paragraphs: (1) business context, (2) safety guardrail including "never delete without confirmation" and "never send without approval", (3) output format preferences.
  2. Sandbox Folder — Create ~/CoWork/ with four subfolders: Inbox/, Output/, Skills/, Business/. Reference this folder in global instructions: "You have permission to read and write within ~/CoWork/ only, unless I explicitly direct you elsewhere."
  3. Business Brain — Create ~/CoWork/Business/business-brain.md. Fill in at minimum five of the six sections with real information about your work. Even rough notes are better than a blank file.
  4. Dynamic Reference — Add one line to global instructions: "If I ask any question about my business, clients, offerings, or brand, check ~/CoWork/Business/business-brain.md first."

Success criteria: Ask Co-Work "Tell me about my business." It should respond with specific, accurate information from your Business Brain — without you providing any context in the prompt.

Knowledge Check
I know what global instructions are and when Co-Work reads them — before every task, every session, every scheduled run
My global instructions include a no-delete safety guardrail and a no-send-without-approval rule
I have created a sandbox folder structure with Inbox, Output, Skills, and Business subfolders
I have created a Business Brain document and linked it via dynamic reference in global instructions
I understand why dynamic loading saves tokens rather than costs them — business context loads on demand, not on every task