A free guide by Mariah Brunner

The Claude
Control
Hierarchy

3 levels that control how Claude behaves in Cowork. Most people only use one.

@itsmariahbrunner

Why Cowork feels inconsistent

Here's something most people don't realize: Cowork has no memory between standalone sessions. Every time you start a new task, it starts from scratch. That's why Claude is great sometimes and completely off other times. It's not random. It's because you're not giving it the right context at the right level.

There are three levels that control how Claude behaves. Most people only set up one (if that). Once you set up all three, Claude stops guessing and starts working the way you need it to. Every time.

Level 1

Global Instructions

Global Instructions are the baseline rules Claude follows in every single session, every task, every project. You set them once and they apply everywhere. This is where you tell Claude who you are and how you always want things done.

How to set it up

Go to Settings → Cowork → Global Instructions → Edit. Write your instructions and save. These load automatically before every task.

Example Global Instructions (copy and customize)

My name is [name]. I'm a [role] at [company]. Tone: Direct and concise. No long preambles or disclaimers. Format: Use bullet points over paragraphs. Default to .docx for documents, .xlsx with formulas for spreadsheets. Process: Always ask clarifying questions before executing. Show me a short plan before writing or editing files. File naming: YYYY-MM-DD_topic_v1 Rules: Don't invent stats. Cite sources or say you don't know. If uncertain, stop and ask. Don't guess.

What to put here vs. somewhere else

Global Instructions should only contain things that are true for ALL your work. Your identity, your communication preferences, your non-negotiables. Anything that's project-specific goes in Level 2.

Level 2

Context Files (Your Folder)

This is where most people drop off, and it's the most important one. When you give Cowork access to a folder, it reads the files in that folder at the start of every session. So you create markdown files that tell Claude everything it needs to know about that specific project.

Different folders = different context = different behavior. Your client work folder has different rules than your internal reporting folder. A marketing project needs different context than a hiring workflow. That's the whole point.

Why this matters so much

Since Cowork has no memory between sessions, these files ARE your memory. Think of them like an onboarding document for a new employee, except this employee reads the document every single morning before starting work and never forgets a word. You write it once, and it compounds over time as you tweak it.

Create a folder for your project and add these three files:

File 1: about-me.md (copy and customize)

# About Me ## Role [Your name], [your title] at [your company] ## What I do [2-3 sentences about your day-to-day responsibilities] ## Current priorities - [Priority 1] - [Priority 2] - [Priority 3] ## My team - [Who you report to] - [Who reports to you or works with you] - [Key stakeholders you interact with] ## Industry context [Anything Claude needs to know about your industry, your clients, or your market]

File 2: voice-and-style.md (copy and customize)

# Voice and Style ## Tone [e.g., Professional but warm. Direct. No fluff. Confident but not arrogant.] ## Writing rules - [e.g., Short sentences. Lead with the point.] - [e.g., Bullet points over paragraphs when possible.] - [e.g., No jargon unless the audience expects it.] ## Words and phrases I use [e.g., "here's the deal", "the short version", "what this means for you"] ## Words and phrases I NEVER use [e.g., "synergy", "leverage", "circle back", "deep dive", "at the end of the day"] ## Examples of writing I like [Paste 1-2 short examples of your best work so Claude can match your voice]

File 3: working-rules.md (copy and customize)

# Working Rules ## Before starting any task - Always ask clarifying questions before executing - Show a short plan before writing or editing files - Read all relevant files in this folder first ## Output format - Documents: .docx - Spreadsheets: .xlsx with formulas (no screenshots of tables) - Save deliverables to /outputs/ with clear filenames ## Quality standards - Don't invent data. Cite sources or say "I don't have this" - Match the tone in voice-and-style.md - If something is ambiguous, ask. Don't assume. ## When you're done - Always produce a change log of what you created or modified - Flag anything you're unsure about

Recommended folder structure

Keep your project folder organized so Claude knows where to find and save things:

/your-project/
  about-me.md
  voice-and-style.md
  working-rules.md
  /templates/ — examples of output you like (past reports, email samples)
  /references/ — brand guidelines, project briefs, read-only context
  /drafts/ — work in progress
  /outputs/ — finished deliverables

Folder instructions override global

If your Global Instructions say "use a casual tone" but your folder's voice-and-style.md says "formal and polished," Claude follows the folder context for that project. Global sets the default. Folder context overrides it when there's a conflict. That's how you get different behavior for different projects without changing your settings every time.

Level 3

Skills

Skills are the specific actions Claude takes, triggered by /slash-commands. You type /call-prep or /draft-brief and Claude runs that exact workflow every single time. No re-explaining. No hoping it remembers. One command, consistent output.

In Cowork, Skills are way more powerful than in regular Chat. They're operational. They govern every file Claude creates, and they combine automatically when a task crosses domains. Ask for a presentation with data analysis, and your presentation skill and your data skill both activate without you doing anything.

How to create a Skill

The easiest way: use Cowork's built-in Skill Creator. It interviews you about what you need, then generates a properly structured skill file. You can also tell Claude directly: "Save this as a Skill called /[name]" and describe what you want. Or build them manually as markdown files in your skills folder.

Example: Create a /weekly-report Skill

"Save this as a Skill called /weekly-report. When I run /weekly-report: 1. Read my emails and calendar from the past 5 business days 2. Pull out the top 5 things that happened (decisions, milestones, blockers) 3. List any open items or follow-ups I still owe someone 4. Draft a short summary I can send to my manager 5. Keep it under 200 words, in my voice, with bullet points Save to /outputs/ as YYYY-MM-DD_weekly-report.docx"

Pro tip: chunk your skills by domain

Don't build one massive "writing" Skill. Build separate ones: /client-email for client communication, /internal-update for team updates, /blog-draft for content. Each one has its own rules, tone, and format. Claude loads only what's relevant per task instead of trying to apply one set of rules to everything.

How the three levels work together

Global Instructions

Control everything. Your identity, your preferences, your non-negotiables. Always on, every session.

Context Files

Define the project. Who it's for, what the rules are, what good output looks like. Changes per folder.

Skills

Tell Claude the work to do. Repeatable actions you trigger with one command. Consistent every time.

When all three are set up, Claude knows who you are (Global), understands the project (Context Files), and knows exactly what to do when you ask (Skills). That's when Cowork stops feeling like a chatbot and starts feeling like someone who actually works with you.

Things most people miss

Always request a change log

Add this to your working-rules.md: "When you're done, produce a change log of every file you created or modified." This way you can review every decision Claude made instead of digging through folders wondering what changed.

Put examples in your /templates/ folder

Claude pattern-matches against examples better than it follows abstract rules. If you want your reports to look a certain way, don't just describe the format. Drop an actual example in /templates/ and tell Claude to reference it. The output quality jumps immediately.

Update your context files as you go

These files compound over time. Every time Claude gets something wrong, add a rule to prevent it. Every time your role evolves, update about-me.md. The people who get the most out of Cowork are the ones who treat these files like living documents, not a one-time setup.

"

Global instructions control everything. Context files define the project. Skills tell Claude the work to do.

Want this set up for your job?

Master AI by Monday

Now you know the hierarchy. The Weekend Claude Bootcamp sets up all three levels for your specific role. You pick your job, follow the steps, and in 2 hours you have Global Instructions, Projects with context, Skills as /slash-commands, and real workflows you'll use every week. All done for you.

25 different roles to choose from. You find yours and go.

Get the Weekend Claude Bootcamp

Follow me for more Claude tips

TikTok Instagram