Most people never make it past level 3. This guide is the ladder: find your level below, then do the one setup step that moves you up. Every starter file you need is on this page.
LVL 1
Prompting
You type, Claude answers. A conversation, nothing more. You're here if Claude feels like a smarter Google.
LVL 2
Context
A CLAUDE.md file teaches it your project, rules, and preferences once. You're here if you never re-explain yourself.
LVL 3
Tools
Claude Code reads files, writes files, runs commands, searches the web. Itself. You're here if Claude DOES instead of suggests, but only on your machine.
LVL 4
MCP
Level 3 gave Claude hands on your computer. MCP gives it keys to everything else: Gmail, Notion, your database, your store.
LVL 5
Skills
Repeatable playbooks saved once, fired on demand. Your weekly report becomes one command.
LVL 6
Subagents
Specialized workers handling pieces in parallel: one researches, one writes, one reviews. You delegate instead of chat.
LVL 7
Agent Teams
A fleet of Claude sessions coordinating with each other on shared work. You manage an AI team. Almost nobody is here yet.
How To Use This Guide
Don't read it all. Find your current level, jump to that section below, and do ONE setup step. Each level takes minutes to days, not months. Climbing one level per week puts you in the top 1% of Claude users by August.
A CLAUDE.md is a plain text file Claude Code reads automatically at the start of every session in that folder. It's the difference between a stranger and a coworker. This is the highest-payoff 10 minutes in this entire guide.
Copy this starter, save it as CLAUDE.md in your project folder (the folder where you run Claude Code), and edit every bracket. Don't aim for perfect, aim for honest.
Copy this CLAUDE.md starter
# CLAUDE.md (save this file in your project folder, then edit every bracket) ## Who I am [Your name, what you do, and what this folder/project is. Example: "I'm Dana, I run a candle business. This folder holds my product copy, supplier info, and monthly numbers."] ## What I'm working on right now [The current goal in one or two lines. Update this weekly. Example: "Preparing the fall product launch: 6 product pages, an email sequence, and a wholesale one-pager."] ## How I like to work - Tone: [e.g. "warm but direct, no corporate filler, short sentences"] - Formatting: [e.g. "give me drafts in plain text I can paste, no markdown tables unless I ask"] - Decisions: [e.g. "give me a recommendation, not a menu of options. If you're unsure, ask one question instead of guessing."] - Pushback: [e.g. "challenge me when my plan has a hole. I'd rather hear it from you than from a customer."] ## Facts you should always know [The stable facts Claude keeps getting wrong or asking about: brand voice rules, key numbers, product names and prices, the people on your team and their roles, words you never use.] ## Never do this [Your hard rules. Examples: "Never use em dashes. Never invent statistics. Never touch the files in /legal. Never draft anything that promises specific income."] ## Lessons learned [Leave this section here. Every time Claude makes a mistake, add one line: "Don't X, because Y." This section is why month two feels psychic. End corrections in chat with: "Update your CLAUDE.md so you don't make that mistake again."]
The Habit That Compounds
Every time Claude gets something wrong, end your correction with: "Update your CLAUDE.md so you don't make that mistake again." It appends the lesson to the file itself. Anthropic's own team works this way. Two weeks of this and Claude stops making YOUR specific mistakes.
Claude Code is the version of Claude with hands: it comes with built-in tools to read your files, write new ones, run commands, and search the web, no setup needed. Regular Claude tells you what to do. Claude Code just does it.
01
Install it from code.claude.com/docs (there's a desktop app version and a terminal version; the docs walk you through both in about 5 minutes). It's included in your Claude subscription.
02
Open it inside a folder that matters to you (your project, your content folder, your business docs). That folder is its workspace.
03
Give it a real job, phrased as a command, not a question: "Organize this folder by project and month, and show me what you moved." Watch it work. That moment is the whole level.
The Phrase That Changes Everything
Questions get suggestions, commands get action. "Can you suggest a cleanup?" makes Claude talk. "Clean this up and show me what changed" makes Claude work. If it keeps describing instead of doing, you asked a question when you meant to give an order.
MCP servers plug your real apps into Claude. The easiest first ones are the official connectors (Gmail, Google Calendar, Google Drive) which you can turn on at claude.ai under Settings, then Connectors, with one click each. In Claude Code, you add MCP servers with one command:
In Claude Code, type claude mcp add and follow the prompts, or browse what's worth adding first in my 24 Things To Install in Claude guide (8 MCP servers, ranked, with links). Start with ONE: the tool you touch most every day.
Then test it with a question only YOUR data can answer: "Find every subscription charge in my Gmail from the last 90 days and list them with prices." When the answer comes back from your real inbox, you've left level 3.
A skill is a folder with a SKILL.md file inside: instructions Claude loads automatically when the task matches. The rule Anthropic's own team uses: if you've done it more than once today, make it a skill.
Copy this template, save it at .claude/skills/weekly-report/SKILL.md inside your project, and edit the brackets. Then just ask Claude Code for your weekly report. It fires automatically.
Copy this SKILL.md template
# Save as: .claude/skills/weekly-report/SKILL.md # (Make the folder name match the skill. This example is a weekly report skill. # Replace every bracket, then in any session just ask for your weekly report.) --- name: weekly-report description: Builds my weekly status report in my exact format. Use when I ask for my weekly report, status update, or end-of-week summary. --- You build my weekly report. Follow this exactly. ## Where to look [Tell it where the raw material lives: "Read progress.txt and the git log from the last 7 days" or "Ask me to paste my task list and calendar for the week."] ## The format (never deviate) 1. WINS: 3 to 5 bullets, each one line, each starting with a verb, numbers included where they exist. 2. IN PROGRESS: what's moving and the realistic finish date. 3. BLOCKED: what's stuck, why, and the specific ask that would unstick it. 4. NEXT WEEK: top 3 priorities only. ## Voice rules [e.g. "Plain words. No 'synergies.' Confident, not boastful. My boss is data-driven: lead with numbers."] ## Quality bar Before finishing, check: would a skeptical manager find anything vague? If yes, fix it first.
What To Build After This One
Steal the pattern for anything you repeat: a content formatter, an inbox triage, a client proposal builder. The structure is always the same: where to look, the exact format, the voice rules, the quality bar. Three skills in, you'll wonder how you worked without them.
Subagents are specialized workers with their own instructions, their own tool permissions, and their own clean context. The easiest first one is a read-only reviewer: it checks Claude's own work without being able to touch anything.
Copy this, save it at .claude/agents/reviewer.md, then after any big task say: "Have the reviewer check this before we call it done." You can also just type /agents in Claude Code to create agents interactively.
Copy this reviewer agent
# Save as: .claude/agents/reviewer.md # A read-only second opinion. It can look at everything and change nothing. --- name: reviewer description: Reviews work for quality and catches problems before I ship. Read-only. tools: Read, Grep, Glob --- You are a meticulous senior reviewer. You may read anything in this project, but you NEVER edit files or run commands. When given work to review: 1. Check it against the project's stated goals in CLAUDE.md. 2. Hunt for the three classic failures: things that contradict the requirements, things that were quietly skipped, and things that would embarrass us in front of a customer. 3. Report findings as: CRITICAL (must fix), WORTH FIXING (should fix), NITPICK (optional). One line each, with the file and location. 4. End with a verdict: ship it, or fix the criticals first. Be specific and brief. No praise padding.
The Two-Word Power Move
On any big task, add "use subagents" to your request. Claude Code splits the work across parallel workers on its own. That's level 6 in two words.
Agent teams are multiple full Claude sessions working the same project: a team lead coordinating teammates, shared task lists, work claimed and synced between them. It's experimental and opt-in, and it's where "using AI" officially becomes "managing an AI team."
Two honest notes. First: you don't need this level for most work, levels 4 to 6 already put you ahead of nearly everyone. Second: the skills that matter at level 7 are the ones you built climbing here: clear specs, good CLAUDE.md hygiene, and verification habits. The best preparation for managing a fleet is being great at managing one.
When you're ready: run parallel sessions in isolated git worktrees with claude --worktree, give each a clearly scoped job, and keep a dedicated session just for reviewing the others' output. Anthropic's power-user guide (the basis for several tricks in this guide) covers the full setup at support.claude.com.
Whatever level you're on today: one level up is one setup step away. Pick it, do it this week, and come back for the next one.
The Only AI Masterclass You Need
If this guide helped, but you’re looking to go deeper, I got you!! My 30-Day Challenge takes you from saving AI tips you never use to actually building with AI, step-by-step.
I show you exactly how I automated two e-commerce brands, my social media, and most of my personal life, then hand you the agents, workflows & systems to do the same. I’m teaching you every single thing I know with one lesson and one build a day.
Join the AI Masterclass →© 2026 Mariah Brunner. All rights reserved.