Open PowerShell
Press the Windows key, type powershell, and press Enter. A dark window opens with a blinking cursor — that's the terminal. It's just a place where you type instructions instead of clicking them.
Windows PowerShell Copyright (C) Microsoft Corporation. All rights reserved. PS C:\Users\you> _
Install Claude Code — one command
Copy the yellow line below, paste it into PowerShell (right-click pastes), and press Enter. It downloads and installs everything on its own. Takes about a minute; if it seems to pause, let it finish.
PS C:\Users\you> irm https://claude.ai/install.ps1 | iex Downloading Claude Code... Installing to C:\Users\you\.local\bin ✓ Claude Code installed successfully
Nothing to click, no license screens — when the ✓ line appears, it's done.
irm "is not recognized," you opened Command Prompt instead of PowerShell. Close it and repeat Step 1 — type powershell, not cmd.Close the window, reopen it, verify
Close PowerShell completely and open a fresh one (Step 1 again). This lets Windows notice the new program. Then type the yellow command — a version number means the install worked.
PS C:\Users\you> claude --version 2.1.x (Claude Code)
Your exact number will differ — any version number = success.
claude doctor in the same window — it checks the install and tells you what's wrong in plain language.Go to your project folder
Claude Code works inside a folder — it can only see files in the folder you start it from. In File Explorer, open the folder you want to work in, click the address bar at the top, and press Ctrl+C to copy the full path. Then in PowerShell, type cd (with a space), paste the path inside quotes, and press Enter.
PS C:\Users\you> cd "C:\Users\you\Documents\My First Agent" PS C:\Users\you\Documents\My First Agent> _
Proof it worked: the blue prompt now shows your folder's name.
Start Claude and sign in (first time only)
Type claude and press Enter. The very first time, it opens your web browser and asks you to log in with your Claude account. Approve it, come back to the terminal, and you're in. Every time after this, claude starts instantly — no login.
PS ...\My First Agent> claude Welcome to Claude Code! Opening your browser to sign in... ✓ Signed in ────────────────────────────────────────── > Ready. Type a task, or /help for commands.
Give it a job — in plain English
No special syntax. Describe what you want as if texting a coworker. Claude proposes a plan, writes the files, and asks your permission before creating or changing anything — you approve each action. Copy the yellow prompt below word-for-word; it's the start of your first project.
> Create one file called dashboard.html — my personal start page. Dark theme. It needs: a live clock, a greeting that changes with the time of day, a list of quick links I can edit, and a to-do list that remembers my items after I close the page. Everything in this one file, no internet needed. I'll build a self-contained start page in dashboard.html: clock + time-based greeting, editable quick links, and a to-do list stored in your browser so items persist. Create dashboard.html? ❯ Yes / No
Press Enter on Yes. Claude writes the whole file — usually in under a minute.
Now the real power: it's a conversation
One-shot file generation is a party trick. The thing that actually changes how you work is the loop: look at the result, say what you want different, watch it change. Three rounds, about ten minutes total.
Round 1 — Open what it built
Without leaving Claude, type the yellow command (the ! runs a normal terminal command) — or just double-click dashboard.html in File Explorer. Your page opens in the browser.
> !start dashboard.html ✓ Opened in your default browser
Round 2 — Change it by describing the change
No hunting through code. Say what's wrong or missing, approve the edit, refresh the browser tab. This is the habit to build.
> The clock should show seconds. Make the to-do items strikethrough when I check them off, and add a button to clear all completed items. Three edits to dashboard.html: • clock → HH:MM:SS • completed to-dos → strikethrough style • "Clear completed" button Edit dashboard.html? ❯ Yes / No
Round 3 — Make it teach you
Claude isn't just a builder, it's a tutor that knows exactly what's in your file. Ask, and terminal anxiety starts turning into understanding.
> Explain how the to-do list remembers my items after I close the page. I'm brand new — no jargon. Your browser has a small notepad built into it called local storage. Every time you add or check off an item, the page writes the whole list to that notepad...
Done for the day? Type /exit — or just close the window. Your files stay. Next time, claude --continue in the same folder picks the conversation back up.
The handful of commands worth knowing
Inside a session, anything starting with / is an instruction to the app itself instead of a task for Claude. There are dozens; these six cover a normal day.
| Command | What it does | When to use it |
|---|---|---|
/help | Lists every command | Whenever you forget one |
/init | Creates a CLAUDE.md memory file for the project | Once, at the start of any real project |
/clear | Wipes the conversation, keeps your files | Switching to an unrelated task |
/compact | Summarizes a long conversation to free up room | Long sessions, when replies slow down |
/model | Switches between faster and smarter models | Big/tricky task → smarter; quick edits → faster |
/exit | Ends the session | You're done (files always stay) |
What's a CLAUDE.md?
A plain text note that lives in your project folder. Claude reads it automatically at the start of every session — it's the project's memory. Anything you're tired of repeating ("always use a dark theme," "explain changes in plain English") goes in there. Run /init and Claude writes the first draft for you; open it in Notepad to add your own rules.
What's a "skill"?
A folder containing a SKILL.md instruction file — a reusable playbook Claude pulls in on its own when the task matches (one for how your team writes reports, one for your website's style, and so on). Personal ones live in C:\Users\you\.claude\skills. File it under "later": you don't need skills for weeks, but when you catch yourself pasting the same instructions into every session, that's the sign to make one — and yes, you can just ask Claude to create the skill for you.
Why is everything a .md file?
.md (Markdown) is just a text file with light formatting — # for headings, - for bullets. It opens in Notepad. Claude reads and writes it natively, which is why instructions, memory, and skills all use it. You never need to "learn Markdown" — write plain sentences and you're already doing it.
The project that matters: build your partner agent
Here's the secret nobody tells beginners: an agent is not an app you download — it's Claude plus a set of files that give it a name, rules, and memory. You just learned about CLAUDE.md. Now you'll use it to build the agent that becomes your long-term partner — and, later, the orchestrator that manages every helper you add.
Most people bolt rules on after something goes wrong. You're going to do it in the opposite order — rules first, powers second. That's the entire trick to an agent you can trust as it grows.
Step 1 — Give it a home
Make a dedicated folder — this folder is your agent. Everything it knows and every rule it follows will live here as plain files you can open in Notepad.
PS C:\Users\you> mkdir "$HOME\Agent"; cd "$HOME\Agent"; claude > Ready. Type a task, or /help for commands.
Step 2 — Paste the founding prompt
This one prompt has Claude build the whole thing — constitution, memory, journal, and growth plan — then interview you and let you name it. It looks long because it's complete, not because it's complicated — your entire job is: copy, paste, press Enter, answer five short questions, and approve each file. Nothing to memorize.
You are becoming my long-term partner agent. Before we do anything else, set yourself up properly by creating these files in this folder. Propose them one at a time so I can approve each. 1. CLAUDE.md — your constitution, with these sections: IDENTITY — Ask me now what I'd like to call you (suggest three name options). Record the name I pick. You are my personal partner agent and, as we grow, the orchestrator of any helper agents we add. HONESTY RULES (non-negotiable) — Never invent facts, numbers, names, quotes, or sources. If you aren't sure, say "I'm not sure," then verify: use web search for anything time-sensitive or outside your knowledge, and tell me where the answer came from. Always keep three things clearly separate: verified fact, your reasoning, and your guess. If you realize you were wrong earlier, say so plainly and correct the record. SAFETY & PRIVACY RULES — Never ask for, store, or write into any file: passwords, PINs, card numbers, or government ID numbers. If I paste one, warn me and leave it out of every file. Before deleting, moving, or overwriting anything — or touching any file outside this folder — stop and explain in plain English what will happen, then wait for my yes. For medical, legal, tax, or financial topics, give me information but remind me to confirm decisions with a qualified professional. HOW WE WORK — Plain English; teach jargon before using it. If my request is ambiguous or risky, ask me one clarifying question instead of guessing. Your job is not to agree with me — push back when I'm about to do something unwise, and tell me why. At the end of every session, update JOURNAL.md. 2. memory/about-me.md — what you know about me. Create it with empty sections (Who I am / Goals / Preferences / Current projects), then interview me: five short questions, one at a time, and fill it in from my answers. Never record anything from the forbidden list above. 3. JOURNAL.md — a dated log. One line per meaningful event: date, what we did, what changed. This is our audit trail. 4. GROWTH.md — our upgrade roadmap, in order, one plain sentence each: (1) custom skills — reusable playbooks; (2) sub-agents — your future team; (3) connectors (MCP) — safe hookups to my email, calendar, and tools. Include this standing rule at the top: "No new capability gets added until its rules are written into CLAUDE.md first." When all four files exist, show me a one-screen summary: your name, the rules now in force, what you remember about me, and three suggested first missions based on my interview answers.
~2 minutes of approvals, and your agent exists — with governance most professionals skip.
Step 3 — What "done" looks like
✓ CLAUDE.md · memory\about-me.md · JOURNAL.md · GROWTH.md
Name: Atlas (your pick)
Rules in force: honesty · privacy · ask-before-changing
I remember: your goals, preferences, current projects
Suggested first missions:
1. Turn your top goal into a 30-day plan
2. Organize the notes for your current project
3. Teach you one thing from GROWTH.md, your pace
Pick a mission — that's your first journey together. From now on, start every session in this folder with claude --continue.
The one mental model that makes this all click
Growing it — without breaking it
You don't need to study anything to level up. When you're ready for the next rung, just ask your agent: "Read GROWTH.md — help me add the next upgrade, and write its rules into CLAUDE.md first." It will walk you through skills, then a sub-agent team it coordinates, then connectors — each gated by the rules-first habit you started with. The agent teaches you how to grow the agent.
Quick reference — the whole guide in five lines
| You want to… | Type this in PowerShell | Success looks like |
|---|---|---|
| Install Claude Code | irm https://claude.ai/install.ps1 | iex | ✓ installed message |
| Check it installed | claude --version (in a new window) | A version number |
| Diagnose a problem | claude doctor | Health report |
| Enter your folder | cd "paste\path\here" | Prompt shows the folder |
| Start working | claude | The > chat prompt |
| Pick up where you left off | claude --continue | Yesterday's conversation, restored |
If something looks wrong
'irm' is not recognized…
FixYou're in Command Prompt (CMD), not PowerShell. Close the window; Windows key → type powershell → Enter. The window title should say "Windows PowerShell."
'claude' is not recognized… (right after installing)
FixYou're still in the old window. Close PowerShell fully, open a fresh one, try again. New programs only appear in windows opened after the install.
cd says the path doesn't exist
FixAlmost always missing quotes around a path with spaces. Retype: cd "C:\full\path" — copy the path straight from Explorer's address bar so there are no typos.
Sign-in page loops or rejects you
FixYou're logged into a free claude.ai account. Claude Code requires Pro, Max, Team, Enterprise, or a Console (API) account — upgrade or switch accounts, then run claude again.
Tempted to "Run as administrator" to fix an error
Don'tAdmin mode installs Claude Code to the wrong user profile and creates permission conflicts that are painful to unwind. Every step in this guide works from a normal window. If you're stuck, run claude doctor first.
This page grows as you do
You need none of the following yet — that's the point. Bookmark this page; as the community levels up, new sections will land here in the same style, in this order:
| Coming | What it unlocks |
|---|---|
| Your first skill | Teach your agent a reusable playbook so you stop repeating instructions |
| Your first sub-agent | Your agent gets a teammate it manages — the start of your orchestrated team |
| Connecting your world | Safely link your agent to your browser, email, and calendar (rules first, as always) |