Zero
to Hero.

1. You don't need to be technical.Plain English and good questions are the method. If you can write an email to a smart colleague, you can do this.
2. Ask questions, don't give orders.The Socratic method produces better results than specific commands. Question → conversation → better answer.
3. Add context.The more context Claude has — files, research, your vision — the better the output. Context is the work.
4. Let it research.Don't just have Claude generate. Have it find and learn from real human content first, then write.
5. Use Plan Mode for big changes.Think before you build. Plan Mode forces a step back.
6. Use multiple terminals.Run research in one, build in another. Work in parallel, not sequence.

1.1 What you need

• A Mac, Windows, or Linux machine — macOS 13+, Windows 10+ with WSL, or Ubuntu 20.04+
• A Claude Pro or Max subscription (or an Anthropic Console account with billing enabled)
• An internet connection

That's it. No developer tools required.

1.2 Install on Mac / Linux

Open Terminal (Spotlight → “Terminal”) and paste this:

$curl -fsSL https://claude.ai/install.sh | bash

1.3 Install on Windows

Open PowerShell and run:

PS>irm https://claude.ai/install.ps1 | iex

1.4 Verify the install

Close and reopen the terminal, then check:

$claude --version
→claude-code 1.18.4 · ready

If you see a version number, you're in. If not, run claude doctor — it diagnoses the common issues automatically.

1.5 Log in

The first time you run claude, it opens a browser window for authentication. Click through, and you're done.

3.1 The Socratic method

Instead of telling Claude what to do, ask it questions.

When you ask, Claude thinks deeper. It asks clarifying questions back: What's the use case? How secure does it need to be? Do you need a backend? You answer in plain language. Together you arrive at a much better solution.

When Claude gives you three options and you don't know which to pick:

3.2 When you hit a bug

1. Tell Claude to fix it. Let it try 2–3 times.
2. Each failed attempt builds Claude's understanding of the problem.
3. If it still can't crack it, escalate:

Works the vast majority of the time. It's literally what real programmers do — search Stack Overflow. Claude just does it for you.

3.3 Keep prompts focused

One task per prompt beats cramming five things into one. Want to redesign the homepage and add a contact form and change the fonts? Three separate prompts. Claude does its best work with focus.

4.1 Local context — drop files in the folder

Anything you put in your project folder, Claude can read: PDFs, Excel sheets, text files, images, any format.

Pro tip: use @ to point at a specific file. Type @ and Claude autocompletes the filename — you don't write the path:

4.2 Web research — send agents out

Claude launches subagents — independent Claude instances, each on its own mission. They search, evaluate, refine, and bring back results. When done:

Now you have a structured docs/ folder Claude can reference while building.

4.3 Advanced research loops

When the agents return, analyze the information, find the gaps, send new agents to fill them. Do this three times. The compounding cycle produces a depth no single search achieves.

4.4 Claude in Chrome

Claude Code can drive your browser — open tabs, click around, fill in forms, read what's on the page. Useful for testing the site you're building, pulling data from apps you're already logged into, and kicking off deep research on Gemini or Perplexity.

Setup is two steps:

1. Install the Claude in Chrome extension from the Chrome Web Store (also works on Microsoft Edge).
2. Launch Claude Code with the --chrome flag — or run /chrome from inside an existing session.

$claude --chrome

To make it the default for every session, run /chrome and pick “Enabled by default”. The downside: browser tools always loaded means more context used, so leave it as a flag if you only reach for it sometimes.

Once it's connected, Claude shares your browser's login state — anywhere you're signed in, it can act. When it hits a CAPTCHA or login wall, it pauses and asks you to handle it manually.

Press Shift + Tab until you see Plan Mode, or type /plan. Now when you prompt:

Instead of changing files immediately, Claude:

1. Analyzes everything.
2. Writes a detailed plan — what to fix, what to improve, in what order.
3. Presents it for approval.
4. You can modify the plan or accept it.
5. On accept, it executes everything at once.

5.1 When to use it

• Starting a new project.
• Making big changes.
• Any time you want to think before acting.
• When you want to understand what Claude will do, before it does it.

Claude Code automatically saves your code state before each change. If a change isn't right:

• Press Esc twice to rewind to the previous checkpoint.
• Or type /rewind to pick which checkpoint to go back to.

When you rewind, choose what to restore:

• Just the code (keep the conversation going)
• Just the conversation (keep the code changes)
• Both

6.1 Git for permanent saves

Checkpoints are temporary. For permanent version control, use Git:

Claude handles the git commands for you. You now have a snapshot you can always return to.

7.1 What it is

Claude Design is Anthropic's visual canvas — chat on the left, design on the right. Powered by Opus 4.7. Available on Pro, Max, Team, and Enterprise plans (research preview).

It builds apps, websites, slide decks, posters, one-pagers — anything visual. Once you've set up your design system once (a one-time, five-minute job), every project after that uses your colors, type, and components automatically.

7.2 Build first, design last

You can start a Claude Design project from screenshots, references, and prompts. That works. But here's the move that pays off:

Build the app, website, or deck in Claude Code first. Focus only on features and functionality. Ignore the design entirely — placeholder copy, default fonts, bare HTML. Get the structure right.

Then push it to GitHub.

Then run that GitHub repo through Claude Design — together with your prompts, screenshots, references, brand guidelines, whatever you've gathered.

Now Claude Design knows every component you actually have. It understands the whole project. The design it gives back fits the real thing — not a guess about what the real thing might look like.

It's the same lesson as Chapter 04: context is the work. Give Design the real codebase and it gives you back something real.

7.3 Same goes for decks

Building a slide deck for the board? Don't open Claude Design first. Open Claude Code, dump your numbers and key points into a markdown file, have Claude draft the speaker notes and rough slide structure.

Then push that to GitHub. Then run it through Claude Design with your brand guidelines.

Same flow, same payoff. The thinking happens in code; the polish happens in Design.

7.4 The handoff back to code

When the design feels right, hit Export → Send to Claude Code. You get a “handoff bundle”: HTML/CSS/JS prototypes, design system tokens, and the chat transcripts that got you there.

Back in Claude Code, point at the bundle and ask:

The chat transcripts are the underrated part — they give Claude the WHY behind every decision, not just the WHAT. The result is closer to what you actually meant.

• Read and process spreadsheets
• Send emails (with confirmation rules in CLAUDE.md)
• Connect to APIs
• Build command-line tools
• Transcribe audio files
• Automate repetitive tasks
• Organize files and folders
• Process images

Another example: 200 fakturas (invoices) as PDFs. Drop them in /regnskap, tell Claude to pull the data into Excel for the accountant. Done — work that would take a bookkeeper a full day, finished while you make coffee.

11.1 What a skill is

A skill is a folder with one file inside it — SKILL.md — holding a short description and a set of instructions. Think of it as a reference booklet on a shelf: Claude only takes it down when the job in front of it needs that booklet.

Here is the part that matters. Claude always sees the one-line description of every skill. It loads the full instructions only when your request actually matches. So you can hand Claude a long, detailed process — a whole house style guide, a release checklist, a tone of voice — and it costs almost nothing in the conversation until the moment it is used. Knowledge on tap, no clutter.

#SKILL.md
---
description: Our writing voice.
             Use when writing copy or UI text.
---
Write in plain language. Short sentences,
active voice. No buzzwords, no exclamation marks.

The lines between the --- markers are the description Claude always sees. Everything below loads only when it is needed. A skill can also be a whole folder — instructions plus templates, examples, even fonts — when one file is not enough.

A skill runs two ways: Claude loads it automatically when your request matches the description, or you run it yourself by typing /skill-name.

11.2 The skills you already have

Claude Code ships with a handful of skills built in — nothing to install. Type / in a session to see them; /loop, /simplify and /debug are three you will use. They behave like any other skill.

11.3 Make your own — with skill-creator

You do not have to learn the SKILL.md format by hand. There is a skill whose entire job is to write skills: skill-creator. It is the one skill worth installing on day one.

$npx skills add https://github.com/anthropics/skills --skill skill-creator

Then describe the process you want to capture — skill-creator asks a few questions and writes the skill file for you.

This is how you stop repeating yourself. A skill is the way to preserve work you have already done. Two clear signs that something should become one:

• You do the same thing over and over — reports, applications, press releases.
• You spent real time getting something specific right — a tone, a structure, a layout — and want next time to be easier.

Take something you built today and turn it into a reusable skill:

Claude writes the skill and saves it in the project. Next time you ask for something similar, it picks the skill up on its own.

11.4 The wider library — skills.sh

Beyond what you build yourself, skills.sh is the community library — hundreds of ready-made skills, from SEO audits to video editing. Every skill there shows its exact install command, always the same shape you saw above: npx skills add, a repository, and a skill name.

Anthropic publishes its own official skills in that library too. Among them are skills that let Claude build Word documents, spreadsheets and slide decks, and fill in PDF forms — Claude Code is not only for websites. They are not bundled in; you add them like any other skill. You can also type /plugin inside Claude Code to browse and install skill bundles, or simply ask Claude in plain language to install one for you.

11.5 Skill, CLAUDE.md, or rule?

Three places to put knowledge — the difference is when each one loads:

• CLAUDE.md — facts that are always true, loaded every session. “This project is Next.js. Deploy with pnpm build.”
• A skill — a procedure you only sometimes need, loaded on demand. A good rule: when a part of your CLAUDE.md has grown into a step-by-step procedure rather than a fact, move it into a skill.
• Rules (.claude/rules/) — like CLAUDE.md, but split into small files, each one scoped so it loads only when you touch certain files.

Skills you write live in ~/.claude/skills/<name>/ for every project, or .claude/skills/<name>/ for just this one.

11.1 Custom slash commands

Reusable prompts you trigger with /your-command-name.

1. Create a folder: .claude/commands/ in your project.
2. Add a markdown file. Example:

#review.md
Review the current codebase for:
1. Code quality issues
2. Performance problems
3. Accessibility concerns
Give me a prioritized list of improvements.

Now in Claude Code, type /review — it runs that prompt. Parameterized commands work too: fix-issue.md with $ARGUMENTS lets you run /fix-issue 42.

11.2 CLIs — most of the time, simpler than MCP

The lowest-friction way to give Claude Code a new ability is usually a command-line tool — GitHub's CLI, the Vercel CLI, the Stripe CLI, AWS, ffmpeg, anything you'd type yourself. Once it's on your PATH, Claude can run it.

Just ask:

Claude installs it (Homebrew on Mac, winget on Windows), then explains the next step. Authentication is the one thing you do yourself — and it has to happen in a new terminal, not the one Claude is running in.

Open a fresh tab and run:

$gh auth login

Pick the defaults, paste the one-time code into the browser when prompted, done. Back in Claude Code, gh works for the rest of the session — Claude can list issues, open PRs, check workflow runs, the lot.

11.3 MCP servers — for databases and live state

MCP (Model Context Protocol) lets Claude talk to external services through a structured API rather than a CLI. The cost is a bit more setup; the win is structured data flowing both ways.

The clearest case: a database. Neon is a serverless Postgres database, and their MCP server lets Claude create databases, run SQL, manage migrations, and optimize queries — all through natural language.

$claude mcp add neon --transport http --scope user https://mcp.neon.tech/mcp

The --scope user flag installs it globally — every Claude Code project can use your Neon databases. Start Claude Code, run /mcp, and a browser window opens for a one-time login. No API keys, no config files.

Then say:

And from there:

• “Create a new Postgres database for my todo app”
• “Add a users table with email, name, and created_at”
• “Show me the slowest queries and suggest indexes”
• “Create a branch so I can test this migration safely”

Other MCP servers worth knowing: Playwright (browser testing), Supabase, Linear, Stripe.

How to find and install them — two easy ways:

1. Inside Claude Code, type /plugin to browse officially supported MCP servers and install with one click.
2. Or visit the MCP server's own website and copy the install command they show you. It usually looks like this:

$claude mcp add supabase --transport http https://mcp.supabase.com/mcp

Paste, hit enter, follow any prompts. Done.

11.4 Hooks — automatic behaviours

Hooks run shell commands automatically when Claude does certain things:

• Auto-format code after every file edit
• Run linting before every commit
• Run tests after code changes

Configure hooks in settings.json, or just ask Claude to set them up.

11.5 Plugins — bundle and share

Plugins package slash commands, subagents, MCP servers, and hooks into a single installable unit. Starter kits for specific workflows. Install with /plugin, or browse community marketplaces.

12.1 Terminal shortcuts

12.2 Useful commands

12.3 Credit-saving tips

• Use Sonnet or Haiku for simple tasks and research — switch with /model
• Use Plan Mode to avoid wasted iterations
• Keep prompts focused — one task at a time
• Use CLAUDE.md rules to prevent unnecessary work
• Close sessions when you're done to avoid idle usage

Not sure where to start? Here are six honest first projects — small enough to finish today, real enough to use tomorrow.

Start simple. Iterate. Let Claude teach you along the way.