Malik Chohra

Posted on Mar 20

Claude Code for beginners: what it is, how to set it up, and why people won’t shut up about it

#vibecoding #ai #anthropic #claude

TL;DR

Claude Code is Anthropic’s AI coding agent that runs in your terminal, reads your whole project, edits files, and loops until the task is done
It reached $1B in six months, partly because non-technical users started using it for SEO, file organization, marketing automation, and yes, literally monitoring tomato plants
Setup takes under 10 minutes: install, authenticate, run claude in your project folder
CLAUDE.md is the most important thing to set up — it gives Claude permanent memory about your codebase
Plan Mode makes Claude think and propose a plan before touching anything — essential before any task that spans multiple files
Claude Skills are installable, reusable workflows — over 25,000 exist on GitHub right now

Before we start, a word:

I started working on an AI native Boilerplate for Mobile Development. A solution that has a scalable and clean architecture, with the best approaches to have a great app performance. Adding more AI context and rules, to make the AI hallucinate less, and keep the clean code. You can use it either with Cursor, Antigravity, Claude Code,… and it gives the amazing results as expected. Also, it has most important features that help you launch your app fast

I m now in the Beta version, where i need people to test it, and gather feedback, you will get access to the code base for free, and i m looking for feedbacks.

Check it here: https://aimobilelauncher.com/

Also, I m working on a newsletter for Tech and non tech people, on how to use AI tools for App development: https://aimeetcode.substack.com/. Subscribe here: https://aimeetcode.substack.com/

If you need custom solution, or a mobile development, contact us in :https://casainnov.com/

Introduction:

Lately, I start to see something cool: Product manager, Designer, CS teams,… and non technical roles have started to use Terminal. I was surprised to see that now most of the people have a terminal open, which at some point was only for tech people who really understand what is going on. Now, the terminal in computers has become the norm

It is highly attributed that all of them started using Claude Code to run their project and to assist them in their work, with custom skills for their specific role / organisation. The trend is getting more and more

I don’t want to talk about combining it with OpenClaw, which took the internet by storm.

Why is everyone suddenly talking about this

I started using it at first when it was announced: as a tech person, Anthropic models are the best at coding, and I use most of their models < depends on the tasks in hand, of course>. And I wasn’t that surprised with Claude Code. It has skills and rules, but I kept the same AI architecture for coding < rules, context, memory bank, indexing…>

I have been using this approach for my aimobilelauncher.com to launch my AI native boilerplate. The plan mode was really helpful, and including skills for it is my roadmap for later

Then Anthropic confirmed $1 billion in annualized revenue in six months, and Fortune ran the story. I read it expecting the usual: 10x engineer productivity, enterprise contracts, vibe coding demos.

The use cases they listed were: booking theatre tickets. Filing taxes. Monitoring tomato plants. Combing through museum archives. Automating Slack messages.

That’s not a developer story. Something else was happening.

Netflix, Spotify, Uber, and Salesforce had all adopted it. Accenture announced they’re training 30,000 staff on the platform. Not 30,000 engineers. 30,000 staff. Anthropic’s own non-technical marketing team runs it to automate Google Ads creative, App Store metadata, email workflows, and SEO.

The reason it spread outside engineering is simpler than most people explain: Claude Code doesn’t need you to know how to code. It needs you to know what you want done. The terminal is just a text box. Files are organized information. Claude Code is an agent that can read, edit, and act on both. Also, the community adopted it very fast, and they started creating solutions on top of it: Open source skills is an example.

## What Claude Code actually is

Claude Code is a terminal-native AI agent from Anthropic. Not a plugin. Not autocomplete. It runs in your terminal with access to your actual files and your actual command line.

Give it a task, and it runs this loop:

Reads your files, searches directories, checks configs
Proposes a plan
Edits files, runs commands
Reads the output — test results, build logs, errors
Fixes what broke and runs again

Step five is what most AI coding tools skip. Copilot writes a suggestion and stops. Cursor helps you write code per session and stops. Claude Code stays in the loop until the job is done — or until it runs into something it genuinely can’t figure out and asks you.

How to install it

You need one of these to start:

A paid Claude subscription (Pro at $20/month minimum)
Or a Claude Console account with active API billing

No free tier. That’s the only real gate.

`// macOS / Linux:
curl -fsSL https://claude.ai/install.sh | bash

// Via npm (still works, but deprecated by Anthropic):

npm install -g @anthropic-ai/claude-code`

After installing:

cd your-project-folder claude

First run asks you to authenticate with a link. It sends a verification code back to the terminal. About 60 seconds total. Then you’re in.

## Set up your workflow: CLAUDE.md

CLAUDE.md is a Markdown file you create at the root of your project. Claude Code reads it at the start of every session.

Without it, every session starts from scratch. Claude doesn’t know your stack, your folder structure, your naming conventions, or the three things you told it never to do two sessions ago. It just scans what it can see and makes assumptions.

That leads to weird architecture suggestions, wrong import paths, and Claude confidently using a pattern that your codebase specifically avoids.

CLAUDE.md fixes this. You write down the things Claude should always know, and it reads them every time.

Here’s a real example for a React Native / Expo project:

`# Project: My App

Stack

React Native with Expo SDK 54
TypeScript (strict mode)
NativeWind for styling
Redux for state management
Supabase for backend

Coding conventions

Functional components only — no class components
TypeScript with explicit return types
Named exports, not default exports
Folder structure: screens/, components/, hooks/, utils/, services/

What never to do

Do not install new libraries without asking first
Do not modify .env files
Do not refactor code unless explicitly asked

Git

Always create a branch before making changes
Conventional commits: feat(), fix(), refactor()

Expo

always using 'npx expo install

You can ask Claude to generate a first draft for you:

Read the project structure and generate a CLAUDE.md file for this codebase.

Plan Mode — think before you touch anything

Most people find out about Plan Mode by watching it not be on and regretting it.

Plan Mode makes Claude read files and ask questions, but it cannot write, edit, or run anything while in this state. It has to think, plan, and explain what it intends to do before execution starts. Then you approve, modify, or cancel.

To activate: press Shift + Tab twice. The mode indicator in the bottom left of the terminal changes.

The practical difference is clearest with an example.

Without Plan Mode, you say:

“Add authentication to this React Native app”

Claude starts writing immediately. It picks an auth library, picks a navigation approach, picks a token storage strategy. By the time you see the output, it’s made four foundational decisions you might disagree with and they’re already in the files.

With Plan Mode:

“Add authentication to this React Native app”

Claude asks: “Which auth provider? How is your navigation structured? Where do you store tokens today? Do you want biometric support?” You answer. Claude writes a plan: specific components, a sequence, the libraries it wants to use. You read it. You push back on anything wrong. You approve. Then it executes with a clear, agreed-upon approach.

I use Plan Mode for anything that touches more than two or three files. It’s not slower — the time you spend reviewing the plan is time you’d spend fixing wrong assumptions later.

For non-technical users: Plan Mode is honestly the feature that makes this feel safe. Nothing happens without you seeing it first.

— -

Commands worth knowing

These come up in almost every session:

A note on /clear: in long sessions, Claude starts to lose track of earlier instructions. Not always

/clear and paste a quick summary of where you are. Claude gets back on track fast. Use /compact if you want to keep a condensed memory of the session.

I usually do the 10 prompt rules: after 10 prompts, compact.

## Claude Skills

This is the part most people skip when they start. That’s usually where the frustration comes from.

A Claude Skill is a Markdown-defined workflow packaged as a ZIP or a GitHub repo. It tells Claude exactly how to execute a specific, repeatable task, step by step, with the rules and output format written in.

Every time you want a repeatable task done without a Skill, you explain the whole process from the start. Every session. Skills replace that. Install once, type one command, and Claude runs the full process the same way every time.

Think about any process you’ve explained to someone more than twice. A Skill is that explanation written down properly, so Claude follows it without you having to be there for every run.

There are over 25,000 community-built Skills on GitHub:

awesome-claude-skills: https://github.com/travisvn/awesome-claude-skills

skills.sh: https://skills.sh/

browsable marketplace: https://github.com/anthropics/skills

Skills cover SEO audits, React Native performance checks, documentation generation, DevOps automation, security reviews, data analysis, and a lot more. The ecosystem is growing fast, and most of it is free.

Anthropic also shipped organization-level Skills deployment in December 2025. Admins can now push Skills workspace-wide with automatic updates. Teams can standardize processes the same way they standardize code.

How a Skill actually works

Each Skill has a SKILL.md file. YAML frontmatter at the top tells Claude when and how to invoke it. The Markdown below is what Claude follows when it runs.

— -
name: rn-performance-audit
description: Audits a React Native screen for performance issues.
Use when reviewing a screen component or when the user asks
about performance, re-renders, or slow UI.
— -
When auditing a React Native screen for performance:

Check for unnecessary re-renders (missing useMemo, useCallback)
Identify heavy computations inside render
Check FlatList vs ScrollView usage
Look for images without proper caching or resizing
Flag synchronous operations on the main thread

Output: a prioritized list of issues with code-level fix suggestions.

Once installed, Claude auto-loads this Skill when you ask about performance, or you invoke it directly with /rn-performance-audit.

— -

Example of A skill that I use: Claude SEO skill

I have started lately using seo skill: https://github.com/AgriciDaniel/claude-seo. It is really helpful for website SEo audit, and then adding fixes

Check how to install it there, and then you do / to call it, and do /seo so it gives you the whole report

I like after that to ask to create a fix plan, using Plan mode for that. and then i ask it to implement it

Things that will catch you off guard

I learned these the hard way, so you don’t have to.

Commit before every Claude session:

Claude can introduce regressions. It fixes one thing and occasionally breaks something in a different file. Not often, but enough. A clean Git commit before you start means rollback is one command. I’ve needed it more than once.

Auto-accept is a trap.

*There’s a — dangerously-skip-permissions flag that lets Claude apply all changes without asking. Don’t use it until you’ve run a task type several times and understand its output well. I let it “tidy up some type definitions” once and got 40 renamed interfaces back. All of them logical. None of them what I wanted. I rolled back with Git and never used that flag on unfamiliar tasks again.

Token costs sneak up on you:

Long sessions, large repos, parallel agents — these burn tokens faster than a focused single-task session. Run /cost inside any session to check where you are. Monitor the Anthropic console when you’re starting out. You want to know what a typical session costs before you schedule one as a weekly routine.

Messy codebases produce messy results:

Claude Code’s ability to understand your project depends entirely on how coherent that project is. Scattered configs, inconsistent patterns, no CLAUDE.md — it still runs, but the suggestions start to drift in weird directions. The architecture discipline makes it more useful. Chaos amplifies chaos.

Rate limits reset on a rolling 5-hour window, not daily:

If you hit your limit at 2 pm, you get a fresh allocation around 7 pm. Once I understood this, I stopped fighting the limits and started working in focused sessions instead. Honestly, a better habit anyway.

FAQ

Do I need to know how to code to use Claude Code?
No. The VS Code extension and Claude Desktop both work without the terminal. You describe what you want in plain language and Claude Code executes. Non-technical users are currently using it for SEO, marketing workflows, file management, and data processing.

What is CLAUDE.md and why does it matter?
It’s a Markdown file in your project root that Claude Code reads at the start of every session. It stores your stack, coding conventions, and what Claude should never do. Without it, every session starts from scratch. It’s the single biggest difference between a good and a bad Claude Code experience.

What is Plan Mode?
A state where Claude can read files and ask questions, but cannot write or run anything. It proposes a full plan first. You review and approve before a single file is touched. Activate with Shift + Tab twice.

What are Claude Skills?
Reusable workflow instructions packaged as Markdown files in a ZIP or GitHub repo. Install once, run on demand. Claude follows the workflow consistently every time without re-explanation.

Where do I find Skills?

awesome-claude-skills on GitHub: https://github.com/travisvn/awesome-claude-skills

How much does Claude Code cost?
It requires a paid Anthropic subscription. Pro is $20/month with roughly 45 messages per 5-hour window. Max 5x is $100/month. No free tier access.

Can I use this for React Native development?
Yes. Claude Code works with any file-based project. For React Native and Expo, the main thing is setting up CLAUDE.md with your stack details and conventions. Once that’s in place, Claude can refactor screens, add features, debug issues, and run Expo CLI commands the same way it does for any other project.

DEV Community