K2 · Claude Code — Beginner's Guide

What is Claude Code?

Claude Code is an AI assistant that helps you write, understand, and improve code. Think of it as a smart coding partner that can work alongside you — reading your project, drafting features, fixing bugs, and running tests, around the clock.

Key capabilities

What it can do:

Understand large projects — read and explain how your code works.
Write code automatically — create new features from your description.
Fix bugs — find and repair problems in your code.
Run tests — check that code works correctly.
Work independently — complete tasks on its own for long stretches.

And beyond coding: data analysis & engineering, system maintenance (SRE), and data-science tasks.

Getting started — the claude.md file

Think of claude.md as an instruction manual you write for Claude about your project.

What to include:

How your team likes to work.
Coding rules and best practices.
Business-specific terms Claude might not know.
A high-level overview of your project structure.

Pro tips: save these files in your project folders; create multiple files for different parts of the project; run /init to have Claude scaffold a starter file; and keep it concise — don't overload it with detail.

Core features

Plan Mode

Claude thinks through the problem, makes a to-do list, and asks for approval before starting. Best for complex tasks where you want oversight.

Shift + Tab

Auto Mode

Claude works independently and makes changes automatically. Best for simple, repetitive tasks.

Extended Thinking

Gives Claude extra time to reason before acting. Cue it with phrases like “think carefully.” Best paired with Plan Mode.

Tab

Checkpoints

An automatic “undo” — your code state is saved before changes, so you can roll back if Claude goes the wrong way.

Escape × 2

Advanced features, made simple

Subagents

Specialized helpers for specific jobs. Main Claude is the general contractor; subagents are the electrician and plumber — e.g. build validator, test runner, code architect.

Hooks

Automatic actions at specific moments — auto-format after edits, notify when a task completes, or run safety checks before risky actions.

Custom Slash Commands

Shortcuts for frequent tasks. Instead of a long instruction, create /review and reuse it.

Plugins

Bundles of settings — your subagents, hooks, and commands — packaged to share with your team in one go.

Best practices for success

1 · Discovery

Ask Claude to explain how features work, use it to learn unfamiliar parts of a project, and lean on it to onboard new team members.

2 · Planning

Still write good design docs. Let Claude brainstorm edge cases. Use Plan Mode + Extended Thinking for complex features.

3 · Implementation

Test-Driven Development is key — write tests first, then code. Without tests, Claude can miss edge cases; with them, success rates climb dramatically. Give it clear validation criteria so it can check its own work.

4 · Context management

Claude works better with focused, relevant information. Use /compaction to compress conversations, route specialized work to subagents, be selective about file reads, and let Claude keep its own memory files for complex plans.

Common problems & fixes

Too agreeable

Says “you're absolutely right” even when you're not. Fix: tell it (in claude.md) to challenge your assumptions and act like a tough code reviewer.

Incomplete understanding

Proceeds before grasping the task. Fix: spend time in Plan Mode, give business context, and share design docs.

Too many files

Creates duplicate or unnecessary files. Fix: prompt it to stop the moment you notice it.

“Reward hacking”

Comments out failing tests instead of fixing them. Fix: use deeper thinking, nudge it not to be lazy, and encourage it to solve the problem properly.

Helpful commands

/init

Creates a starter claude.md file.

/compaction

Compresses the conversation to save space.

/agents

Configure specialized subagents.

/memory

Add important info to remember.

Shift + Tab

Toggle Plan Mode on / off.

Tab

Toggle Extended Thinking on / off.

Escape (twice)

Rewind / undo Claude's recent work.

Where you can use it

Terminal — the command-line interface (the original way).
VS Code — a sidebar panel in your editor.
JetBrains IDEs — IntelliJ, PyCharm, and the rest.
GitHub Actions — tag Claude for PR reviews.
Custom apps — embed it with the Claude Agents SDK.

Tips for non-technical people

How to talk to Claude: be clear and specific, don't worry about perfect technical terms, and feel free to say “I don't understand, explain it simpler.”

When to use it: understanding what existing code does, building features from a description, finding and fixing bugs, and learning how a project works.

When to be careful: always review its work (it's an assistant, not a replacement), use Plan Mode for important tasks, double-check before deploying to production, and ask questions if something looks off.

✓

Remember

The short version

Start simple — basic tasks first, then work up to complex ones.
Use tests — writing them first dramatically improves results.
Plan first — use Plan Mode for anything complex, and review before proceeding.
Save checkpoints — don't fear breaking things; you can always rewind.
It's a partnership — Claude is powerful, but you're still the developer in charge.

Source

Notes based on the Claude 301 Training webinar by Anthropic's Applied AI team. View the original Markdown · official docs at docs.anthropic.com.