How to use Claude Code?

What is Claude Code?

Claude Code is a terminal-based agentic coding tool developed by Anthropic that integrates directly with the Claude AI model family. It allows developers to interact with Claude in their terminal to perform coding tasks, such as exploring codebases, editing files, and managing Git. Claude Code is designed to be a powerful, flexible, and customizable tool for coding workflows. 

Claude Code: Deep coding at terminal velocity \ Anthropic

Transform hours of debugging into seconds with a single command. Experience coding at thought-speed with Claude's AI that understands your entire codebase—no more context switching, just breakthrough results.

www.anthropic.com

Install

Install npm

# download nvm :
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash

# install Node.js :
nvm install 22

# Node.js version:
node -v # "v22.18.0"
nvm current # "v22.18.0"

# npm version:
npm -v # 10.9.3

Install Claude Code

# Install Claude Code CLI
npm install -g @anthropic-ai/claude-code
claude --version   # version

Account creation and payment

Type of use

Payment

Claude Code: Deep coding at terminal velocity \ Anthropic

Transform hours of debugging into seconds with a single command. Experience coding at thought-speed with Claude's AI that understands your entire codebase—no more context switching, just breakthrough results.

www.anthropic.com

1. Click 'Try Claude'
2. Enter your email address and password → Create an account
3. Enter card at payment stage

First run, OAuth authentication

cd ~/my-project
claude
/login

1. When the browser opens, Anthropic permission request → Authorize
2. If you see the message Logged in as you@email.com in the terminal, it is successful.

After running the /status command:

Jetbrain Plugin

You can find Claude Code by searching for Plugins.

You can check the diff of Claude Code's code manipulation.

Representative workflow

Code Description

> what does the payment processing system do?
> find where user permissions are checked
> explain how the caching layer works

Automating Git-related tasks

> commit my changes
> create a pr
> which commit added tests for markdown back in December?
> rebase on main and resolve any merge conflicts

Modify some code

> add input validation to the signup form
> refactor the logger to use the new API
> fix the race condition in the worker queue

Write a test

> run tests for the auth module and fix failures
> find and fix security vulnerabilities
> explain why this test is failing

How to Use Extended Thinking Mode

1. First, explain the task to Claude and ask him to gather context from the project.
2. Afterwards, prompt them to think more deeply and produce results using the expression "think hard"

> think about how we should architect the new payment service
> think hard about the edge cases in our authentication flow

Claude Project Memory

Memory type

Claude supports two main memory types

1. User memory (~/.claude/CLAUDE.md)
   • Personal coding preferences, custom shortcut definitions, etc. 
2. Project memory (./CLAUDE.md)
   • Project architecture and coding standards that need to be shared across the team.

• The key point is that user memory is not managed by git, while project memory is.
• Management is done via the/memorycommand or direct file editing.

Working process

• The Claude code recursively searches memory. It currently reads all distinct CLAUDE.md or CLAUDE.local.md files, starting from the carpet workspace (cwd) and ending with the elevator (/).
• Example
  • For example, when claude is executed from ./claude-code-test/utils/, the entire memory of ./claude-code-test/CLAUDE.md and ./claude-code-test/utils/CLAUDE.md is loaded.

  •   claude-code-test/
      ├── CLAUDE.md
      ├── README.md
      ├── test.py
      └── utils/ (claude 실행 위치)
          ├── CLAUDE.md
          ├── __init__.py
          └── simple_utils.py

  • Conversely, when the execution location is claude-code-test/, ./CLAUDE.mdin that directory is loaded and, when necessary, the memory of ./claude-code-test/utils/CLAUDE.md is loaded.
• Summation
  

  Area	Key
  Memory Type	(1) Project memory ./CLAUDE.md: Defines rules and workflows shared by the team. (2) User memory ~/.claude/CLAUDE.md: Defines personal preferred rules.
  Memory navigation rules	Recursively loads all CLAUDE.md and CLAUDE.local.md files from the execution directory (cwd) up to the project root (/). CLAUDE.md files in subfolders are included only when reading files in those folders.
  Add memory quickly	If you start the prompt in the terminal with #, a dialog box will appear asking, “Where do you want to store this sentence in memory?”
  Edit directly	Running the /memory command during a session opens the memory file in the default editor for editing and organizing.

A Complete Guide to Setting Claude Code Permissions

Claude Code prompts the user for permission when performing critical tasks, such as executing shell commands, modifying files, or accessing external APIs.

This can be configured in one of four ways

1. Real-time permission mode
2. View/edit permissions during a session (edit local project)
   1. claude > /allowed-tools
3. Edit the settings.json file
4. Temporarily granting permissions via CLI
   • claude —-allowed-tools “…”

Real-time authorization method (default)

When Claude tries to do something, it asks:

"Can I run npm install?"

• Allow: Run only once
• Always allow: Added to the allow list (can be run automatically)
• Deny: Not run

This response internally updates the configuration.

View/edit permissions during a session

Entering /allowed-tools in the Claude dialog box will display a list of currently allowed tools. You can add or remove them directly here.

Edit the Settings.json file

Create a settings.json file for the project or for all users.

Path

• ~/.claude/settings.json (Treated as personal user settings)
• .claude/settings.local.json(Included in .gitignore and not shared)
• .claude/settings.json (Shared as a project file within the team)

Example

See supported commands

Security - Anthropic

Learn about Claude Code's security safeguards and best practices for safe usage.

docs.anthropic.com

{
  "permissions": {
    "allow": [
      "Bash(npm run lint)",
      "Bash(npm run test:*)",
      "Read(~/.zshrc)"
    ],
    "deny": [
      "Bash(curl *)"
    ]
  }
}

• Fields
  • allow: Actions to automatically allow
  • deny: Actions to unconditionally deny
• Commands
  • Bash(...): Shell commands
  • Read(...) / Edit(...): File reading/writing

Temporary permission using CLI method

# Allow only specific tools
claude --allowedTools "Bash(npm run dev),WebFetch(domain:example.com)"

# Certain commands are rejected
claude --disallowedTools "Bash(rm -rf *)"

If the conversation session is long

Use the built-in summary prompt

• Create a file (.md) containing a summary of the conversation history
  • /compact summarize and create xxx_history.md
• Open a new claude session:
  1. claude 

  2. Based on the information in @xxx_history.md, please analyze the current status of our conversation and identify key topics, decisions, and unresolved issues.

(optional) Use custom prompts

1. Download the custom summary prompt and save it as compact_custom_prompt.md.
2. If the conversation history in interactive mode is long, use the prompt above to save the summarized content as an .md file.
   • @compact_custom_prompt.md summarize and create xxx_history.md
3. Open the file in a new session and continue the task.
   1. claude 

   2. Based on the information in @xxx_history.md, please analyze the current status of our conversation and identify the key topics, decisions, and unresolved issues.

General workflow management

Shift-Tab Mode

Claude Code allows you to select different work modes using Shift-Tab. There are three modes:

1. Plan Mode: Claude describes the plan and waits for approval before executing the task.
2. Auto Accept Mode: Claude automatically accepts all proposed tasks.
3. Normal Mode: This is the default mode, requiring approval for each task.

Revert to previous command

There are two ways to revert to a previous command or state in Claude Code:

Using the ESC key

1. Pressing the ESC key displays a list of previous chats.
2. Press the ESC key again to select a previous chat.

Note: Reverting to a previous state using this method will delete the history (future history) after the current state. The result will be undone, but you will not be able to return to the future.

Use Git commits.

A safer approach is to ask Claude Code to commit every time you make a change. This allows you to seamlessly switch between previous and current versions using Git.

# prompt
> Please commit these changes with a descriptive message before proceeding

This method allows you to safely revert to a previous state if something goes wrong while working, and even roll back to a future version if necessary.

Thanks for reading. If you found this helpful, please like and share. 😄