Claude Code

Claude Code is Anthropic’s agentic coding tool. It runs in the terminal, reads your codebase, edits files, runs commands, and manages git - all through natural language.

What It Is

An AI agent that lives in your development environment. Unlike chat-based coding assistants that work on snippets, Claude Code operates on your actual project: reading files, searching code, making edits, running tests, and committing changes.

Available as:

Terminal CLI - npm install -g @anthropic-ai/claude-code, then claude in any project
VS Code extension - sidebar panel with inline diffs and code actions
Web - claude.ai/code for browser-based sessions

The Agent Loop

Claude Code follows the same pattern as every agentic system:

User prompt
    |
    v
+-------------------+
| Claude decides:   |
| - read a file?    |<--+
| - edit code?      |   |
| - run a command?  |   |
| - respond?        |   |
+-------------------+   |
    |                   |
    v                   |
+-------------------+   |
| Execute tool      |   |
| (Read, Edit,      |---+
|  Bash, Grep...)   |
+-------------------+
    |
    v
Final response to user

Claude receives your prompt, decides what action to take, executes it, observes the result, and repeats until it has a complete answer. A single prompt might trigger dozens of tool calls.

Built-in Tools

Tool	Purpose
Read	Read file contents
Edit	Make targeted edits to files
Write	Create new files
Bash	Run shell commands
Glob	Find files by pattern
Grep	Search file contents
Agent	Spawn subagents for parallel work
WebSearch	Search the web
WebFetch	Fetch web page content

Project Conventions

CLAUDE.md

The most important file for configuring Claude Code’s behavior in your project. Place it at the project root:

# Project Name

## Build & Test
- `npm run build` to build
- `npm test` to run tests

## Code Style
- Use TypeScript strict mode
- Prefer named exports over default exports

## Architecture
- src/api/ contains REST endpoints
- src/db/ contains database queries

Claude reads CLAUDE.md at the start of every session. It shapes how Claude approaches your codebase.

.claude/ Directory

.claude/
├── skills/           # Custom skills (SKILL.md files)
├── settings.json     # Permission rules, model preferences
└── settings.local.json  # Personal overrides (gitignored)

Permission Model

Claude Code asks before taking actions. You control the level of autonomy:

Level	Read files	Edit files	Run commands
Default	Auto	Ask first time	Ask always
Allowlisted	Auto	Auto (for allowed patterns)	Auto (for allowed commands)
Full autonomy	Auto	Auto	Auto

Configure in .claude/settings.json:

{
  "permissions": {
    "allow": [
      "Bash(npm test)",
      "Bash(npm run build)",
      "Edit"
    ]
  }
}

Wildcard patterns work: Bash(git *) allows all git commands. Bash(*-h*) allows any command with a help flag.

Three Interaction Modes

Terminal - most powerful, full tool access:

cd my-project
claude

VS Code - sidebar panel with inline diffs, code lens actions, and file-level suggestions.

Web (claude.ai/code) - browser-based sessions with the same agentic loop. Useful for working from any device.

Key Commands

Command	What it does
`/help`	Show available commands
`/clear`	Reset conversation context
`/compact`	Summarize and compress context
Cmd+K	Topic switcher (jump between tasks)
Esc Esc	Undo last change (checkpoint restore)
`/rewind`	Rewind to a specific checkpoint
`/teleport`	Continue session at claude.ai/code

Claude Code quickstart - install and configure
Claude Code features - checkpoints, hooks, subagents, agent teams