AI Explained
Claude Code, feature by feature
A coding agent in your terminal, editor and browser, and the pieces you add to make it yours.
The map
One session, six ways to extend it
Claude Code is a coding agent you talk to. What makes it yours is what you put around it. Play the diagram to see what each piece is for.
Press Play to walk through it, or Listen to have it read to you.
Read all the steps
- You ask for something in plain words. Claude Code reads the code, edits files and runs commands to do it.
- CLAUDE.md is read at the start of every session: your project's standing rules and facts.
- Skills are instructions and workflows that load only when they are needed, or when you call one by name.
- A subagent does a side task in its own context and hands back a summary, so the main conversation stays clean.
- MCP servers connect it to outside systems: your tracker, your database, your design tool.
- Hooks run your script every time an event happens, such as after an edit. Use them for rules that must always hold.
- A plugin packs skills, hooks, subagents and MCP servers together so a team can install them in one step.
Start using it
From nothing to a first task
Seven steps, with the commands from the documentation. Copy a command, paste it in your terminal.
Install it
On macOS, Linux or WSL, run the installer in a terminal. Windows has its own PowerShell and WinGet commands on the same page [1].
curl -fsSL https://claude.ai/install.sh | bashStart it inside a project
Open a terminal in your project folder and run it. It asks you to log in the first time [1].
claudeAsk for something real
Describe the outcome in plain words and let it plan, edit and check. The documentation's own first example is below [1].
claude "write tests for the auth module, run them, and fix any failures"Write down the standing rules
Add a CLAUDE.md file at the project root with your coding standards, architecture decisions and review checklist. It is read at the start of every session [1].
Connect your tools
With MCP it can read your design documents, update tickets or use your own tooling [1].
Script it
Pipe text into it or run it in CI with the -p flag [1].
git diff main --name-only | claude -p "review these changed files for security issues"No terminal?
Use the VS Code extension, the JetBrains plugin, the Code tab of the desktop app, or claude.ai/code in a browser. The same engine is behind each [1].
Which one?
Pick the right extension point
Choose what you are trying to do. The answer is the documentation's own guidance [2].
The features
Feature by feature
As the documentation described them on the date below. Claude Code changes often: follow the links in the sources for the current details.
What it is
An agentic coding tool that reads your codebase, edits files, runs commands and integrates with your development tools [1].
Where it runs
In the terminal, in VS Code and JetBrains editors, in a desktop app and in the browser, and from GitHub Actions and GitLab pipelines. The same engine is behind each, so your project rules, settings and MCP servers carry across them [1].
CLAUDE.md
A file of standing instructions that is loaded in every session. The docs call it context, not enforced configuration, and advise keeping it under 200 lines [3] [2].
Auto memory
Claude can also keep its own notes from your corrections and preferences, stored for each repository [3].
Skills
Reusable knowledge and workflows that load on demand, or when you call one with a slash and its name. A skill is a SKILL.md file in a folder of its own [4].
Subagents
Isolated workers with their own context, system prompt, tools and permissions. Built in are Explore, for read-only search, and Plan, for research in plan mode; you can write your own [5].
Hooks
Your script, HTTP request or prompt run automatically at an event: before a tool is used, after it, when a session starts or stops, and more [6].
MCP servers
The way to connect Claude Code to your own tools and data through the Model Context Protocol [2].
Plugins and marketplaces
A plugin bundles skills, hooks, subagents and MCP servers. A marketplace is a list of plugins a team can install from [2] [7].
Permission modes and plan mode
Permission modes control whether Claude asks before acting. Plan mode is read-only: it researches and proposes before anything is changed [7] [5].
Sandboxing
A built-in sandbox restricts which files and network hosts its shell commands can reach [7].
Running it from code
The Agent SDK runs Claude Code programmatically from the command line, Python or TypeScript [7].
Many at once
Agent teams coordinate several instances with shared tasks, worktrees keep parallel sessions from colliding, and dynamic workflows run many subagents from a script [7].
Sources
Where this comes from
Read 2026-11-15, unless a source gives its own date. A number in brackets points to the page a statement was read on. Sentences with no number, the small demonstrations and the tools are this center's own explanations.