The map of extensions: when each one is useful
This lesson is still being tested: it improves with learners' feedback. If something is unclear, it is the lesson that needs fixing, not you.
Follow this lesson with Claude Code
With the course companion installed, open Claude Code in your working folder and type:
/formation-ia:suivre m07-l01 enIt reads the up-to-date lesson on this site and guides you step by step, without changing anything without your consent. Install the companion. Claude Code is not included in the free plan: an eligible Claude subscription, or a Console account billed by usage for the terminal.
By the end of this lesson, you will be able to
- tell apart the main families of Claude Code extensions: instructions, skill, connector (MCP), hook, subagent, plugin;
- start from a real need to choose the simplest one that fits;
- know what to check before installing anything.
Why it matters to you
Every day you come across “incredible skills” and “agents that do everything”. Yet most needs are solved with the simplest extension, and some extensions give a lot of power to strangers. This map helps you choose without getting lost or putting yourself at risk.
Before you start
| You need | Claude Code, permissions (module 4) and CLAUDE.md (module 5) |
| Duration | 30 minutes |
| Cost | Included in your subscription for this lesson |
Understand it in 3 minutes
From the simplest to the most powerful (the course’s own ranking, close to the order in which most teams add these extensions, according to the official documentation checked on 24/10/2026):
| Extension | What it is | Example of a need | What it can do |
|---|---|---|---|
| Instructions (CLAUDE.md) | Your instructions, read again at every session | “Keep your answers short, ask before changing anything” | Guide the behavior |
| Skill | A folder with a SKILL.md file: instructions for a specific task, loaded only when they are needed, or called with /name | “Always write my meeting notes the same way” | Guide a task; some also run commands or allow tools |
| Connector (MCP server) | A bridge to an outside tool or outside data | “Read my lessons on the course website”, “Access my calendar” | Read, and sometimes act, in another service |
| Subagent | A specialized assistant that Claude launches for a task, with its own conversation | “Do a search without cluttering my conversation” | Work separately, then report back |
| Hook | A command that fires automatically at a precise moment (for example before an action) | “Always block this action”, “Let me know when Claude is waiting” | Act at a moment fixed in advance, without the AI choosing to do it |
| Plugin | A package that can contain several of the items above, installed from a catalog | “Install a whole set of tools at once” | Everything its items contain |
The course’s rule: take the simplest extension that is enough. An instruction in CLAUDE.md before a skill, a skill before a connector, and a hook only if you understand exactly what it runs.
Before installing an extension that does not come from you, three questions:
- Who made it, and where does it come from?
- What can it do: read, write, send, run commands, access which accounts?
- How much does it cost, to install and to use?
Do it together
Example: this course’s companion
The course’s Claude Code companion is itself a plugin that contains two items:
- a skill named “suivre” (French for “follow”): the instructor’s guidelines;
- an MCP connector: the address of the course server, which only reads lessons.
It contains no hook and no added program: its skill is launched by typing /formation-ia:suivre. This is the kind of check you will do for any extension. Its “Continue in Claude Code” page says what it does, what it does not do and what it sends.
Ask Claude Code to explain an extension to you
In your working folder, in Manual mode (Claude Code reads without asking in this folder; it usually asks for your approval to read elsewhere, except with some read-only commands, and to edit a file or run a command that does something other than read, unless you already allowed that action):
Explain to me in 10 lines the difference between a skill, an MCP connector, a hook and a plugin in Claude Code, with a concrete example for each, suited to a beginner.
Compare its answer with the table in this lesson. If there is a difference, the official documentation is right.
Your turn
Your decision sheet. Take 3 real needs (for example from your “My requests” note or your to-do list) and, for each one, fill in a line:
| My need | The simplest extension | Why not a more powerful one |
|---|---|---|
Check your work
It worked if:
- each of your 3 needs has a chosen extension, with a reason;
- at least one need is solved simply with CLAUDE.md or a skill;
- you can name the 3 questions to ask before installing an extension.
Common mistakes:
- installing a plugin “just to see” without looking at what it contains;
- using a hook for something an instruction is enough to get;
- thinking an extension is free because installing it is: some connectors go through paid services.
Keep it and pick it up later
- Keep your decision sheet: the need that calls for a skill will be the exercise of the next lesson.
Going further
optional
- The official documentation, in English: code.claude.com/docs/en/skills, /mcp, /hooks-guide, /sub-agents, /plugins/marketplace-reference.
New words
5 words explained
| Word | What it means |
|---|---|
| Extension | What you add to Claude Code to give it instructions, access or automations |
| Skill | Instructions for a task, loaded when they are needed |
| Connector, MCP server | A bridge to an outside tool or outside data |
| Hook | A command that fires by itself at a precise moment |
| Plugin | A package of extensions installed from a catalog |
Check your understanding
3 questions, no grade. An explanation appears after each answer.
What's next?
Lesson 2 · Your first skill: turning a request you repeat into a skill.
Module 7:
✏️ Do the exercisesMy space: your progress, your exercises and your last lesson.
Sources and checks
6 sources
Date, version, sources Written on 24/10/2026. Checks, all in the official Claude Code documentation on 24/10/2026:
- CLAUDE.md read again at every session, instructions and not enforced rules: code.claude.com/docs/en/memory;
- skills, loaded when they are needed, called with /name: code.claude.com/docs/en/skills;
- MCP servers: code.claude.com/docs/en/mcp;
- hooks, commands that fire automatically: code.claude.com/docs/en/hooks-guide;
- subagents, with their own context: code.claude.com/docs/en/sub-agents;
- plugins and catalogs: code.claude.com/docs/en/plugins/marketplace-reference;
- the order in which most teams add these extensions: code.claude.com/docs/en/features-overview.
Sources checked: