Module 15 — How Claude Code Works (It All Comes Together)
🎯 Goal: Put every "under the hood" idea together into a clear mental picture of what's actually happening each time you talk to Claude Code. ⏱️ Time: ~20 min of reading. The satisfying payoff module.
You've met the pieces: a frozen brain (13), tools, context, and skills (14). Here's how they assemble into the tool on your laptop.
The one-sentence version
Claude Code is the Claude model (the brain) wrapped in software that gives it hands (tools), manages its desk (context), and hands it the right playbooks (skills) — so you can get real work done just by talking.
The model alone is a brain in a sealed room. Everything that makes it useful on your actual files is the wrapper Anthropic built around it. That wrapper is Claude Code.
What's in the box
| Piece | What it is | Which module |
|---|---|---|
| The brain | A Claude model — predicts text, reasons, decides what to do | 13 |
| The hands (tools) | Read/write files, run terminal commands, search, and more | 14 |
| The desk (context) | Your messages, files read, and tool results — kept tidy | 14 |
| The playbooks (skills) | Specialist instructions loaded when your task matches | 14 |
The memory note (CLAUDE.md) |
Standing instructions kept on the desk every time | 11 |
| You | The director — you set the goal and check the results | every module |
The loop: what happens when you press Enter
This is the heart of it. When you type a request, Claude Code runs a little cycle — often called an agent loop — over and over until the job is done:
- You ask. Your request goes onto the desk (the context).
- The brain reads the desk and decides. It might answer directly, or it might realize it needs to do something first — read a file, run a command.
- It reaches for a tool. It says, in effect, "read every PDF in this folder."
- The tool runs, the result comes back onto the desk.
- The brain reads the new information and decides again. Done? Or another step needed?
- Repeat steps 2–5 until the task is complete — then it replies to you.
Analogy: A chef (the brain) at a counter (the desk). You give the order. The chef grabs a utensil (tool), checks the result, grabs another, tastes, adjusts — looping until the dish is ready — then plates it for you. You didn't describe each knife stroke; you described the dish.
That loop, running invisibly, is why a single sentence from you — "pull the client names and dates out of these 50 PDFs into a spreadsheet" — turns into the model reading each file, writing a program, running it, checking it, and handing you the result.
You've been using all of this already
Every habit this course taught maps onto a piece of the machinery. This is the "aha":
| What you do | What's really happening |
|---|---|
| Write a clear, specific request | Prompt engineering — a good brief for the brain |
Keep one task per chat; use /clear |
Context engineering — keeping the desk tidy |
| Approve/deny those permission prompts | You're the safety gate on the tools (hands) before they act |
Keep a CLAUDE.md in a project |
Pinning standing facts onto the desk every time |
| Claude handles a PDF/Excel job smoothly | A skill (playbook) loaded automatically |
| Verify the output before trusting it | Catching the brain's confident-but-wrong moments |
| Work on copies, commit before big changes | Safety nets, because tools take real actions |
None of it was arbitrary. It all follows from how the thing actually works.
Why this gives such a good "out of the box" experience
Plenty of tools give you a raw model in a chat box. Claude Code feels different because Anthropic pre-wired the whole machine for you:
- The brain is a strong, current Claude model — chosen for you.
- The hands (file tools, terminal, search) come ready, with safety prompts.
- The desk is managed automatically — it keeps relevant context and trims the rest so you don't think about limits.
- Good skills and defaults are built in, so common jobs just work.
- The agent loop runs itself — you don't orchestrate steps.
So you, a non-programmer, get to operate at the level of "here's what I want" — and a great deal of careful engineering underneath turns that into done work. That's the magic, and now it isn't magic to you: it's a brain, some hands, a tidy desk, the right playbooks, and a loop — with you directing.
✅ Takeaways — and you've finished the deep dive
- Claude Code = a Claude model + tools + context management + skills, wrapped so you can just talk.
- Each request runs an agent loop: think → use a tool → read the result → repeat → answer.
- Every good habit you learned is you working with that machinery — directing the brain, guarding the hands, tidying the desk, and checking the result.
- The "out of the box" feel is real engineering, not magic — and you now understand it well enough to use it with genuine confidence.
That's the whole picture. Bring it back to the practical: Recipe Cookbook · Making It a Habit · Start over from the top