---
title: "Install Claude Code"
description: "The terminal agent — and the moment your skill becomes operational."
order: 22
duration: "10 min"
chapter: "06-delegation-not-prompting"
type: lesson
---

## Install Claude Code

This chapter introduces the third tool in the progression: Claude Code, a terminal-native agent. You have used Claude.ai in a browser, the Claude mobile app for capture, and Cowork for data analysis. Claude Code is different from all of them. It has full access to your filesystem, your shell, your git history, and every tool on your machine.

Install it:

```bash
curl -fsSL https://claude.ai/install.sh | bash
```

(On Windows, use PowerShell: `irm https://claude.ai/install.ps1 | iex`. Homebrew users can `brew install --cask claude-code`.)

Then open a terminal, navigate to a project directory — any directory where you have files — and run:

```bash
claude
```

That is it. Claude Code starts and you are in a conversation, but unlike the browser or mobile app, this conversation is grounded in your actual environment. Claude can see the files around it.

<div class="exercise">
  <div class="callout-label">Try This</div>
  <p>Install Claude Code and run <code>claude</code> in a project directory. Ask: "What files are in this directory?" Claude will actually look — it reads your filesystem, not guessing from context. Try "What's in the README?" or "Show me the most recently modified files." Notice the difference: Claude is not just responding from general knowledge. It is looking at your actual machine.</p>
</div>

Cornwall Market: Sarah opens Claude Code in the directory where she has been building her project. She asks, "What files are here?" Claude lists everything — including the `.claude/skills/` folder where she saved her categorization SKILL.md from Chapter 04. Claude sees it. Claude can read it. And that changes everything.

Notice what also happens: if this project has a `CLAUDE.md` file (the project memory pattern from Chapter 04), Claude reads that too. Every session starts by loading CLAUDE.md and any skills in the project. The distillation pipeline is already working — your accumulated knowledge from previous conversations persists automatically.

## Your skill meets your environment

Take the SKILL.md you wrote in Chapter 04. Place it in a `.claude/skills/` directory inside your project:

```
your-project/
  .claude/
    skills/
      categorization/
        SKILL.md
```

Start a new Claude Code session. Claude discovers the skill automatically and loads it the moment a question matches its description — without you mentioning it, without you pasting anything into the conversation.

Now ask Claude a domain-specific question. One that requires the knowledge in your skill.

Cornwall Market: Sarah starts a new Claude Code session and asks, "How should I categorize a $6,000 renovation payment for Chen's Bakery in Q4?"

Claude responds using the skill. It knows that Chen's Bakery Supplies is account 5120, not to be confused with Chen's Produce at 5100. It knows that building improvements go to 6250 and need to be tracked by store location. It knows that Q4 means cross-referencing contractor payments against year-to-date totals for 1099 reporting, and a $6,000 payment definitely crosses the threshold. It flags all of this without Sarah having to explain any of it.

This is the moment the skill becomes real. It is not a document anymore — it is operational knowledge that Claude uses automatically.

And the distillation continues in real time. When Sarah corrects Claude — "No, that should be 6250 not 6100, it's a building improvement" — she can say "update CLAUDE.md with that correction" or "update the categorization skill." The correction moves from conversation into persistent knowledge. Every interaction is a chance to refine the system.

This is the meta-pattern: **every discovery gets documented back into the system.** Future sessions inherit all previous learnings. The project teaches Claude how to work on it.

<div class="exercise">
  <div class="callout-label">Try This</div>
  <p>Place your SKILL.md from Chapter 04 into a <code>.claude/skills/</code> directory in your project. Start a new Claude Code session in that directory. Ask a domain-specific question — something that requires the knowledge in your skill. Does Claude reference your rules? Does it get the details right? If it misses something, update your SKILL.md and try again.</p>
</div>

## Check your understanding

Answer in your own words — write it down before opening the key. Your tutor grades against the criteria and generates fresh variants on retries.

**Q1.** Your skill file exists but a new Claude Code session ignores it. List the two things you would check, in order, and why each matters.

<details>
<summary>Answer key — attempt every question first</summary>

## Answer key

### Q1

**Model answer:** First, location: the skill must be in .claude/skills/<name>/SKILL.md inside the project (or ~/.claude/skills/) — Claude only scans known locations. Second, the description trigger: if it doesn't match the phrasing of your question, the skill never loads. Location determines discovery; description determines activation.

**Pass criteria:** both checks present (path/location and description matching); ordering or role of each explained

</details>


**Next:** [The Delegation Shift](/mastering-claude/06-delegation-not-prompting/23-the-delegation-shift/)