This is your first real loop with the agent: create a project, wire it to Git, and give the agent its instruction file.
Throughout this page every step shows two paths: via the agent (Desktop app or VS Code — you describe, the agent does, you review) and via the terminal (you run the commands yourself). Both are legitimate; a developer ends up using both. Whichever you pick, the rule from Week 1's mental model applies: read and understand every command before it runs.
https://www.youtube.com/watch?v=gbetp6D7J_Q&list=PLmWCw1CzcFilebjK89WLb5cAvM8K0cLB3&index=4
Via the agent: create a new folder (Finder / Explorer, or ask the agent to create it), open it in the Claude Desktop app (Code section) or in VS Code, and ask the agent to propose a simple structure for a small static site (or a project matching your main track). Ask it to explain why it structured things that way — don't accept a layout you can't explain.
Via the terminal:
mkdir my-first-agent-project
cd my-first-agent-project
…then start claude from that folder and continue the same way.
Commit and branch theory is covered in your main track; here we only connect the project.
Via the agent: ask it to initialize Git, then create an empty repository on GitHub (no README) and give the agent the URL: "connect this repo as origin and push the first commit." The agent will propose the commands — read each one before approving. This is exactly the review habit that Week 4 builds on.
Via the terminal:
git init
git remote add origin <your-repo-url>
git add .
git commit -m "first commit"
git push -u origin main
Either way, refresh the GitHub page — your project is there. This loop (change → commit → push) is the safety net for everything the agent will do later: any change can be reviewed as a diff and rolled back.
https://www.youtube.com/watch?v=O0FGCxkHM-U&list=PLmWCw1CzcFilebjK89WLb5cAvM8K0cLB3&index=5
CLAUDE.md is a markdown file in the project root that Claude Code reads at the start of every session. It is how you turn a generic agent into your project's agent: stack, conventions, commands, and boundaries live here. It works the same in Desktop, VS Code, and the terminal.
What to put in it:
A minimal example for this week's project:
# My First Agent Project
## Stack
Static site: HTML + CSS, no frameworks.
## Commands
Open index.html in the browser to check the result.
## Conventions
- Semantic HTML tags
- CSS in styles.css, no inline styles
- Comments in English
## Do not
- Do not add JS frameworks or build tools
- Do not commit files in /notes
Write yours (or draft it with the agent and edit), then test that it works: ask the agent to do something that violates one of your rules — e.g. "add this style inline." It should refuse or flag the rule. Seeing the file actually steer the agent is the point of this exercise.
<aside> 💭
CLAUDE.md is a living document. Every time the agent does something you didn't want, ask: "should this be a rule in CLAUDE.md?"
</aside>
The HackYourFuture curriculum is licensed under CC BY-NC-SA 4.0 *https://hackyourfuture.net/*
Built with ❤️ by the HackYourFuture community · Thank you, contributors
Found a mistake or have a suggestion? Let us know in the feedback form.