Free · No fee, ever · No email required to read

A free course on Claude in your terminal

For people who write code and still use AI in a browser tab — copying functions in, answers out.

No fee. Nothing to buy. Read it right here.

Watch first · 4 min

Installing it, logging in, picking a model, and building a page from one instruction. If you've never opened a terminal for this, start here — then read the lesson below, which is about using it well.

Lesson 1

Give it the repo

You have a bug. You open a chat, paste the function, describe what's going wrong. It gives you something reasonable, you paste it back, and it fails differently. So you paste in the new error. And the file it hasn't seen. And the test output.

Twenty minutes later you're the courier between two things that could just talk to each other.

The change

Run Claude in the folder your code is in, and it reads the files itself.

cd your-project

claude

Instead of pasting a function, you say:

the login test is failing. figure out why.

It opens the test. It opens the file the test imports. It runs the test and reads the real error — not your summary of the error. If the cause is three files away, it goes and looks.

You never described the bug. You pointed at it.

Why that's a bigger deal than it sounds

Pasting a snippet means you already picked which file matters, which twenty lines of it matter, and what the error "basically says." You made those calls before you knew what the bug was.

That's why the answers are so often confidently aimed at the wrong thing. You can't paste the part you don't yet know is the problem. Give it the repo and you stop pre-deciding — that's the entire lesson.

Try it on something real

A toy example will look unimpressive, because a toy fits in a chat box. Pick something with sprawl: a test failing for a reason you haven't chased down, a config change that has to land in four files, a dependency bump that breaks something two layers away.

It reads and edits your files, runs commands, runs your tests, and reads what came back — so it can check its own work. It asks before it changes anything on your machine, and you can say no.

One thing to be careful about

It can edit files and run commands, so start on a project where that's fine. Two habits and it stops being a worry:

  • Work in git, so any edit is one command away from being undone
  • Read what it proposes before you accept it — especially anything that deletes, force-pushes, or touches credentials

And don't hand it real keys or real customer data. True of every AI tool, truer of one that can run things.

What's next

Lesson 2 is what you say after the first answer. Most people take it or bin it. Both are a mistake — the second message is usually where the actual work happens.

This course is free and it stays free. Lessons go up here whether or not you leave an email — the list below just tells you when a new one is live, and that's all it does.

Questions?

Email me directly and I'll answer.

armani@hamtek.dev