You know that moment. You get an idea, you’re ready to fire up Claude or Copilot and just go…

But first, the dance.

mkdir... uh... book-thing? reading-app? Whatever, pick something. cd into it. git init. Now you’re staring at an empty folder trying to explain to an AI what you want to build, when you’re not even sure yourself.

Two weeks later you have 600 lines of generated code and you forgot what you were even trying to make.

I had it backwards.

I was rushing through the thinking - what am I building, what’s the scope, when do I stop - and spending time on the clerical stuff. Making folders, initializing git, creating the same files I always need.

So I flipped it. Slow down the thinking. Speed up the boring setup. Get to a place where I’m actually ready to talk to Claude Code.

That’s why I made mint-cli.

Here’s what it does:

You type mint. It asks what you’re building. Suggests some names. You pick one.

Now you’re in your project. Git’s initialized. And a file called DECISIONS.md is already open.

This is where I slow down.

Before I write any code, before I prompt any AI, I answer three questions:

• Why am I building this?
• What am I not building?
• When is this done?

That’s it. Three prompts.

The middle one is the most important. What I’m not building is permission to ignore stuff. It keeps the scope small. It keeps me honest.

And now when I point Claude Code at this project, it has context. It knows what I want, what I don’t want, and when to stop. We’re starting from the same place.

30 seconds of setup. A few minutes of actual thinking. And now I’m ready to build.

mint-cli on GitHub →