BP, Bella Pivo signaturefolders, not code
vibe with vibe coding · episode 01

It was never code. It’s folders.

If Claude keeps forgetting you, drifting, or making things up, you’re not bad at prompting. It’s where you’re talking to it.

build your website what’s the idea?
ep 01 · the cheat sheet

one project = one folder. one folder = one Claude that actually knows you.

▸ ~/Desktop/my-website
▸ ~/Desktop/summer-class
▸ ~/Desktop/job-hunt
1

The whole idea, in three cards

The app and the terminal run the same model. They are not the same setup.

the problem

The app forgets you

It runs on a generic instruction file written for everybody. You tune it all chat, get it perfect, and it drifts.

the fix

One project, one folder

Every thing you’re working on gets its own folder on your desktop. Open Claude inside it and it already knows the job.

~/Desktop/website-redesign~/Desktop/summer-class
the magic

CLAUDE.md is your rules

One file, read every single time, in plain English:

“You are a web designer helping me redesign my site. Keep it warm, never corporate. Never publish without asking me first.”

“Imagine that the folders are exactly like the project system that lives in the desktop app. But it hallucinates more when it’s in the desktop app… when you use terminal, you are making the Claude brain very specific to you.”

the whole idea, in one line
2

You only need to ask yourself three questions

Every system you’ll ever build is the same three steps. Engineers call it ETL: extract, transform, load. It’s the pattern under almost every data system, including the AI tools you already use. Once you see it, you start thinking like a systems engineer.

01 · extract

What’s the data?

your calendar + to-do list
02 · transform · the only AI part

What should it become?

drop what’s done, pick your top 3
03 · load

Where does it show up?

a text at 8am

↑ one example: your morning.

“That’s kind of just the system you need to understand. What’s the data, what do you want it to look like, and then where do you want it to show up so that you can best interact with it. That’s kind of it.”

the frame, unedited
3

Start your first project

Build a badass personal website in five minutes. Fill this in, copy your prompt, paste it into Claude with your resume or LinkedIn PDF attached. It saves as you type, only in your browser.

1/3where you’re headed
I’d talk to about
Topics you keep coming back to. Hobbies count.
2/3about you
Your job, side projects, classes, clubs. It all counts.
That’s how LinkedIn found me. It was a cold email.
3/3the look
real palettes, so it doesn’t look AI-made.
your prompt, writing itself

      
then:
  1. Open claude.ai (free works). Attach your resume, or your LinkedIn as a PDF (on your profile: More → Save to PDF).
  2. Paste your prompt and send. Your site shows up on the side. Don’t love something? Just tell it.
  3. Hit Share to publish it and get a link, or download the file.
  4. Want me to react to it in my content? Send me your site →
want the terminal version?install it, what goes in the folder, and how to not get overwhelmed. tap to open

1 · Install it (about ten minutes)

1 · install it

$ curl -fsSL https://claude.ai/install.sh | bash
installs Claude Code. no npm, no homebrew needed
$ claude
opens a browser to sign in, once

If claude comes back “command not found” right after it installs, that’s a PATH thing, not a you thing. Quit terminal, reopen it, try again.

2 · make your first project

$ cd ~/Desktop
$ mkdir website-redesign && cd website-redesign
$ claude

Then type, in plain English:

“Set this folder up as a project for redesigning my website. Write a CLAUDE.md that says who you are and what I care about. Ask me anything you need.”

one thing to decide first: which account you sign in with

  • Work or school account. More protections on paper, but your usage is likely reported back to the institution.
  • Personal paid account. Yours alone. Nobody at work sees what you’re building.
  • If you’re making anything personal, or anything you might one day own, use the personal one.

2 · What goes inside the folder

Only CLAUDE.md gets read every time. Everything else waits until it’s relevant.

~/Desktop/summer-class/your project, on your desktop
├── CLAUDE.mdread EVERY time · who it is, what you want, your rules
├── .claude/the engine room
│   ├── skills/a saved way of doing one job · runs when triggered
│   │   └── build-a-lecture.md“when I say make a lecture, do these steps”
│   ├── agents/specialists · opened only when you call one
│   └── settings.jsonpermissions · what it may run without asking
├── memory/facts worth keeping · pulled in when relevant
├── syllabus/your actual stuff
└── lectures/your actual stuff
always read read only when needed your files, untouched until you ask

if you’ve never opened terminal

  • Terminal is just a text box for your computer. That’s the whole thing. There’s no hidden skill you’re missing.
  • Press ⌘ + space, type terminal, hit enter.
  • You never have to learn commands. Claude writes them, you approve them.

skills and memory, plainly

A skill is a job you do the same way every time, written down once.

Memory on its own is unreliable. Claude keeps highlights, not every message, so build your own memory folder and tell it what’s worth saving.

3 · You don’t build the app. You build the leaves.

too big

“Redesign my whole website”

Nowhere to start. So you don’t start.

a leaf

“Write the new About page in my voice, using these three old bios”

One task. Doable tonight. Then the next leaf.

The tree was never the thing you build. It’s just the map.

The tree was never the thing you build. It’s just the map.