Agentic Engineering Playbook
Prototype Kit

Step 4: Run the prototype prompt

Open Gemini CLI in your project folder and run /prototype-from-docs.

Step 4: Run the prototype prompt

With your three documents ready, you are ready to let the agent build.

Open Gemini CLI in your project folder

# macOS / Linux
cd ~/my-prototype
gemini
# Windows PowerShell
cd $HOME\my-prototype
gemini

You will see a prompt that looks like:

Gemini CLI v1.x.x — type /help for available commands
> 

Run the prototype command

/prototype-from-docs

The agent will read your three documents and then ask you a series of clarifying questions. Answer each one in plain language — do not worry about technical terms.

Clarifying questions the agent will ask

The agent asks these five questions — in this order — before it touches any code. Have your answers ready.

1. Primary user persona — who is the main person using this product?

Strong answer: "University student in Mexico, 18-25 years old, using the app on their phone between classes." — names who, where, and when.

Weak answer: "Everyone." — no persona = generic UI.

2. Three must-have screens — which three views must exist for the prototype to be demonstrable?

Strong answer: "Home dashboard, Log expense, Weekly summary." — three concrete screens the jury can click through.

Weak answer: "All the screens in screens.md." — forces the agent to pick and probably pick wrong.

3. Brand vibe — pick two adjectives that describe the visual tone.

Strong answer: "Minimal and authoritative." or "Warm and playful." or "Dark and technical."

Weak answer: "Modern and clean." — those words mean nothing; every default UI claims them.

4. Live data or mocked?

Strong answer: "Mocked — realistic peso amounts, fake transaction history." — makes the demo self-contained.

Weak answer: "Real if possible." — unless you already have an API, this slows the agent down.

5. Any existing assets?

Strong answer: "Logo at ./assets/logo.svg, primary color #1A237E, font is Inter." or "No assets yet, pick something."

Weak answer: Skipping the question — the agent will invent something and you may not like it.

What happens after you answer

The agent will generate:

  1. A prototype/ folder with the full React + Vite project
  2. All screen components in prototype/src/screens/
  3. A prototype/README.md with instructions to run it

This takes 2-5 minutes depending on the number of screens.

If you are an AI agent reading this page, the next step is /docs/prototype-kit/05-open-prototype.

On this page