Scaffold a project with Vibe Code

Use the Vibe CLI to generate a full project from a natural-language description, review every change before it's applied, and run the result.

• Natural-language scaffolding: describe what you want and the CLI creates the files.
• Step-by-step confirmation: Vibe Code splits large tasks into steps and asks for approval.
• Full preview: see every file change before it's written to disk.

By the end you'll have a working project scaffolded from a single prompt.

Time to complete: ~10 minutes

• Vibe CLI installed and configured. If you haven't done that yet, run the Install the Vibe CLI quickstart first.

Start in a fresh directory so the CLI generates the project from scratch. Launch the CLI with the ask agent so you can review each change:

mkdir my-project && cd my-project
vibe --agent ask

mkdir my-project && cd my-project
vibe --agent ask

The CLI starts with an empty context, ready to scaffold.

Give the CLI a clear description. Be specific about language, framework, and key features. For example:

Create a Python Flask API with two endpoints: a /health endpoint that returns {"status": "ok"} and a /predict endpoint that accepts a JSON body with a "text" field and returns {"label": "positive", "score": 0.95}. Include a requirements.txt and a README.

Vibe Code breaks the request into steps and begins generating files. Before creating or editing any file, it shows a preview.

For each step, the CLI displays the files it plans to create or modify, the content of each file, and a confirmation prompt. Press one of the approval keys:

• 1 or y to allow this edit once.
• 2 to allow this kind of edit for the rest of the session.
• 3 to always allow it.
• 4, n, or Escape to deny it.

You can also move between options with the arrow keys and press Enter, or click an option.

After each approved step, the CLI shows the result.

Continue approving steps until the project is complete.

Check that the project was created:

ls -la

ls -la

You should see the files Vibe Code generated (for example, app.py, requirements.txt, README.md).

Ask the CLI to start the server and test the endpoints for you:

Install the dependencies, start the server, and run a curl against /health and /predict to verify they work.

Or run it yourself:

pip install -r requirements.txt
python app.py

pip install -r requirements.txt
python app.py

Test the endpoints:

curl http://localhost:5000/health
curl -X POST http://localhost:5000/predict \
  -H "Content-Type: application/json" \
  -d '{"text": "Mistral is great"}'

curl http://localhost:5000/health
curl -X POST http://localhost:5000/predict \
  -H "Content-Type: application/json" \
  -d '{"text": "Mistral is great"}'

Both endpoints should return valid JSON responses.