The Vibe-Coding Kit

The method, not
just the idea

An idea is a starting line. The kit is the sequenced path from it to a running MVP: ten phases, a written prompt for each, the context files that keep your agent on scope, and a checkpoint that says when it truly counts as shipped.

What is in every kit

01

The sequence

Ten phases in the order that actually works, so you never have to guess what tonight's work is.

02

The prompts

A written prompt per phase, tuned to the idea and the stack, and ready to hand to your agent.

03

The context files

AGENTS.md, a one-page PRD, and a schema, so the agent builds this product and not a generic one.

04

The checkpoints

Each phase names what must be true before it counts as shipped.

05

The blockers

Pre-written fixes for the exact places where builders tend to stall on a build of this particular kind.

06

The launch plan

Phases nine and ten are distribution, because a booted MVP on its own is not yet a business.

What one phase looks like

You never face the whole build. You face tonight: one phase, one prompt, and a checkpoint that tells you when it is genuinely done. Then the next one unlocks.

This is the difference between a prompt library and a method.

Phase 4 of 10Now
Data model and sign-in
The prompt
Create the schema from the PRD: users, products, ingredients, scans. Wire email sign-in. Enforce access rules so one account cannot read another's scans.
Checkpoint
Two accounts cannot see each other's data
Passes the checkpoint, and the next phase opens

The order is the expertise

1

Data before features

Get the model and sign-in right first. Every feature after it is cheap. Do it late and you end up rewriting almost everything.

2

One core loop before breadth

Ship the single thing the product is for, end to end, before adding a second. Most builds die somewhere in adding that second thing.

3

Money before polish

Wire monetization while the code is small. Bolting it on later is the tax that kills momentum.

4

Users before perfection

Phase ten is the first ten users. A perfect app nobody sees is the most common way this ends.

Written for where you will actually stall

Every kit ships fixes for the specific places this kind of build breaks, written before you hit them. Not generic advice, the actual failure of this phase on this stack.

+Access rules leak: the wrong user can read a scan
+Seed data looks unrealistic and hides a bug
+Sign-in works locally but breaks on deploy

Proven before it reaches you

No kit publishes until it has been built end to end on a real stack until the MVP boots, and that run is logged. If it still fails you, the ship guarantee covers it.

Read the guarantee

You keep all of it

Your repo

The code is yours from the first commit. Nothing of ours runs inside it.

Your stack

Your Vercel, your Supabase, your keys. We never hold them.

Your revenue

Whatever you charge is yours. We take no cut of anything you ship.

Frequently asked

Do I have to use the prompts as written?

No. They are a strong starting point, not a cage. Most builders paste them, then steer from there.

Which agent is it written for?

Claude Code and Cursor primarily. The files are plain markdown, so any capable agent can read them.

What if my agent does it differently?

Fine, as long as the phase's checkpoint passes. The kit cares that it works, not that it matches a diff.

Do I own the code?

Completely. It is your repo, your stack, your product. There is no runtime of ours in what you ship.

What if a phase is broken?

Report it and it gets a 48-hour fix priority, backed by the ship guarantee we publish.

How long does a kit take?

Difficulty is an honest estimate of evenings. A 2 out of 5 is a real two-week build for one person with an agent.

Start free

Open a kit and see the whole path

Today's idea is unlocked in full, free, with no account. The method is the same on every one.

No card required · your code stays yours