Ship an app-builder prompt that actually builds.
Turns an app idea in plain words into the product, pages, data model, interactions, and what to build first.
Inside: The product / pages / data-model / interactions / build-order scaffold.
Install it in one line, or paste it in.
~/.claude/skills/ and runs automatically when it is relevant.Connect your context. Set it to your motion.
| Set this | What it is | Example |
|---|---|---|
| Target builder | The tool you will paste into | Lovablev0BoltReplit |
| Stack preference | Any framework or service you want it to use | Next.jsSupabaseTailwind |
| Look and feel | Your default visual direction | "Cleanlots of whitespaceone accent color" |
Everything the skill does, in full.
Takes an app idea in plain words and returns a build prompt structured the way AI app builders want it: the one-sentence product, the pages, the data model, the core interactions, the styling, and what to build first. A vague prompt gives you a vague app. This one gives the builder a spec.
- 11. Name the product in one sentence: what it is and who it is for.
- List the pages or screens, each with its job in a phrase.
- Define the data model: the few objects and their fields. Builders go sideways without this.
- Spell out the core interactions: what happens when the user clicks the main button.
- Set the styling: the visual direction and the one or two brand colors.
- Sequence the build: "Start with the data model and the main page. Get that working before anything else." This stops the builder from trying to do everything at once and breaking.
- Add guardrails: no auth in v1, mock the data, keep it one page until it works.
Build a single-page web app: an outreach-experiment tracker for a salesperson.
Pages: one main page, a table of experiments with an "add experiment" form.
Data model:
- Experiment { name, channel, hypothesis, sent, replies, status, notes }
Interactions:
- "Add experiment" opens a form, saves a row to the table
- Each row shows a computed reply rate (replies / sent)
- A status pill: running, won, killed
Styling: clean, lots of whitespace, one orange accent, system font.
Build order: first the data model and the table with mock rows. Get that
rendering. Then the add form. Then the computed reply rate. No auth in v1.
Where an operator takes this next.
The prompt is written. Here is where an operator takes it once the manual version proves out.
The prompt gets you a working v1; the wiring is what makes it the tool you actually use.
Once the app builder ships a working version, push it straight to Vercel so you are testing on a real URL instead of a preview pane.
Wire the tracker to pull from a live sheet or Airtable base once the data model holds up, so v1 becomes the real tool, not a demo.
Have a scheduled Claude task log your actual outreach sends into the tracker's data store daily, so the experiment table fills itself.
One skill is the on-ramp.
A single skill does one job. Chained into a playbook, or run as a full build, it becomes a system. Here is where this one plugs in.