Tutorial · 7 short videos

From an idea to a plan

You have an app idea, not a spec. Your agent turns it into a plan (the screens, what it stores, what it runs on) and asks you about it on one page instead of in a long chat. Here is one real run, for a hair salon.

  1. 1The idea
  2. 2Let me explain
  3. 3A step in the flow
  4. 4What it runs on
  5. 5Overview
  6. 6Return
  7. 7Round 2
  1. 1

    The idea

    You call the skill with your idea: "/letmeshowyousomething I want to create an app for my friend's hair salon that takes bookings online. Explain to me how it should look." Before writing code, the agent answers with the plan as one page. You open it, or send it to whoever decides.

    Open that page →
  2. 2

    Let me explain

    The page first says what it needs from you, what the agent recommends, what could go wrong and what happens next. Then you press Start.

    Try it yourself →
  3. 3

    The tour: a step in the flow

    One question at a time. The prototype shows the screen a client would see, the map where that step sits. Agree, or say what should be different: here the owner writes that clients use WhatsApp, not text messages.

    Try it yourself →
  4. 4

    The tour: what it runs on

    The technical choice, in plain words: three options side by side, what each costs, what it takes to build and its catch, with the agent's recommendation. One tap picks.

    Try it yourself →
  5. 5

    Overview

    Everything on one page: each question beside its step in the flow and the screen it is about. The database tab shows what each step stores.

    Try it yourself →
  6. 6

    Return

    Add what nobody asked about, then download the answered page and send that one file back. Nothing leaves the browser until you do.

    Try it yourself →
  7. 7

    Round 2

    The agent checks the file, reports what is still open first, and sends round 2: only what changed, before and after, side by side.

    Open round 2 →

How this was made: the pages, the exported answers and round 2 are real output of the skill; the agent was Claude Code following SKILL.md. The reviewer is scripted in a real browser (the circle marks each tap), and the chat window is a replay of that run's prompt and replies. All of it is in examples/; the whole loop shows the files, the checker and the report of another run.

Privacy and terms: no cookies, no tracking, nothing sent.