Can I vibecode Arcade?

Build me a click-through demo maker to replace Arcade. Requirements:

- Node.js 22 + TypeScript in one repo: a Chrome extension (Manifest V3, built with Vite) and a Fastify
  server on http://127.0.0.1:4720 that also serves the editor.
- Recording: the toolbar button starts a session. A content script catches clicks on pointerdown in the
  capture phase and sends the viewport-relative click position, the element's label (aria-label, text or
  placeholder, 60 characters) and the URL; the service worker calls `chrome.tabs.captureVisibleTab` and
  posts the PNG. Never record typed text and skip password fields.
- Storage: one folder per demo in ~/Demos/<slug>/ with demo.json and steps/01.png, 02.png and so on.
- Editor: reorder and delete steps; per step move the hotspot (a rectangle in percent of the image), set a
  callout title, text and position, add a chapter label, crop, and draw blur boxes that are applied at
  export while the originals stay untouched.
- Player: one static `player.html` with vanilla JS inside a browser frame: the screenshot, a pulsing
  hotspot and a callout. Clicking the hotspot or pressing the right arrow advances, Esc restarts, a
  progress bar shows the step, and autoplay is optional.
- Export: a self-contained folder (index.html, demo.json, images) for any static host, a single HTML file
  with base64 images, and an <iframe> embed snippet. A Share button opens the folder with `npx serve`.
- Honest limits: Chrome allows about two captureVisibleTab calls per second, so merge clicks less than
  600 ms apart; screenshots only, so no text editing after capture; cross-origin iframes aren't captured.
- Single user. No accounts, no telemetry, local-first; the server listens on 127.0.0.1 only.
- Out of scope: video recording, AI voiceover and copy, view analytics, branching and forms, team
  workspaces, a desktop app.
- README: building and loading the unpacked extension, starting the server, where demos live, and a
  warning that screenshots keep whatever was on screen, so use blur boxes.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.