Skip to content

How it works ​

A run, step by step ​

  1. Preflight. A script collects your OS, the git scope, the save folders, a running dev server and which tools are installed, as JSON.
  2. Questions. One dialog: what to make, where to save it, which style, and whether to screenshot your app.
  3. Storyboard. Claude reads the diff and the code around it (routes, components, copy, translation files) and writes a change brief, the scenes and the message.
  4. Approval. You see the storyboard and can render it, shorten it, add detail or change the style.
  5. Render. Screenshots, diagrams, slides and video, all on your machine.
  6. QA. Every output is checked, then Claude looks at each frame and fixes what is off, at most twice.
  7. Delivery. You get the folder, the message and a few follow-up ideas.

Rendering ​

  • Scenes are HTML templates driven only by CSS variables, so a style switch is a re-render, not a rewrite. Text that would overflow is shrunk to fit before anything is captured.
  • Video is a HyperFrames composition animated with GSAP and rendered deterministically in headless Chrome with FFmpeg. Without Node 22 or HyperFrames, a Playwright engine seeks the same timeline frame by frame instead.
  • Slides are screenshots of the same scenes at twice the resolution.
  • Diagrams go through mermaid-cli with the style's colours converted to hex and the fonts embedded with their licences.
  • Screenshots come from Playwright: numbered callouts, a neutral browser frame, masked personal data, and an optional "before" shot from a temporary git worktree.

Everything runs locally. Telemetry in the rendering tools is switched off and nothing is uploaded.

The tool folder ​

Rendering tools live in their own cache folder with pinned versions: HyperFrames, Playwright with one shared Chromium, mermaid-cli, GSAP, culori, rough.js and the open fonts each style needs. Your project's dependencies and lockfiles are never touched.