Browser-agent workflow

Some capabilities are intentionally browser-only because their output depends on browser graphics APIs or interactive composition.

Use the browser for

• Moodboard PNG up to 4800 × 6000
• Animation GIF packages
• Live logo shaders and shader GIFs
• PNG output from OpenGraph, terminal, logos, and templates
• Local font, image, logo, and shader inputs
• Visual placement, opacity, scale, and zoom inspection
• Shared Brand settings that update all browser-local designs

Discovery before operation

1. Fetch /api/catalog.
2. Select the exact tool id or visible name.
3. Open /studio.
4. Choose Starter, GT, or create a local project.
5. Open the matching tool with navigation or Studio search.
6. Configure fields in the inspector.
7. Verify the canvas.
8. Activate the visible download action.

Reliable interaction

• Prefer visible labels and accessible names over coordinates.
• Wait for imported images, fonts, and shader previews to finish rendering.
• Use the canvas at 100% for final inspection.
• Confirm surface and logo reversal in both light and dark compositions.
• Treat browser download completion as the terminal success condition.
• Do not delete projects unless the user explicitly requests deletion.

Persistence boundary

Browser projects persist in localStorage. HTTP requests do not read or mutate those projects. If an agent needs an API artifact from a browser identity, it must serialize the required foundation fields into the request.

Choosing API versus browser

If /api/generate can represent the artifact, prefer it. Browser automation is appropriate when visual judgment or browser-only rendering is part of the requested result.