In the Skill's Own Terms
This skill drives Stitch MCP through four flows: generating a new screen from text, recreating an image or mockup, editing an existing screen, and generating variants. Before any generation or edit call it runs a prompt enhancement pipeline that resolves the project and design system, swaps vague wording for UI/UX terminology from its design mappings, and formats the prompt around layout, content and structure, using a template with PLATFORM and PAGE STRUCTURE sections.
Visual styling stays at project level, so generation prompts must carry no hex codes, fonts or roundness values. AI feedback is surfaced after every call and assets are downloaded to .stitch/designs.
What it produces
- The generated screen's HTML and screenshot, downloaded into the .stitch/designs directory.
- The AI text description and suggestions returned in outputComponents, shown to the user.
- Variant screens for the chosen direction, in batches of one to five.
How It Works
- 01Enhance the Prompt Before Anything Else
The pipeline is mandatory, not optional: the agent rewrites the user's request before it is allowed to call a Stitch generation or editing tool.
- 02Check What the Project Already Styles
If the project already has a design system, colors, fonts and roundness are applied at project level, so the enhanced prompt must leave visual styling out entirely.
- 03Format the Prompt as Platform Plus Page Sections
New screens follow a template that states overall purpose first, then PLATFORM, then a numbered PAGE STRUCTURE running from Header through Footer.
- 04Pick One of the Four Flows
The request is routed to Generate from Text, Generate from Image, Edit, or Generate Variants before any tool is called.