dmonteroh

google-stitch-ai

"Create DESIGN.md summaries from Google Stitch projects or offline assets for UI design workflows, and refine Stitch-ready UI prompts using extracted design tokens."

dmonteroh 1 Updated 3mo ago

Resources

1
GitHub

Install

npx skillscat add dmonteroh/curated-agent-skills/google-stitch-ai

Install via the SkillsCat registry.

SKILL.md

Google Stitch AI

Provides two workflows for Stitch tasks:

  • Design System Synthesis: Generate DESIGN.md from Stitch projects or provided assets.
  • Prompt Enhancement: Turn vague UI ideas into structured, Stitch-optimized prompts.

Use this skill when

  • A DESIGN.md is needed to capture a Stitch project’s design language.
  • A prompt needs enhancement for Stitch or similar UI generators.
  • Stitch-exported HTML/CSS and screenshots need a structured summary.

Do not use this skill when

  • The request needs user research, product strategy, or a full UX design brief.
  • The request requires inventing a brand identity without assets or inputs.
  • The request depends on cross-skill dependencies or multi-agent orchestration.

Activation cues (trigger phrases)

  • "Create a DESIGN.md from this Stitch project"
  • "Summarize the Stitch design system"
  • "Polish this Stitch prompt"
  • "Turn this UI idea into a Stitch-ready prompt"

Required inputs

  • Mode A (DESIGN.md): Stitch project data (exported files or offline assets), screenshots, and project title.
  • Mode B (Prompt Enhancement): Draft prompt, target platform (web/mobile), and optional DESIGN.md.

Decision points

  • If the request includes both a design summary and prompt refinement, complete Mode A first, then Mode B.
  • If Stitch project data is provided, use it for accurate theme data.
  • If only offline HTML/CSS and screenshots are provided, rely on those sources.
  • If key inputs are missing (palette, typography, layout), request them before final output.

Mode A: DESIGN.md Synthesis

Step-by-step

  1. Choose retrieval path.
    • If Stitch project data is provided, collect metadata + screen HTML/screenshot.
    • If offline, collect HTML/CSS export + screenshots + project title.
    • Output: asset bundle summary.
  2. Extract visual theme details.
    • Identify primary/secondary colors, background/surface colors, text colors.
    • Capture typography (font family, weights, sizes, line heights).
    • Output: theme notes list.
  3. Catalog components and states.
    • Buttons, inputs, cards, nav, modals, tables, etc.
    • Note states (hover, active, disabled) if visible.
    • Output: component inventory.
  4. Draft DESIGN.md.
    • Use the template below; fill with extracted tokens and observations.
    • Output: DESIGN.md file.
  5. Validate gaps.
    • If missing critical details, list follow-up questions.
    • Output: question list (if needed).

Output: DESIGN.md structure

# Design System: <Project Title>
**Project ID:** <ID>

## 1. Visual Theme & Atmosphere

## 2. Color Palette & Roles

## 3. Typography Rules

## 4. Component Stylings

Mode B: Prompt Enhancement

Step-by-step

  1. Clarify intent and platform.
    • Identify page type, user goal, and platform (web/mobile).
    • Output: clarified intent line.
  2. Pull design tokens (if available).
    • Use DESIGN.md to populate colors, typography, spacing, and components.
    • Output: token summary block.
  3. Structure the prompt.
    • Purpose + vibe (one line).
    • DESIGN SYSTEM block (if available).
    • Page structure (numbered layout sections).
    • Specific changes (only if editing an existing design).
    • Output: enhanced prompt in Markdown.
  4. Check for missing details.
    • Request missing platform or target audience info.
    • Output: question list (if needed).

Output contract

  • Mode A: Provide DESIGN.md, plus any open questions for missing tokens.
  • Mode B: Provide enhanced prompt in Markdown and state whether DESIGN.md was used.
  • If DESIGN.md is missing, recommend creating it via Mode A.
  • End with the reporting format described below.

Common pitfalls to avoid

  • Inventing brand colors or typography without evidence
  • Mixing web and mobile patterns in one prompt
  • Overloading the prompt with unrelated features
  • Ignoring accessibility contrast or text hierarchy

Examples

Example input → output (prompt enhancement)

Input: "Make a clean dashboard for sales KPIs."

Output (excerpt):

Purpose/Vibe: Clean, professional sales KPI dashboard for executives.

DESIGN SYSTEM
- Primary: #1F4ED8, Secondary: #93B5FF
- Typography: Inter (H1 32/40, H2 24/32, Body 16/24)

Page Structure
1. Top nav with logo, account, date range filter
2. KPI summary cards (Revenue, ARR, Churn)
3. Line chart for revenue trend
4. Table of top accounts

Reporting format

Provide results in this structure:

## Output
- Mode: DESIGN.md | Prompt
- Inputs used: Project data | Offline assets (list assets)
- Files created/updated: <paths or none>
- Open questions: <list or none>

Resources (Optional)

  • UI/UX keyword palette: references/KEYWORDS.md
  • Reference index: references/README.md