oimiragieo/enhance-prompt

Enhance Prompt

Transform vague UI and feature requests into structured, optimized prompts with design system awareness. Based on the Google Labs Stitch Skills prompt enhancement pipeline.

When to Apply

Use this skill when:

• A user request describes a UI feature vaguely ("make a nice login page")
• You need to generate implementation prompts from loose requirements
• Translating design mockups or wireframes into actionable specs
• Integrating design system tokens into component specifications
• Scaffolding page layouts from high-level descriptions

4-Step Enhancement Pipeline

Step 1: Assess Intent

Analyze the raw request to identify:

• Core objective: What is the user trying to build?
• Implicit requirements: What is assumed but not stated?
• Target audience: Who will use this UI?
• Interaction patterns: What user flows are implied?

Example:

Raw: "Add a settings page"
Assessment:
  - Core: User preferences management interface
  - Implicit: Navigation integration, form validation, save/cancel actions
  - Audience: Authenticated users
  - Interactions: View current settings, modify, save, receive confirmation

Step 2: Check DESIGN.md

Look for a DESIGN.md or equivalent design system file in the project root or docs directory. Extract:

• Color tokens: Primary, secondary, accent, semantic colors
• Typography scale: Font families, sizes, weights, line heights
• Spacing system: Base unit, scale (4px, 8px, 12px, 16px, etc.)
• Component inventory: Available pre-built components
• Layout patterns: Grid system, breakpoints, container widths

If no design system file exists, generate reasonable defaults based on the project's existing patterns (check tailwind.config.*, theme.*, or CSS custom properties).

Lookup order:

1. DESIGN.md in project root
2. docs/design-system.md
3. tailwind.config.ts / tailwind.config.js theme section
4. CSS custom properties in global styles
5. shadcn/ui theme configuration (if present)

Step 3: Apply Enhancements

Transform the raw request by applying these enhancement layers:

3a. Add Specificity

Replace vague terms with specific component names and patterns:

| Vague Term | Specific Replacement | | --------------- | ------------------------------------------------------------------------- | | "nice" | Clean layout with consistent spacing, proper visual hierarchy | | "modern" | Card-based layout, rounded corners, subtle shadows, micro-interactions | | "responsive" | Mobile-first grid (1-col mobile, 2-col tablet, 3-col desktop) | | "fast" | Skeleton loading states, optimistic updates, prefetched data | | "user-friendly" | Clear labels, inline validation, helpful error messages, focus management | | "beautiful" | Design-system-compliant colors, balanced whitespace, typography scale | | "simple" | Single-column layout, minimal navigation, progressive disclosure | | "interactive" | Hover states, transitions, keyboard navigation, ARIA attributes |

3b. Inject Design Tokens

Insert design system values into the specification:

Enhanced: "Use the primary color (--color-primary / colors.primary.500)
for the submit button. Apply spacing-4 (16px) between form fields.
Use the heading-2 typography preset for section titles."

3c. Scaffold Layout Structure

Generate a structural layout for the page/component:

Page Layout:
  Header: Breadcrumb navigation + page title
  Body:
    Sidebar (desktop only): Section navigation
    Main Content:
      Section 1: [Primary feature area]
      Section 2: [Secondary feature area]
      Section 3: [Related actions]
  Footer: Save/Cancel actions (sticky on mobile)

3d. Add Interaction Specifications

Specify loading states, error states, empty states, and transitions:

• Loading: Skeleton placeholder matching final layout
• Error: Inline error messages with retry action
• Empty: Helpful illustration + primary action CTA
• Success: Toast notification + state update
• Transitions: 200ms ease-in-out for element appearance

Step 4: Format Output

Produce the enhanced prompt in this structured format:

## Enhanced Specification: [Feature Name]

### Objective

[1-2 sentences describing the precise goal]

### Design System Integration

- Colors: [specific tokens]
- Typography: [specific presets]
- Spacing: [specific scale values]
- Components: [reusable components to use]

### Layout Structure

[ASCII or markdown layout diagram]

### Component Breakdown

1. [Component name] - [purpose] - [key props/state]
2. [Component name] - [purpose] - [key props/state]

### Interaction States

- Loading: [description]
- Error: [description]
- Empty: [description]
- Success: [description]

### Accessibility Requirements

- Keyboard navigation: [specifics]
- Screen reader: [ARIA labels/roles]
- Focus management: [focus trap/restore behavior]

### Responsive Behavior

- Mobile (< 640px): [layout]
- Tablet (640-1024px): [layout]
- Desktop (> 1024px): [layout]

Complete Example

Raw Request: "Build a dashboard for our analytics"

Enhanced Output:

## Enhanced Specification: Analytics Dashboard

### Objective

Build a data visualization dashboard displaying key business metrics
with real-time updates, date range filtering, and exportable reports.

### Design System Integration

- Colors: --color-primary for active states, --color-success/warning/danger
  for metric status indicators
- Typography: heading-1 for page title, heading-3 for card titles,
  body-sm for metric labels
- Spacing: gap-6 between metric cards, padding-4 inside cards
- Components: Card, Select (date picker), Button, Badge, Skeleton

### Layout Structure

[Nav]
[Page Title + Date Range Selector + Export Button]
[KPI Card] [KPI Card] [KPI Card] [KPI Card] <- 4-col grid
[Line Chart (2/3 width)] [Pie Chart (1/3 width)] <- 2-col split
[Data Table (full width)] <- sortable, paginated

### Component Breakdown

1. KPICard - Displays single metric with trend indicator - { label, value, change, status }
2. DateRangeSelector - Preset ranges + custom date picker - { range, onRangeChange }
3. TrendChart - Line/area chart for time series data - { data, dateRange, metric }
4. DistributionChart - Pie/donut for categorical data - { data, metric }
5. MetricsTable - Sortable, paginated data table - { columns, data, sortBy, page }

### Interaction States

- Loading: Skeleton cards (pulse animation) matching final dimensions
- Error: Error banner with retry button, individual chart error boundaries
- Empty: "No data for selected range" with suggestion to expand range
- Success: Smooth number transitions (count-up animation) on data refresh

### Accessibility Requirements

- Keyboard navigation: Tab through cards, charts have data table alternatives
- Screen reader: Chart descriptions via aria-label, live region for metric updates
- Focus management: Date picker focus trap, return focus on close

### Responsive Behavior

- Mobile (< 640px): Single column, KPI cards stack vertically, charts full width
- Tablet (640-1024px): 2-col KPI grid, charts stack vertically
- Desktop (> 1024px): 4-col KPI grid, side-by-side charts

Anti-Patterns

• Do NOT skip Step 2 (design system check) -- prompts without design tokens produce inconsistent UI
• Do NOT leave vague terms unresolved -- "nice" is not a specification
• Do NOT generate prompts without interaction states -- every view has loading, error, and empty states
• Do NOT ignore accessibility -- every enhanced prompt must include keyboard and screen reader specs
• Do NOT hardcode pixel values -- always reference design system tokens or relative units

Iron Laws

1. ALWAYS analyze the original prompt for ambiguities and implicit assumptions before enhancing — never enhance a prompt whose scope you haven't verified with the caller.
2. NEVER add requirements that weren't implicit or explicit in the original — enhancement clarifies and structures; it does not invent scope.
3. ALWAYS preserve the original intent — an enhanced prompt that redirects to a different goal is a rewrite, not an enhancement.
4. NEVER produce an enhanced prompt longer than necessary to resolve ambiguities — verbosity in prompts reduces AI response quality; keep it focused.
5. ALWAYS include success criteria in the enhanced prompt — a prompt without acceptance criteria cannot be evaluated as done or not done.

Memory Protocol (MANDATORY)

After using this skill, record learnings using MemoryRecord:

MemoryRecord({
  type: 'pattern',
  text: 'enhance-prompt: [describe what design system pattern or prompt structure worked well]',
  area: 'frontend',
});

Write decisions to .claude/context/memory/decisions.md when choosing between design system approaches. Write issues to .claude/context/memory/issues.md when DESIGN.md is absent and fallback patterns were used.