VAEA UI — Design System Enforcer

Single entry point for all design quality on ACM-AI. Orchestrates existing design skills through VAEA brand constraints.

Activation Modes

Mode 1: Auto Context Injection (on file touch)

When ANY agent modifies frontend/**/*.tsx, frontend/**/*.ts, or frontend/**/*.css:

1. Load .impeccable.md from project root — this is the design authority
2. Load hard rules from references/acm-patterns.md (read that file now)
3. Apply VAEA constraints to all design decisions
4. No review, no checklist — just ensure the agent HAS the knowledge

Mode 2: Explicit Sub-Command (on invocation)

Invoke with /vaea-ui <command> or /vaea-ui <command> <file-or-scope>.

Mode 3: Story Completion Gate (on story-complete)

When a UI story is marked complete, run the mandatory checklist from references/design-checklist.md (read that file now). BLOCK completion if any critical item fails.

Context Gathering Protocol

Before ANY design work, execute these steps in order:

Step 1: Load Design Context

Read .impeccable.md at the project root. This file contains:

• Users and personas (compliance officers, facility managers, assessors)
• Brand personality: Professional, Transparent, Calm
• Emotional tone: Confidence & Trust
• References: Salesforce + Xero
• Anti-references: generic SaaS, legacy .gov.au, over-designed, dense ERP
• 5 design principles
• Full color system with coral expansion rules
• Typography and font loading requirements
• Responsive strategy (true responsive, card view on mobile for grids)
• Motion direction (invest in framer-motion)
• Accessibility target (WCAG 2.2 AAA aspirational)

If .impeccable.md does not exist, run /teach-impeccable first. STOP and do not proceed without design context.

Step 2: Load ACM Component Patterns

Read references/acm-patterns.md in THIS skill directory. Contains hard rules with code examples for:

• AG Grid theming (must use .ag-theme-custom from globals.css)
• Risk badge dual encoding (icon + color, never color alone)
• Mobile card fallback (card view below md breakpoint)
• Font loading (next/font, not Google Fonts link tags)
• ARIA patterns (specific to ACM components)
• Coral usage (CTAs, badges, notifications — expanded role)
• Government design patterns (12px radius, teal shadows, acknowledgment footer)

Step 3: Load Dependency Skills

Load these skills for their specialized knowledge, filtered through VAEA constraints:

Curated Command Sheet

10 commands tuned for ACM-AI's "Professional, Transparent, Calm" personality.

Evaluation

/vaea-ui critique [file|scope]

Evaluate design effectiveness. Loads critique skill internally.

• Assesses: visual hierarchy, information architecture, emotional resonance, design quality
• ACM lens: Does it inspire confidence? Is data scannable? Does it feel like Salesforce/Xero?
• Leads to: /vaea-ui polish, /vaea-ui arrange, /vaea-ui typeset

/vaea-ui audit [file|scope]

Comprehensive quality audit. Loads audit skill internally.

• Checks: accessibility (WCAG 2.2 AAA), performance, theming, responsive, anti-patterns
• ACM lens: Dual-encoded risk indicators? AG Grid theme from globals.css? Mobile card fallback?
• Leads to: /vaea-ui normalize, /vaea-ui harden, /vaea-ui optimize, /vaea-ui adapt

Quality

/vaea-ui polish [file|scope]

Final quality pass before shipping.

• Fixes: alignment, spacing, consistency, detail issues
• ACM lens: Design token usage (no raw hex/px), VAEA shadow system, consistent border-radius

/vaea-ui harden [file|scope]

Improve interface resilience.

• Fixes: error handling, text overflow, edge cases, loading/empty/error states
• ACM lens: ConnectionGuard patterns, ErrorBoundary on every page, Skeleton matching final layout

/vaea-ui normalize [file|scope]

Normalize to match VAEA design system.

• Fixes: inconsistent token usage, off-brand colors, wrong component variants
• ACM lens: oklch tokens only, VAEA teal palette, branded teal-tinted dark mode

/vaea-ui optimize [file|scope]

Improve interface performance.

• Fixes: loading speed, rendering, animations, images, bundle size
• ACM lens: AG Grid lazy loading, React Query staleTime tuning, next/font migration, route prefetching

Adaptation

/vaea-ui adapt [file|scope]

Adapt for different screen sizes and devices.

• Fixes: responsive behavior, touch targets, grid-to-card transitions
• ACM lens: AG Grid to card view below md, Sheet for mobile chat, collapsible sidebar, pinned columns

/vaea-ui clarify [file|scope]

Improve UX copy and microcopy.

• Fixes: labels, error messages, instructions, tooltips, empty states
• ACM lens: Government tone (direct, factual, no jargon), status messages state facts, errors explain next step

Layout & Typography

/vaea-ui arrange [file|scope]

Improve layout, spacing, and visual rhythm.

• Fixes: monotonous grids, inconsistent spacing, weak visual hierarchy
• ACM lens: 4px spacing grid, BentoGrid for dashboards, data density over whitespace in grids

/vaea-ui typeset [file|scope]

Improve typography.

• Fixes: font choices, hierarchy, sizing, weight consistency, readability
• ACM lens: Inter for body, JetBrains Mono for data, Text component variants, tabular-nums for numbers

Hard Rules (Non-Negotiable)

These rules MUST be followed. Violations block story completion.

1. Design Tokens Only

NEVER: className="text-[#53A69D]" or style={{ color: '#53A69D' }}
ALWAYS: className="text-primary" or className="text-vaea-teal-300"

2. AG Grid Theme from globals.css

NEVER: <style jsx global>{`.ag-theme-alpine { ... }`}</style>
ALWAYS: className="ag-theme-custom" (defined once in globals.css)

3. Risk Indicators Dual Encoded

NEVER: <Badge className="bg-risk-high">High</Badge>
ALWAYS: <Badge className="bg-risk-high"><XCircle className="h-3 w-3 mr-1" />High</Badge>

4. Mobile Card Fallback for Grids

// Below md: card stack. Above md: AG Grid.
{isMobile ? <RecordCardList records={records} /> : <DataGrid ... />}

5. Font Loading via next/font

NEVER: <link href="https://fonts.googleapis.com/css2?family=Inter..." />
ALWAYS: import { Inter } from 'next/font/google'

6. Coral for CTAs and Important Actions

NEVER: Primary CTA using only teal
ALWAYS: Important actions, notification dots, critical badges use vaea-coral

7. ARIA on All Interactive Elements

NEVER: <button onClick={...}><Search /></button>
ALWAYS: <button onClick={...} aria-label="Search records"><Search /></button>

8. Text Component for Typography

NEVER: <h2 className="text-2xl font-bold">Title</h2>
ALWAYS: <Text variant="h2">Title</Text>

9. Teal-Tinted Dark Mode

NEVER: Dark mode backgrounds using neutral grey (oklch(0.15 0 0))
ALWAYS: Dark mode backgrounds with teal tint (oklch(0.175 0.025 170))

10. Aboriginal Acknowledgment in Footer

Victorian government requirement. Always present via branding.ts config.

Story Completion Gate

When a UI story is marked complete, the agent MUST run the mandatory design checklist. Read references/design-checklist.md for the full checklist.

Quick summary — ALL must pass:

• Design tokens used (no raw hex/px values)
• Responsive: tested at mobile (< 640px), tablet (768px), desktop (1280px+)
• Dark mode: verified in both light and dark themes
• ARIA: all interactive elements have labels, roles, states
• Risk indicators: dual encoded (icon + color) if applicable
• AG Grid: uses .ag-theme-custom from globals.css, no inline styles
• Typography: uses <Text> component variants, not raw heading tags
• Loading states: Skeleton components matching final layout structure
• Error states: ErrorBoundary + meaningful fallback on every page/section
• Mobile grids: card view fallback below md breakpoint if AG Grid is used

If ANY critical item fails, the story is INCOMPLETE. Fix before marking done.

AI Slop Test

Before submitting any UI work, ask yourself:

"If someone saw this and said 'AI made this,' would they believe it immediately?"

If yes, that's the problem. Check for:

• Generic card-in-card nesting
• Uniform card grids with no visual hierarchy
• Purple/blue gradient text
• Glassmorphism everywhere
• Centered hero metric layout
• Sparklines as decoration
• Bounce/elastic animations
• Dark mode with glowing neon accents

ACM-AI should feel like it was designed by a government UX team that cares deeply about data quality — not generated by an AI coding assistant.

File Reference