Skip to content

Repository files navigation

Frontend Design Intelligence

Design intelligence for coding agents.
A reusable Pi Agent skill system for translating vague user intent, auditing, redesigning, animating, and portfolio-polishing modern web applications using curated design knowledge, motion patterns, framework recipes, Playwright browser inspection, and visual evaluation rubrics.

Pi Agent Agent Skills TypeScript React Next.js Tailwind CSS License: MIT


Supported Technology Ecosystem

Platform / Framework Motion & Animation Styling & Tooling Agent Runtime
TypeScript React Framer Motion GSAP Tailwind CSS CSS3 Pi
Next.js Astro Canvas 2D WebGL PostCSS Playwright Agent Skills

Overview

AI coding agents excel at generating functional code, but frequently produce visually mediocre, unrefined, or unmistakably AI-generated web interfaces. They default to purple gradients, glowing borders, floating ambient blur orbs, and generic SaaS marketing copy ("Revolutionize your workflow").

frontend-design-intelligence equips Pi coding agents with a master Creative Director and a structured design/motion intelligence layer. You communicate in normal, informal language ("Make this look legit", "Parang mas premium sana", "This looks vibe-coded", "Make it look like something I can show a mayor"). The system translates your intent into a structured design strategy, inspects code and browser viewports, and orchestrates specialist skills to execute production-grade polish.

Orchestration Architecture Flow

graph TD
    A[Informal / Layman Prompt] --> B[Creative Director Skill]
    
    B --> C[Intent Translation & Design Brief]
    C --> D[Product Context Classification]
    D --> E[Playwright Browser Inspection]
    
    E --> F{Orchestrated Specialist Skills}
    F --> G[Frontend Audit]
    F --> H[Hero Redesign]
    F --> I[UI Polish]
    F --> J[Motion Pass]
    F --> K[Portfolio Polish]

    G & H & I & J & K --> L[Domain Context Guides & Anti-Slop Rules]
    G & H & I & J & K --> M[Knowledge Base & Recipes]

    M --> N[Playwright Visual QA & Recapture]
    N --> O[Validated Code Changes & Creative Director Report]
Loading

Core Pi Agent Skills

This repository provides seven portable Pi skills in skills/:

Skill Command Role Description Modifies Code? Best Used For
/skill:creative-director Orchestrator Master intent-translation and design director skill. Translates vague/layman prompts into structured design briefs, classifies product context, and orchestrates specialist skills. Yes Vague or informal design requests ("Make it look legit", "Vibe-coded", "Pitch to a mayor").
/skill:portfolio-polish Specialist Full frontend improvement workflow. Audits stack, typography, layout, motion, and storytelling before executing targeted code improvements. Yes Preparing projects for portfolio showcases, demos, client pitches, or senior hiring review.
/skill:frontend-audit Specialist Non-destructive design analysis. Scores project against 14 rubric dimensions, captures Playwright screenshots, and flags anti-slop violations. No Initial code reviews and visual health checks without modifying files.
/skill:motion-pass Specialist Targeted animation refinement. Adds spring physics, staggered entrances, microinteractions, and reduced-motion support. Yes Smoothing out janky transitions, card hover effects, and scroll interactions.
/skill:hero-redesign Specialist Specialized hero section art direction. Replaces generic slogans with technical/civic copy, stack badges, and live mockups. Yes Redesigning landing page headers, developer tool heroes, and LGU portal intros.
/skill:ui-polish Specialist Micro-refinement for buttons, cards, inputs, tables, forms, modals, and empty/loading states. Yes Tightening 4px/8px grid spacing, border definitions, and focus rings.
/skill:study-reference Specialist Deconstructs external URLs, design references, and motion benchmarks into normalized JSON pattern entries. No Extracting reusable principles from external design systems or web showcases.

Natural-Language Intent Translation

The creative-director skill translates subjective human language into precise technical strategies:

  • "Make this look more legit." → Increase perceived craft: replace placeholder fonts with Geist/Inter, enforce 1px neutral borders, fix 4px/8px grid alignment, add stack badges and real technical metrics.
  • "Parang mas premium sana." / "Make it look expensive." → Establish visual restraint: reduce color count to 1 primary neutral + 1 accent hue, increase vertical whitespace, apply high-contrast typography (tracking-tight), add spring hover feedback.
  • "This still looks vibe-coded." → Purge AI-slop visual clichés: remove purple/pink radial background orbs, glowing rainbow borders, uniform 24px rounded bento grids, and low-contrast backdrop-blur.
  • "Make it look like something I can show a mayor." → Establish civic trust & LGU authority: apply municipal service discovery layout, emergency advisory banner, clear contact directory, mobile-first thumb navigation, and WCAG AA contrast.
  • "Make it more alive but don't overdo it." → Add functional, restrained motion: spring hover feedback on buttons (active:scale-[0.98]), staggered headline word reveals, smooth metric counter rolls, and prefers-reduced-motion fallbacks.

Context-Specific Design Intelligence

Located in knowledge/contexts/, the system applies domain-specific principles rather than forcing a generic SaaS style onto every project:

  • LGU & Government Portals (lgu-government.md): TRUST > NOVELTY, SERVICE DISCOVERY > MARKETING, ACCESSIBILITY > SPECTACLE, LOCAL IDENTITY > GENERIC STARTUP BRANDING, MOBILE USABILITY > FANCY DESKTOP EFFECTS.
  • Analytics & Data Dashboards (dashboard.md): High density, font-mono tabular-nums numbers, chart entrances, slide-over detail drawers.
  • Developer Tools & CLI Web Apps (developer-tool.md): Terminal code blocks, copy micro-interactions, dark mode #09090B, monospace typography.
  • AI / ML Products (ai-ml-product.md): Telemetry over marketing hype, P99 latency ms, confidence score %, transparent prompt streams.
  • Developer Portfolios (portfolio.md): Tech stack badges, architecture diagrams, measured performance metrics, live interactive demo frames.

The Anti-AI-Slop Philosophy

This system strictly enforces design restraint. It flags and eliminates common AI visual clichés:

  • Banned: Glowing purple/pink background radial orbs, glowing rainbow border gradients, uniform 24px+ rounded bento grids, low-contrast glassmorphism (backdrop-blur bg-white/10), generic slogans ("Powered by AI").
  • Required: High-contrast neutral bases (bg-zinc-950 / bg-white), crisp 1px borders (border-zinc-800), modular type scales (Geist, Inter Display), single restrained accent hue, explicit 5-state interaction controls, and 100% prefers-reduced-motion compliance.

Jitter Motion Research Corpus

Jitter (https://jitter.video/) serves as the initial motion research corpus for this repository. We reverse-engineered public Jitter UI templates across buttons, toggles, navigation rails, device showcases, and kinetic text into framework-agnostic web primitive recipes.

All extracted patterns are stored as structured JSON records in knowledge/references/jitter/ detailing motion primitives, choreography, spring parameters, and framework feasibility.


14-Dimension Visual Quality Rubric

Projects are evaluated out of 100 points across 14 dimensions defined in evals/rubric.md:

  1. Visual Hierarchy (/10)
  2. Typography (/10)
  3. Layout & Composition (/10)
  4. Spacing & Rhythm (/10)
  5. Color Discipline (/10)
  6. Interaction Quality (/10)
  7. Motion Quality (/10)
  8. Responsiveness (/10)
  9. Accessibility (/10)
  10. Content Clarity (/10)
  11. Technical Storytelling (/10)
  12. Originality (/10)
  13. Perceived Craft (/10)
  14. Portfolio Readiness (/10)

Installation & Usage

1. Global Installation (Recommended)

Makes the skill package available across all Pi projects on your computer:

pi install git:github.com/kerwinarlan/frontend-design-intelligence

2. Project-Local Installation

Adds the skills to the current project's .pi/settings.json:

pi install git:github.com/kerwinarlan/frontend-design-intelligence -l

3. Quick Start Examples

Natural Language Request (Creative Director)

/skill:creative-director

Bro this still looks super vibe-coded. Make it feel like a real LGU portal I can pitch to a mayor, but keep it modern and portfolio-worthy. Don't overdo animations.

Non-Destructive Design Audit

/skill:frontend-audit

Targeted Hero Redesign

/skill:hero-redesign

4. CLI Audit & Browser Inspection

# Static codebase & anti-slop audit CLI:
npm run audit -- /path/to/target-project

# Playwright browser inspection & screenshot capture:
npx tsx scripts/browser-inspector.ts http://localhost:3000

Repository Structure

frontend-design-intelligence/
├── README.md
├── AGENTS.md
├── LICENSE
├── package.json
├── package-lock.json
├── tsconfig.json
│
├── skills/                     # Canonical Pi Agent Skills (7 Skills)
│   ├── creative-director/      # Master Intent-Translation & Orchestration Skill
│   │   ├── SKILL.md
│   │   └── references/         # Brief Schema, Intent Guide, Intervention Levels
│   ├── portfolio-polish/
│   ├── frontend-audit/
│   ├── motion-pass/
│   ├── hero-redesign/
│   ├── ui-polish/
│   └── study-reference/
│
├── .pi/                        # Local Project Config
│   ├── settings.json
│   └── skills -> ../skills     # Symlink to root skills/
│
├── docs/
│   ├── ARCHITECTURE.md
│   ├── TAXONOMY.md
│   ├── ROADMAP.md
│   ├── DESIGN_PRINCIPLES.md
│   ├── ANTI_SLOP.md
│   ├── PORTFOLIO_DESIGN.md
│   └── CONTRIBUTING_KNOWLEDGE.md
│
├── knowledge/
│   ├── contexts/               # LGU Portal, Dashboard, DevTool, AI/ML, Portfolio Context Guides
│   ├── fundamentals/           # Typography, Layout, Responsive, Color, Motion, etc.
│   ├── patterns/               # Heroes, Cards, Navigation, Dashboards, Microinteractions
│   ├── references/jitter/      # Structured JSON reference databases & STATUS.md
│   └── recipes/                # Framework-aware code recipes (React, Tailwind, Motion, GSAP)
│
├── evals/
│   ├── rubric.md               # 14-Dimension Visual Quality Rubric
│   ├── eval-fixture.ts         # Code Evaluation Logic
│   ├── fixtures/               # Runnable Before vs After Web Application Fixture & Intent Tests
│   └── real-projects/          # Non-destructive audits of real repositories
│
├── scripts/
│   ├── audit-project.ts        # CLI Audit Tool
│   ├── browser-inspector.ts    # Playwright Screenshot & Overflow Inspector
│   ├── validate-skills.ts      # Pi Skill Specification Validator
│   └── validate-knowledge.ts   # JSON Reference Schema Validator
│
└── .github/
    ├── REPOSITORY_METADATA.md
    └── workflows/ci.yml

Development & Verification

Run the validation suite to verify all 7 Pi skills, YAML frontmatter, JSON schemas, and browser fixtures:

npm run validate
npx tsx evals/fixtures/run-fixture-eval.ts

Research & Reference Attribution

  1. Third-Party Trademarks: Jitter, Framer, GSAP, Lottie, Apple, Vercel, and related product names referenced herein belong to their respective copyright holders.
  2. No Proprietary Asset Redistribution: This repository contains no proprietary template files, assets, source code, or media from Jitter or third parties.
  3. Original Analytical Work: All patterns, taxonomies, and code recipes represent original analytical abstractions created for educational and design-intelligence purposes under the MIT License.

License

MIT License © 2025 Kerwin Arlan

About

Design, motion, Canvas 2D, and WebGL intelligence for coding agents. Audit, redesign, animate, and portfolio-polish frontend projects with reusable Pi skills.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages