@umangtalpara/ui-agent
v1.0.8
Published
Autonomous multi-agent UI software development factory
Maintainers
Readme
@umangtalpara/ui-agent
Autonomous Multi-Agent Responsive UI Software Factory CLI
@umangtalpara/ui-agent is an NPM package and CLI tool designed to bootstrap any blank project directory into a collaborative AI-agentic workspace. It sets up files, rules, skills, and workflows that allow AI coding tools (Google Antigravity, Claude Code, Cursor, Codex, etc.) to act as a cooperative software development team to automatically build beautiful, fully responsive, and animated user interfaces from a PRD or Figma/design source.
🌟 Key Benefits
- 🤖 Fully Agentic & Automatic: Orchestrates 5 specialized agent personas (Orchestrator, Designer, Developer, QA, and Auditor) working in a cooperative loop.
- 💻 Universal IDE Support: Ready-to-use configurations for all major AI coding interfaces:
- Google Antigravity (
.agents/workspace rules, skills, and workflows) - Cursor (
.cursor/rules/MDC rules) - Claude Code (
.claude/skill stubs)
- Google Antigravity (
- 📐 Umang Talpara's Premium Design System: Tailors generated code to follow high-end enterprise layouts (HSL Navy/Cyan themes, glassmorphism card gutters, Outfit/Inter typography hierarchy, fluid responsive scaling, and smooth hover micro-animations).
- 📱 Strict Mobile-First & Responsive Focus: Enforces viewport meta settings, fluid grid grids, and mobile-friendly tap targets of at least
48px * 48px. - 🔍 Built-in Quality Auditing: Includes a validation suite to check for placeholders, responsive grid configurations, and accessibility violations before production.
🚀 Installation & Initialization
You can initialize a project directly using npx:
# 1. Create your new UI project directory
mkdir my-premium-ui
cd my-premium-ui
# 2. Initialize a blank package.json
npm init -y
# 3. Boot the Agentic Workspace
npx @umangtalpara/ui-agent init📖 Step-by-Step Usage Guide
Once you have initialized the workspace using the init command, follow these steps to configure your designs and launch the automated agent loop:
Step 1: Input Your UI Requirements / Design Reference
Fill in one or both of the configuration files under the doc/ directory to instruct the agent:
- If you have visual mockups, Figma links, or wireframe blueprints (
doc/design.md): Opendoc/design.mdand enter:- Links to your Figma mockups or visual boards.
- Markdown image references to layout wireframe screenshots (e.g.
). - Descriptions of the design layout (e.g., "Left-aligned brand logo, center navbar items, floating cards on hover").
- If you have structured text requirements or a feature list (
doc/prd.md): Opendoc/prd.mdand fill in:- Core Vision: The purpose of the website or application.
- Page Inventory: List every page/screen (e.g. Landing Homepage, User Dashboard, Contact Page).
- Layout Sections: What specific sections should go on each page? (e.g. Hero banner, 3-column benefit cards, email sign-up footer).
- Theme Options: Color tokens (navy base/cyan accent HSL colors) and animation specifications.
Step 2: Install UI Styling Dependencies
The CLI has already injected the necessary packages in your package.json. Install them using your preferred package manager:
npm install
# Or: pnpm install / yarn install / bun installStep 3: Trigger the AI Agent Build Loop
Open the project directory in your AI Coder of choice and run the starting prompt:
1. In Claude Code (Terminal Coder)
Start Claude Code in your workspace root:
claudeIn the Claude chat interface, send the start command:
"I have configured my design inputs in doc/design.md (or doc/prd.md). Ingest the requirements, run the design-to-prd workflow, and begin code generation."
2. In Cursor (Composer or Chat)
Open the workspace folder in Cursor. Open Cursor Composer (Ctrl+I / Cmd+I) or Chat (Ctrl+L / Cmd+L) and send the start command:
"Please read doc/design.md and doc/prd.md. Follow the rules in AGENTS.md, run the planning phase, and generate the React/Tailwind components."
3. In Google Antigravity (Gemini Extension)
Open the workspace folder in your IDE with the Google Antigravity extension enabled. In the agent chat box, send:
"Ingest the requirements in doc/prd.md and run the /prd-to-ui-plan workflow to generate the component inventory and begin coding."
Step 4: Monitor & Validate Progress
- Watch the Agent Progress: Open
.ai/project-management/progress.mdor run:
This outputs the currently active tasks, phase status, and logs.npx ui-agent status - Run Responsive Layout Verification: Once the agent writes components into the
codebase/directory, verify that spacing, touch bounds (>= 48px), and responsive viewport settings meet WCAG guidelines by running:npx ui-agent validate
🛠️ CLI Command Reference
1. ui-agent init
Copies all agent personas, standard skills, workflows, and rule files into your directory.
- Injects dependencies into
package.json(tailwindcss,postcss,autoprefixer,framer-motion,lucide-react,clsx,tailwind-merge). - Dynamically configures or appends standard exclusions to
.gitignore. - Detects lockfiles (
pnpm-lock.yaml,yarn.lock,bun.lockb) to output custom dependency installation commands.
2. ui-agent status
Outputs the current execution phase, active tasks, and status reports of the running agents:
npx ui-agent status3. ui-agent validate
Runs automated audit checks on the generated code, reporting errors or recommendations:
npx ui-agent validate- Viewport Tag Check: Ensures standard
<meta name="viewport" content="width=device-width...">tags exist. - Responsive Prefix Scan: Scans for Tailwind utility breakpoints (
sm:,md:,lg:). - Touch Target Audit: Flags interactive tags (
<button>,<a>, andonClickhooks) that lack spacing classes ensuring48pxboundaries. - Placeholder Detector: Alerts you if
TODO,lorem ipsum, or standard placeholders remain.
🤖 The Cooperative Agent Team
When initialized, the following personas are configured to cooperatively build your project:
┌──────────────────────────────┐
│ SUPER AGENT │
│ Orchestrator & Validator │
└──────────────┬───────────────┘
│
┌───────────────────────┼───────────────────────┐
▼ ▼ ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ DESIGN AGENT │ │ DEVELOPER AGENT│ │ QA AGENT │
│ Figma & Tokens │ │ React Coder │ │ Breakpoint & a11y│
└─────────────────┘ └─────────────────┘ └─────────────────┘
│
▼
┌─────────────────┐
│ REVIEW AGENT │
│ Aesthetic Audit │
└─────────────────┘- Super Agent (Orchestrator): Ingests PRDs and design specifications, constructs the roadmap, logs progress, delegates tasks, and validates each development phase.
- Design Agent (Figma Spec Parser): Translates design requirements, wireframes, or Figma links into structural tokens and components inventory.
- UI Developer Agent (Vite/React Coder): Writes components (
components/ui), pages, CSS frameworks, and coordinates motion wrappers (Framer Motion). - Responsive QA Agent (Breakpoint Tester): Validates layouts across device viewports (
320px,768px,1024px,1440px) and asserts WCAG compliance. - Code Review Agent (Aesthetic Auditor): Inspects naming rules, ensures clean HSL color systems, and checks for placeholder removal.
📂 Generated Workspace Layout
├── AGENTS.md ← Master workspace directive and layout pointers
├── CLAUDE.md ← Claude Code adapter
├── GEMINI.md ← Google Antigravity adapter
├── README.md ← Human introduction
│
├── .ai/ ← Base configurations, personas, and workflows
│ ├── settings.json ← UI development stack & directory map
│ ├── context/ ← Coding-rules, ui-guidelines, and naming conventions
│ ├── agents/ ← Base personas and directives for each agent
│ ├── workflows/ ← Phased checklists for design-to-prd, coding, and review
│ └── project-management/ ← Progress logs and phase status sheets
│
├── .agents/ ← Google Antigravity rule and workflow mappings
├── .cursor/rules/ ← Cursor MDC attach-rules (core, ui, coding)
├── .claude/skills/ ← Claude skill facades
│
├── doc/ ← Input specs (prd.md, design.md, how-to-use-ai.md)
└── codebase/ ← Output folder where the UI components and pages are generated📄 License
MIT © Umang Talpara
