promptplay

Transform natural language prompts into playable games using Claude

What is promptplay?

promptplay is a creative playground that leverages the Claude Agent SDK to turn natural language descriptions into fully functional games. Describe the game you want to build—a Snake game, Flappy Bird, a puzzle game—and Claude will write the complete implementation for you.

How It Works

Web Interface (Recommended)

Your Game Idea
    ↓
Web Form (index.html)
    ↓
/generate endpoint (server.py)
    ↓
Claude Agent SDK generates HTML/CSS/JS
    ↓
Saved to output.html
    ↓
Rendered in iframe on page 🎮

CLI Mode

Edit agent.py with your game idea
    ↓
python agent.py
    ↓
Claude generates complete game code
    ↓
Saved to output.html
    ↓
Open in browser 🎮

The agent automatically wraps your game idea with strict instructions to ensure Claude returns only clean HTML/CSS/JavaScript code without explanations or tool usage.

Quick Start

Prerequisites

• Python 3.10+
• An Anthropic API key (get one here)

Installation

# Clone the repository
git clone https://github.com/sausi-7/promptplay.git
cd promptplay

# Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Set up your API key
cp .env.example .env
# Edit .env and paste your ANTHROPIC_API_KEY

Running

Option 1: Web Server (Recommended)

Start the FastAPI server:

python server.py
# Server runs on http://localhost:8000

Open your browser and describe your game idea. The generated game will render instantly in the page.

Option 2a: Command Line - Quick Mode

Pass a game idea directly or use the default tic-tac-toe prompt:

python agent.py "A cozy memory card game with three difficulty levels" --output output.html
# Streams progress and generated code, then saves output.html
open output.html

Use --verbose to include detailed stream logs while Claude is generating.

Option 2b: Command Line - Conversational Mode (Recommended)

For better games with clarifications:

python conversation_agent.py
# Interactive mode:
# 1. Describe your game idea
# 2. Answer Claude's clarifying questions  
# 3. Get a tailored game
# 4. Request refinements if needed
# 5. Save with full conversation memory

Game Generation Modes

One-Shot Mode (agent.py)

• ⚡ Fast: 5-10 seconds
• 🎯 Simple: Just edit and run
• 📡 Streaming: Shows progress and generated code as it arrives
• 📝 Use when: You have a clear idea and want it quickly
• 🔄 Can't iterate: One generation per run

Conversational Mode (conversation_agent.py) ⭐ Recommended

• 💬 Interactive: Claude asks clarifying questions
• 🎮 Better quality: Games tailored to your specs
• 🔧 Refinable: Ask for changes, get improved versions
• 📚 Memory: Full conversation saved as JSON
• ⏱️ Takes: 30-60 seconds + optional refinements

Example Conversational Flow:

You: "Make a puzzle game"
Claude: "What type of puzzle? (sliding, matching, word, logic?)"
You: "Matching cards with difficulty levels"
Claude: "Generated! Try it."
You: "Make it harder - fewer cards to match"
Claude: "Refined! Check it out."

Examples

Try these prompts by editing the game_idea string in agent.py (one-shot) or describing them to conversation_agent.py (conversational):

Prompt	Output
"Build me a snake game in single index.html"	Classic Snake with keyboard controls
"Create a Flappy Bird-style game"	Bird dodging obstacles
"Build a memory matching card game"	Click-to-flip matching game
"Make a simple Pong game"	Two-player paddle game

Architecture

Key Components

• agent.py - Core game generation using Claude Agent SDK

  • build_prompt() - Wraps user ideas with strict instructions to ensure HTML-only output
  • generate_game_html() - Calls Claude to generate complete game code
  • htmlParser() - Extracts clean HTML from Claude's response

• server.py - FastAPI web server with two endpoints

  • POST /generate - Accepts a game idea, runs the agent, returns success/failure
  • GET /game - Serves the generated output.html file
  • GET / - Serves the web UI

• index.html - Web interface for submitting game ideas

  • Form for entering game descriptions
  • Iframe rendering of generated games
  • Error handling and loading states

Customization

Change the Game Idea (CLI)

Edit the game_idea in agent.py:

game_idea = "Your detailed game description here"
html = await generate_game_html(game_idea)

Adjust the Model

Change the model parameter in agent.py:

model="claude-opus-4-8"      # For complex/3D games
model="claude-sonnet-4-6"    # Balanced performance
model="claude-haiku-4-5-20251001"  # Fast, lightweight

Modify Prompt Instructions

Edit build_prompt() in agent.py to customize how game ideas are converted to instructions.

Roadmap

• Web UI to input prompts and play games directly
• FastAPI server with endpoints for game generation
• Iframe-based game rendering (no popup blockers)
• Multi-file game support (separate HTML, CSS, JS)
• Prompt gallery: community-submitted game ideas
• CLI flags to customize model, effort, and output
• Game examples directory with pre-generated games
• Streaming output for real-time generation feedback
• Game save/download functionality
• Share generated games via URL

Contributing

We love contributions! Whether you're fixing bugs, adding features, or sharing cool game prompts—all are welcome.

Getting Started

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-game-idea)
3. Make your changes
4. Commit with a clear message (git commit -m "Add support for X")
5. Push to your fork and open a PR

See CONTRIBUTING.md for detailed guidelines.

Good First Issues

• Add a new example prompt to the README
• Create a prompt template for different game genres
• Add error handling for API failures
• Write tests for the agent module
• Improve documentation

License

This project is licensed under the MIT License—see LICENSE for details.

Questions?

Open an issue on GitHub or reach out to @sausi-7.

---

Built with ❤️ and the Claude Agent SDK