🎮 GameDev

AI-powered game development workbench. Natural language -> automated code review, config modification, test generation, and more.

Features

• Agent Loop + Skill: One agent loop handles all tasks, guided by Skill .md files
• Deterministic Modification: Config/code changes via LLM intent parsing + Pydantic validation + code execution
• MCP Multi-Server: Dynamic tool discovery across FileSystem, Git, and custom game dev tools
• Progressive Disclosure UI: Chat-first interface, smart recommendations after scan
• Engine Agnostic: Abstract scanner + tool mapping (Unity first)

Quick Start

git clone https://github.com/Firefly0237/GameDev.git
cd GameDev
python -m venv venv
source venv/bin/activate
cp .env.example .env  # Fill in DEEPSEEK_API_KEY
pip install -r requirements.txt
npm install -g @modelcontextprotocol/server-filesystem
pip install mcp-server-git
python create_test_project.py
streamlit run app.py

Architecture

User Input -> Skill Match -> Intent Router (3-way): ├── Deterministic (modify_config / modify_code) │ -> LLM parse -> Pydantic validate -> code execute ├── Agent Loop (review / test / translate / analyze) │ -> Plan-Execute: LLM plans steps -> tool calls -> verify └── Supervisor (generate_system / summarize_requirement) -> Agent Loop with multi-step planning

Tech Stack

Python · LangGraph (Checkpoint) · MCP · DeepSeek API · Pydantic · Streamlit · SQLite · Docker

Project Structure

• app.py: Streamlit main entry with scan, chat, recommendations, and history
• pages/: Shared action helpers, skill execution page, and Git panel
• config/: Environment settings and logger
• database/: SQLite task logs and LangGraph checkpoint storage
• scanner/: Engine-agnostic base scanner and Unity scanner implementation
• agents/: LLM client factory
• mcp_tools/: Custom GameDev MCP server and multi-server client manager
• context/: Skill loading, project schema loading, and built-in Skill .md files
• schemas/: Pydantic output models for structured execution
• graphs/: Router, agent loop, deterministic executor, and safety layer
• logs/: Runtime logs
• output/: Generated output artifacts
• test_project/: Sample Unity project for validation

License

MIT