OpenGame

Why we included this project

Most code agents handle isolated programming tasks well but fall apart when asked to build a whole playable game, where many files have to stay consistent and the scene wiring has to actually work. OpenGame tackles that directly: a template library that grows from experience, paired with a debugging skill that records verified fixes, lets the agent scaffold a stable project and repair integration errors instead of patching one syntax bug at a time. That makes it worth a look for teams experimenting with AI-assisted game prototyping, indie developers who want a fast route from concept to something runnable, and researchers watching how agents manage cross-file state and interactive verification. The project also ships a game-specialized coding model, GameCoder-27B, plus an evaluation pipeline that scores output on build health, visual usability, and intent alignment using headless browser execution and a vision-language judge. If you want a concrete reference for how far an agent can get in a domain that mixes creativity with real engineering constraints, this is it.