OpenGame

OpenGame is an AI-driven game generation framework that converts natural language game descriptions into playable Phaser 3 browser games through an LLM-powered tool chain.

Repository: github.com/leigest519/OpenGame License: MIT (inferred from package.json) Core Stack: TypeScript, Phaser 3, Node.js, LLM APIs

What It Does

You type a game idea in plain English (e.g., "a Mario-style platformer where you play as a cat"), and OpenGame:

Classifies the game archetype based on physics (gravity? perspective? movement type?)
Generates a Technical Game Design Document (GDD) tailored to the archetype
Scaffolds the project from pre-built templates
Generates assets (images, audio, tilemaps) via AI
Implements code using the template's Hook / Behavior system
Builds and tests the final game

Architecture

CLI + Core SDK

packages/cli — Command-line interface (opengame command)
packages/core — Tool chain: GDD Generator, game type classifier, asset generator, tilemap generator, code generator
packages/sdk-typescript — TypeScript SDK for programmatic use

Template System (`agent-test/templates/`)

Five archetype families, each with its own BaseScene hierarchy, entity behaviors, and design rules:

Archetype	Physics	Perspective	Examples
platformer	Gravity ON	Side view	Mario, Terraria
top_down	Zero gravity	Top-down	Zelda, Isaac
grid_logic	Discrete steps	Top-down	Sokoban, Fire Emblem
tower_defense	Path-following	Top-down	Kingdom Rush
ui_heavy	None	N/A	Slay the Spire, Visual Novels

AI Tool Chain (`packages/core/src/tools/`)

Tool	Purpose	Key File
`classify-game-type`	Physics-first archetype classification	`game-type-classifier.ts`
`generate-gdd`	Technical GDD generation with 6 sections	`generate-gdd.ts`
`generate_game_assets`	AI image/audio generation	(asset pipeline)
`generate_tilemap`	ASCII → tilemap JSON	(tilemap pipeline)
`todoWrite`	Task tracking	`todoWrite.ts`

Key Design Philosophy

"Constraints Over Generation"

The LLM does NOT write code from scratch. It fills templates under strict constraints. The Template API is the boundary of capabilities; Prompt Engineering guards the boundary.

This prevents LLM hallucinations from producing non-compiling code — a common failure mode in AI code generation.

Workflow

User Description
    ↓
[Phase 1] classify-game-type → archetype
    ↓
[Phase 2] generate-gdd → GAME_DESIGN.md (6 sections)
    ↓
[Phase 3] generate_game_assets + generate_tilemap → assets
    ↓
[Phase 4] Merge config → gameConfig.json
    ↓
[Phase 5] Code implementation → TypeScript files
    ↓
[Phase 6] Build + Test + Verify

platformer-vs-topdown-basescene — Deep dive into BaseScene architecture differences
ai-gdd-prompt-engineering — GDD generation prompt engineering patterns
game-design-document — General GDD templates and guidelines
phaser-vue-integration — Phaser 3 + Vue 3 integration patterns