game-creator
Go from game idea to deployed, monetized browser game in minutes — not hours.
Tell your AI coding agent "make a space invaders game" and get a fully playable 2D or 3D browser game with pixel art, procedural audio, and automated QA. Deploy instantly to the web. Monetize with Play.fun. Share on Moltbook.
Works with 40+ AI coding agents — Claude Code, Cursor, Windsurf, Cline, and more via npx skills add.
Owner: OpusGameLabs
Install
npx skills add OpusGameLabs/game-creator
Target a specific agent:
npx skills add OpusGameLabs/game-creator -a claude-code
npx skills add OpusGameLabs/game-creator -a cursor
npx skills add OpusGameLabs/game-creator -a codex
npx skills add OpusGameLabs/game-creator -a antigravity
Quick Start
# Build a complete 2D game (scaffold → pixel art → design → audio → deploy → monetize)
# QA subagent runs after every step (build, runtime, gameplay, architecture, visual)
/game-creator:make-game 2d my-game
# Build from a tweet
/game-creator:make-game https://x.com/user/status/123456
# Or run individual steps on an existing game:
/game-creator:add-assets # Replace shapes with pixel art sprites
/game-creator:design-game # Add visual polish (particles, juice, transitions)
/game-creator:add-audio # Add chiptune music and sound effects
/game-creator:add-feature jetpack # Add a gameplay feature
/game-creator:improve-game # Audit + implement highest-impact improvements
/game-creator:monetize-game # Register on Play.fun, add SDK, get monetized URL
/game-creator:review-game # Code review for architecture + best practices
/game-creator:qa-game # Add Playwright QA tests
How It Works
/make-game is an orchestrator that delegates all code writing to subagents and runs a QA subagent after every code-modifying step:
Step 0 Parse args, create task list ← main thread
Step 1 Scaffold game from template ← code subagent → QA subagent
Step 1.5 Add pixel art sprites (2D only) ← code subagent → QA subagent
Step 2 Visual polish (particles, juice, transitions) ← code subagent → QA subagent
Step 3 Audio (Strudel.cc BGM + SFX) ← code subagent → QA subagent
Step 4 Deploy to here.now ← main thread (instant publish)
Step 5 Monetize with Play.fun ← main thread (interactive auth)
The QA subagent runs 5 phases per step: build check, runtime check (headless Chromium), gameplay verification (iterate client with game-specific actions), architecture validation, and visual review (Playwright MCP screenshots). If any phase fails, an autofix subagent patches the code and QA re-runs (up to 3 attempts per step).
Architecture
Every game follows the same architecture, whether 2D or 3D:
src/
├── core/
│ ├── EventBus.js # Singleton pub/sub — all cross-module communication
│ ├── GameState.js # Centralized state with reset() for clean restarts
│ ├── Constants.js # Every magic number, color, timing, speed
│ └── GameConfig.js # Engine config (Phaser or Three.js)
├── scenes/ # Scene lifecycle (Phaser) or states (Three.js)
├── entities/ # Game objects (player, enemies, obstacles)
├── systems/ # Background, particles, spawners
├── sprites/ # Pixel art data (palette, matrices, tiles)
├── audio/ # Strudel.cc procedural audio
│ ├── AudioManager.js # Init, play, stop wrapper
│ ├── AudioBridge.js # Wires EventBus → AudioManager
│ ├── music.js # Background music patterns
│ └── sfx.js # Sound effect patterns
├── playfun.js # Play.fun SDK integration
└── main.js # Entry point + render_game_to_text() + advanceTime()
EventBus — Modules never import each other. All communication via pub/sub with domain:action event names.
GameState — Single source of truth. reset() gives a clean slate for restarts.
Constants — Zero hardcoded values in game logic. Sizes are proportional (GAME.WIDTH * ratio), with DPR-aware scaling for retina displays.
render_game_to_text() — Every game exposes window.render_game_to_text() returning a JSON string of current game state, so AI agents can read the game without interpreting pixels.
Skills
Reference Skills (loaded automatically by Claude when relevant)
PlayableIntelligence