Specialized roles
Planning, implementation, asset, and QA responsibilities remain explicit instead of collapsing into one prompt.
Back to Open SourceEarly-stage infrastructure
ArcadeRelay
Turn ideas into playable games with AI.
An experimental harness designed to connect implementation, play, evaluation, and iteration through AI agents, turning game ideas into working prototypes.
From brief to playable build
ArcadeRelay relays work across specialized agents and pauses at explicit checkpoints for human direction.
- 01
Brief
Start from one game idea and the experience it should create.
- 02
Plan
Creative, game-design, technical, art, and audio roles shape an executable plan.
- 03
Build
Gameplay and UI roles turn the approved plan into a browser-playable Phaser prototype.
- 04
Playtest and review
Design, art, and QA roles inspect the build and run real browser playtests.
Design decisions
Visible checkpoints
Human approval stays at concept selection, prototype feedback, and final delivery.
Engine direction
The current path targets Phaser 3 while the long-term contract aims to support multiple game runtimes.
Current status
The public repository currently contains the Claude Code harness, agent prompts, workflow definitions, and generated-output directories. It targets Phaser 3, TypeScript, and Vite today and should be presented as early-stage infrastructure.
Direction
The long-term goal is an engine-free harness that can preserve the same idea-to-playable workflow across different game runtimes.