#Three.js Game Skills: Turning an AI Coding Agent Into a Browser Game Studio

An open-source skill pack attempts to turn Codex or Claude Code into a small Three.js game studio with a single prompt, routing work across nine specialized roles.

#What it is

Three.js Game Skills is an open-source repository that packages a set of self-contained skills for AI coding agents. The idea is that instead of asking one agent to build an entire game from scratch, you install the skill pack, call a single director command, and describe the game you want. The director then routes the work across nine specialized skills covering gameplay, graphics, user interface, debugging, release QA, and optional 3D, image, and audio generation.

The workflow is built around producing playable browser games, not static 3D scenes to look at. It starts with a playable loop, then upgrades visuals, HUD, effects, and audio on top of that foundation. The repo supports both Codex and Claude Code, and the core works without paid APIs by falling back to local or procedural assets.

#Why it matters

• It demonstrates a practical multi-agent pattern: one director routes to specialized skills instead of forcing you to remember and invoke each one manually.
• The workflow prioritizes a playable game first, which is harder and more useful than generating a pretty 3D demo.
• Generated projects include seeded randomness, deterministic test hooks, and Playwright templates for smoke tests, visual regression, and bot playtests.
• The agent is expected to self-check before declaring a task complete, which is a meaningful step beyond raw code generation.
• The core runs without paid APIs, with optional integration for Tripo (3D models), Gemini (images), and ElevenLabs (audio) when keys are available.

#How it works

You install the skill pack and invoke the threejs-game-director with a description of your game. The director selects the appropriate specialist rather than making you choose. The nine skills cover gameplay logic, AAA-style graphics, UI, debugging, and release QA, plus optional branches for 3D model generation, image generation, and audio generation. Each skill is self-contained, so the director can compose them without tight coupling.

The QA layer is what sets this apart. Games come out with seeded RNG so tests are reproducible, fixed test hooks for deterministic behavior, and Playwright templates that let the agent run smoke tests, visual regression checks, and bot playtests. The agent must verify its own output before it can claim the job is done.

#Caveats

The repo is MIT licensed and has several demo games running in the browser, but it is still a skill pack layered on top of AI coding agents, not a mature game engine. Output quality depends heavily on the underlying model and your prompt clarity. The optional asset generation branches require paid API keys, and without them you fall back to local or procedural assets, which may not match the visual quality you see in polished demos.

#Who it's for

This is for developers experimenting with AI-assisted game development who want a structured, test-backed workflow rather than ad-hoc prompting, and who are comfortable with Three.js and agent tooling.

If you want to see what a disciplined, multi-skill approach to AI game generation looks like, Three.js Game Skills is worth a serious look, especially for the QA-first design.