Meng To open-sources Three.js game skills
Designer Meng To released three open-source agent skills for Three.js game development following a 24-hour build of a cyberpunk driving demo using GPT-6.2 Astra. The modular skills provide structured workflows for coding agents to generate animated point-cloud mesh scans, lens raindrop shaders, and full-body 3D character selectors.
Packaging domain-specific Three.js implementation patterns into versioned agent skills shows how AI coding is maturing from generic autocomplete into modular, repeatable graphics engineering.
- –Solves procedural WebGL complexity by giving agents precise mathematical heuristics and shader logic rather than letting them guess vertex manipulation.
- –Demonstrates how frontier reasoning models like Astra can coordinate complex multi-system 3D web games within tight timeframes when guided by structured playbooks.
- –The point-cloud scan and raindrop refraction skills bridge raw Three.js mechanics with high-end aesthetic fidelity, minimizing common AI-generated visual slop.
- –Establishes a practical template for sharing specialized agent workflows across Claude Code, Cursor, and Codex ecosystems.
DISCOVERED
26m ago
2026-11-21
PUBLISHED
54m ago
2026-11-21
RELEVANCE
AUTHOR
MengTo