VibeGame: Adversarial Multi-Agent Team with AI-Native Engine for Full Game Dev
机器之心 · wechat · 2026-08-31
Nanjing University and NTU introduce VibeGame, a framework enabling sustained game development from prompts. Key innovations include:
- AI-Native Engine: Built on Phaser, it represents all states as type-checked JSON files, allowing agents to pause execution frame-by-frame, inject inputs, and read source code for full control.
- Adversarial Agent Team: Eight roles (Producer, Designer, Artist, Architect, Coder, Auditor, Player, QA) work with a strict rule: output is verified by a non-generating role. The workflow aligns intent, parallelizes dev, and iterates via adversarial feedback.
- Self-Evolution: The system accumulates experience as Skeletons, Modules, and Contracts without model updates, making the team more efficient over time.
It supports creation from scratch, concept-based dev, and arbitrary editing of existing projects.
More from coding & agent
- Open source workflow: Local context layer for AI coding agents — No_Professional_4310 · 2026-08-31
- Claude Code on WoW Addons: Success Reading Code, Failures Detecting API Changes — This_Cell_1829 · 2026-08-31
- Grok reminder: Underestimating UX hinders Agent adoption — nikvassev · 2026-08-31
- Open-Source Project Runs a Never-Ending AI Livestream on Twitch with FastH3 — VoidAsuka · 2026-08-31
- Building a Self-Improving Hermes Agent Org on Four Pillars — Saboo_Shubham_ · 2026-08-31
- How would you build an orchestrator-based multi-agent software development pipeline? — juniorrafael · 2026-08-31