htdt/godogen

Autonomous game development for Godot, Bevy, and Babylon.js with Claude Code and Codex.

Watch the demos · Prompts

Describe a game. The agent builds it, generates assets, runs the engine, and proves the result — as a live game you watch and steer, or as a recorded video when you're not there. It reads the situation and decides which, in the run.

This repo is not a game. It is the source for a generator that produces games: godogen -> game repo -> game. You publish into a fresh game repo — choosing engine and host-agent flavor — then the agent runs inside that repo and builds the actual game from a short engine guide.

A published repo is intentionally thin: a runtime manifest, a one-page engine guide, and the asset-generation skill. The agent recreates everything else (project scaffold, capture tooling) from the guide.

• prompts/runtime.md — the runtime manifest
• asset-gen/ — the cross-engine asset-generation skill
• engines/babylon.md, engines/godot.md, engines/bevy.md — per-engine guides
• publish.sh — renders the runtime layout for the chosen engine and host agent

Engine and host agent (Claude vs Codex) are publish-time render choices, not separate source trees.

• Godot 4 — C#/.NET projects with build-time scene generation, runtime scripts, and Jolt physics.
• Bevy — Rust/Bevy projects with code-first ECS scenes and offscreen capture.
• Babylon.js — TypeScript/Vite browser games served at a live URL.
• Asset generation — Gemini or xAI Grok for images (whichever key is set; with both, quality-critical assets are generated on each and the better kept), Tripo3D for image-to-3D and rigged biped animation; animated sprites via Grok video with loop detection and background removal.
• Proof over claims — the agent judges results from the running game (a live URL or a recorded clip), not from a clean compile, so visible defects drive the next iteration.
• You choose your involvement — watch the live game (a Babylon.js URL, or a Godot/Bevy project you run) and steer at decision points, or leave the run unattended and get a 15–20s proof recording at the end. The agent takes its cue from how you frame the task.

• Godot 4 (.NET build) on PATH for Godot projects
• Rust/Cargo for Bevy projects
• Node.js 20+ and npm (22.12+ for Babylon.js projects), plus the Tripo CLI: npm install -g tripo-cli
• Chrome or Chromium with hardware WebGL2 for Babylon.js browser capture
• Python 3 with pip
• API keys as environment variables:
  • GOOGLE_API_KEY — Google AI Studio for Gemini image generation
  • XAI_API_KEY — xAI Grok for image generation and animated-sprite video (either image key is enough)
  • TRIPO_API_KEY — Tripo for 3D generation (used by the tripo CLI)
• System packages from setup.md: vulkan-tools, xvfb, ffmpeg, imagemagick, plus platform-specific extras
• Tested on Ubuntu, Debian, and macOS
• Claude Code or Codex

Pick the engine and host agent:

./publish.sh --engine godot   --agent claude --out ~/my-game       # CLAUDE.md + .claude/skills/
./publish.sh --engine babylon --agent codex  --out ~/my-game       # AGENTS.md + .agents/skills/
./publish.sh --engine bevy    --agent claude --out ~/my-game

Pass --force to wipe existing contents at the target before re-publishing.

A full generation run can take hours, so it's convenient to offload it to a server — ideally a GPU instance, since engine rendering and video capture are much faster with hardware acceleration.

• Keep the session alive across SSH drops with tmux or screen.
• Enable remote control so you can check in and steer the run from any device — both Claude Code and Codex have official remote-control interfaces.

See CHANGELOG.md.

Follow progress: @alex_erm