Sandbox
@htdt/godogen

Game development harness for Claude Code and Codex

Godogen turns a short game idea into an agent-run game repo. The published repo contains a runtime manifest, an engine guide, and the asset-generation skill, while the agent recreates the rest of the project from those files. It supports Godot 4, Bevy, and Babylon.js, and it is built around proof from the running game instead of just a successful build. You can watch the agent work live or let it produce a short recording at the end.

6,819 stars633 forksPythonUpdated 12d ago
Godogen: From a Text Prompt to a Godot 4 Game
Alex Ermolov21k views • 6 months ago
Who it's for

Builders who want Claude Code or Codex to build, run, and verify games from a small set of project instructions.

What it delivers

You can hand an agent a game idea and get a working game with visible proof instead of a pile of code claims.

What it does

Engine-specific guides

Includes `engines/godot.md`, `engines/bevy.md`, and `engines/babylon.md` for the three supported game stacks.

Runtime manifest

Uses `prompts/runtime.md` as the thin source of truth that gets published into the game repo.

Asset generation skill

Packages `asset-gen/` for cross-engine asset creation, including `SKILL.md` and supporting files.

Publish step

`publish.sh` renders the chosen layout for Claude or Codex into a fresh game repository.

Proof-based iteration

The agent checks the running game or recorded clip and uses visible defects to drive the next pass.

Live or unattended runs

You can steer a live Babylon.js URL or Godot/Bevy run, or leave it unattended and review a proof recording.

How to get it

  1. 1Pick 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

README

Godogen

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

Watch the video

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.

Source layout

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.

What the agent does

  • 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 for precise references and characters, xAI Grok for textures and simple objects, 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.

Getting started

Prerequisites

  • Godot 4 (.NET build) on PATH for Godot projects
  • Rust/Cargo for Bevy projects
  • Node.js 22.12+ and npm for Babylon.js projects
  • Chrome or Chromium with hardware WebGL2 for Babylon.js browser capture
  • Python 3 with pip
  • API keys as environment variables:
    • GOOGLE_API_KEYGoogle AI Studio for Gemini image generation
    • XAI_API_KEYxAI Grok for image/video generation
    • TRIPO3D_API_KEYTripo3D for 3D generation
  • System packages from setup.md: vulkan-tools, xvfb, ffmpeg, imagemagick, plus platform-specific extras
  • Tested on Ubuntu, Debian, and macOS
  • Claude Code or Codex

Publish a game repo

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.

Running on a server

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.

Changelog

See CHANGELOG.md.

Follow progress: @alex_erm

Files in the repo

Repository payload15 top-level entries
  • .github
  • asset-gen
  • docs
  • engines
  • prompts
  • scripts
  • .gitignore
  • AGENTS.md
  • CHANGELOG.md
  • CLAUDE.md
  • CONTRIBUTING.md
  • LICENSE.md
  • publish.sh
  • README.md
  • setup.md

Discussion (0)

Ask about usage, or say what you built with it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More harnesses

affaan-m/
ECC
affaan-m/ECCHarnesses

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

258k
ruvnet/rufloHarnesses

🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated

72k

Practical patterns, starters & CLI tools for loop engineering with AI coding agents. Design systems that prompt and orchestrate agents (inspired by Addy Osmani and Boris Cherny). Includes loop-audit, loop-init, loop-cost.

11k