Autonomous text-to-game pipeline for Godot, powered by Claude Code,Codex,Opencode
# Add to your Claude Code skills
git clone https://github.com/RandallLiuXin/GodotMakerGuides for using ai agents skills like GodotMaker.
Last scanned: 6/5/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-06-05T08:08:05.355Z",
"npmAuditRan": true,
"pipAuditRan": true
}English | 中文
In game development, especially early ideation and market validation stages, teams often come up with more ideas than they can develop. The usual approach is to discuss and pick one idea for development, only to find out after a couple of weeks that the idea is not viable, making the previous discussion and development efforts wasted. I built GodotMaker to allow individuals to turn their ideas into playable prototypes quickly, validating the feasibility and fun of their ideas. This will significantly speed up the early stages of game development, helping developers find truly worthwhile game concepts faster.
Many tools promise that AI can help you make games. Once you actually try to build with them, the same problems tend to show up:
GodotMaker takes a different path: bring the game idea, let it shape that idea into a GDD, then let agents run through planning, implementation, tests, gameplay runs, screenshots, evaluation, and fixes. When the run finishes, you review a real Godot project on your disk.
The code is yours. The GodotMaker framework is source-available, the workflow is local-first. Want a better game? Refine the idea or GDD and run another iteration.
External agent runtimes and model providers, such as Claude Code, Codex, Gemini, OpenAI, xAI, or Tripo, may have their own pricing, quotas, and data policies. GodotMaker keeps the framework source-available, the workflow local-first, and the generated project on your machine.
During a run, GodotMaker agents keep moving the design forward:
GDD.md, tasks, scenes, systems, assets, and acceptance criteriaA small prototype usually takes about 5-8 hours of agent runtime. You do not need to manually drive each stage or keep an eye on it the whole time; the workflow is designed to keep going on its own.
npm install -g godotmaker-cli
mkdir my-game
cd my-game
# Bring your game idea, then run:
godotmaker-cli --agent claude-code
Use Codex or OpenCode for the same workflow:
godotmaker-cli --agent codex
godotmaker-cli --agent opencode
The CLI drives the workflow from idea capture and GDD planning to a playable Godot prototype. Agent selection resolves in this order: --agent, project .godotmaker/config.yaml, CLI-global ~/.godotmaker/cli/config.yaml, then the default runner. Advanced users can still run the underlying role commands directly in Claude Code (/gm-*), Codex ($gm-*), or OpenCode (/gm-*).
For framework development:
git clone https://github.com/RandallLiuXin/GodotMaker.git
cd GodotMaker
pip install -r tools/requirements.txt
python tools/check_env.py
| Tool | Why |
|---|---|
| Godot 4.5+ | Runs the generated project |
| Claude Code, Codex, or OpenCode | Command-line agent runtime for CLI-driven runs; desktop chat apps are not sufficient |
| Node.js 22+ | Runs godotmaker-cli and Godot MCP tooling |
| Python 3.10+ | Runs GodotMaker helper scripts |
| Git 2.30+ | Enables local history and agent worktrees |
You only need to set an API key when your project configuration selects an API provider. Image generation and visual QA can use runtime-native providers only when the selected agent runtime supports them; OpenCode projects should configure codex or API-backed image/VQA providers.
GodotMaker is preparing for a public alpha under a source-available license. The CLI, agent-runner support, visual QA, and packaging workflow are evolving quickly. Core gameplay generation is usually the most reliable part of the flow. Because AI output is still variable and the art pipeline is still alpha, a completed run may still need a follow-up pass with a coding agent to correct visual wiring issues such as the wrong atlas region, missing animation setup, or incomplete asset binding.
Current boundaries:
| Capability | Current status | Notes |
|---|---|---|
| 2D only | The current framework targets 2D Godot games. | Do not start with a 3D game prompt. Add 3D content manually after generation if needed. |
| Level generation | Level-based games can be built, but automatic level design is still limited. | Treat generated levels as draft placeholders and expect to adjust layouts manually. |
| Puzzle design and balance | The agent can implement puzzle mechanics, scoring rules, economy hooks, and win/loss logic, but it does not have human taste for whether a puzzle is clever or whether a numeric curve feels good. | Use GodotMaker to build the functional system, then fill in puzzle content, level data, and balance values yourself. |
| Art pipeline | The art pipeline is alpha. It can generate and bind draft assets, but it may occasionally choose the wrong atlas region, miss an animation configuration, or need a follow-up coding-agent repair pass. | Review the generated project visually. Replace assets under assets/ or rerun /gm-asset when the visual direction is important. |
| Pixel art | Pixel art style is not currently supported by the art pipeline, but support is planned. | Use non-pixel 2D styles for generated art, or provide your own pixel art assets manually for now. |
| TileMap | TileMap is not currently supported. Tile-based terrain, tilesets, tile atlases, and grid-based map generation may not be completed automatically. | Use hand-authored scenes or simple generated layouts for now. TileMap support is planned later. |
| Audio generation | Not supported by the current pipeline. Audio rows are treated as user-provided or deferred. | Provide music/SFX manually and wire them into the project yourself until the audio workflow ships. |
| Long automated runs | GodotMaker is cost-sensitive because it runs coding agents for a long time. The current internal baseline is that Codex Pro can handle long prototype runs comfortably, with capacity still left over in typical use. | If you are using a different runtime or plan, start with smaller prompts and watch the first full run. |
| Non-converging runs | A very small number of projects may fail to converge. In this context, "not converging" means the project still cannot pass acceptance after at least 5 build/fix/evaluate loops. | Please share the local failure case if this happens. These reports are useful for improving the framework. |
A dedicated art-production UI is planned to make curation, slicing, replacement, and review more reliable.
If this direction
GodotMaker is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by RandallLiuXin. Autonomous text-to-game pipeline for Godot, powered by Claude Code,Codex,Opencode. It has 469 GitHub stars.
Yes. GodotMaker passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/RandallLiuXin/GodotMaker" and add it to your Claude Code skills directory (see the Installation section above).
GodotMaker is primarily written in Python. It is open-source under RandallLiuXin on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh GodotMaker against similar tools.
No comments yet. Be the first to share your thoughts!