How to use Claude to build playable Godot games
Godogen combines two Claude Code skills with a custom Godot reference system, headless scene construction, and separate Gemini Flash screenshot evaluation to generate playable Godot 4 projects.
news.ycombinator.com
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Build [game concept] as a Godot 4 project. Use [GDScript references] and load needed APIs selectively. Construct scenes with headless scripts and correct node ownership. Generate [assets], implement gameplay, and have a separate visual QA agent compare screenshots with [reference image].
Reach people exploring AI agents
See placements →What happened
The author describes complete playable Godot 4 projects and visual QA that catches z-fighting, floating objects, physics explosions, and inappropriate grid-like placement.
Reported by htdt on Hacker NewsHow the agent did it
- Built a custom GDScript specification, API reference, and engine quirks database.
- Used headless scripts to construct scene graphs and serialize .tscn files.
- Added separate Gemini Flash visual QA using rendered screenshots and reference images.
Where a person stepped in
The author describes agent-based visual QA and careful prompting, but does not specify human review of each generated game.
A person set the goal and reviewed the result.
Godogen is a pipeline that takes a text prompt and produces a complete, playable Godot 4 project. Its author describes roughly a year of development and four major rewrites. The pipeline designs architecture, generates 2D or 3D assets, writes GDScript, and tests the game visually.
To address unreliable GDScript generation, the author built a reference system containing a handwritten language specification, API documentation converted from Godot's XML source, and a database of engine quirks. The agent loads specific APIs as needed rather than putting all roughly 850 classes into context. Scenes are built by headless scripts that construct node graphs in memory and serialize them into .tscn files. Prompting distinguishes build-time APIs from runtime features and addresses node ownership needed for saving.
A separate Gemini Flash agent performs visual QA. It sees rendered screenshots and a generated reference image, without access to code, to detect problems such as floating objects or physics explosions. Two Claude Code skills divide orchestration from execution, with task execution in forked contexts. Who it's for: developers experimenting with prompt-driven Godot projects who need engine-specific references and visual evaluation alongside code generation.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · htdt
Frequently asked questions
Why does Godogen load Godot references selectively?
The author says roughly 850 Godot classes would overwhelm context. The agent lazy-loads only the APIs it needs from the custom reference system.
How does the visual QA agent evaluate games?
A separate Gemini Flash agent sees screenshots from the running engine and compares them with a generated reference image. It receives no code and catches visual problems text analysis misses.
More details
Remaining steps
- Organized orchestration and task execution into two Claude Code skills.
About this account
Claude Code skills orchestrate architecture, assets, code, and testing; a separate Gemini Flash agent evaluates screenshots without reading the implementation.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47400868. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: htdt on Hacker News
Source published 16 Mar 2026
Source checked 14 Oct 2026
Tools and topics
Claude Code · Godot 4 · Gemini Flash
Godot · GDScript · Visual QA
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.