OpenClaw skill conversion of htdt/godogen — Claude Code skills that build complete Godot 4 projects from a game description.
Original project by @alex_erm. All credit for the skill design, prompts, GDScript reference, and tooling goes to them.
The original godogen is a set of Claude Code skills that orchestrate an AI pipeline to generate complete Godot 4 games from natural language. You describe what you want, and it designs the architecture, generates art, writes every line of code, captures screenshots from the running engine, and fixes what doesn't look right.
This repo is a direct conversion of those skills to the OpenClaw AgentSkill format, making them usable with OpenClaw's agent runtime and sub-agent orchestration.
${CLAUDE_SKILL_DIR}/path references → skill-relative pathsSkill(skill="godot-task")withcontext: fork→ OpenClaw sub-agent pattern (sessions_spawn)$ARGUMENTSplaceholder → explicit task passing via spawn message- Teleforge (Telegram bridge) references → removed (OpenClaw handles messaging natively)
- Claude Code frontmatter (
context: fork) → OpenClaw frontmatter (name+descriptiononly) tools/directories →scripts/(OpenClaw convention)- Root-level
.mdfiles →references/(OpenClaw convention)
No changes were made to the actual skill logic, prompts, GDScript reference, Python tooling, or Godot API docs. The conversion is purely structural.
The main pipeline skill. Generates a visual target, decomposes the game into tasks, scaffolds the Godot project, plans and generates assets, then delegates each task to a sub-agent.
Runs a single development task from the plan. Generates scenes and scripts, validates compilation, creates test harnesses, captures screenshots, and runs visual QA against the reference image. Spawned by the orchestrator for each task with a fresh context.
godogen/
├── SKILL.md # Orchestrator instructions
├── references/
│ ├── visual-target.md # Reference image generation
│ ├── decomposer.md # Game → task decomposition
│ ├── scaffold.md # Project architecture & skeleton
│ ├── asset-planner.md # Asset budgeting & planning
│ └── asset-gen.md # Image & 3D model generation
└── scripts/
├── asset_gen.py # Gemini image + Tripo3D GLB generation
├── rembg_matting.py # Background removal
├── spritesheet_slice.py # Sprite sheet processing
├── spritesheet_template.py # Sprite sheet grid template
├── tripo3d.py # Image-to-3D model conversion
└── requirements.txt
godot-task/
├── SKILL.md # Task executor instructions
├── references/
│ ├── gdscript.md # GDScript syntax reference
│ ├── quirks.md # Known Godot gotchas
│ ├── scene-generation.md # Headless scene building
│ ├── script-generation.md # Runtime script patterns
│ ├── coordination.md # Scene + script ordering
│ ├── test-harness.md # Visual test harness creation
│ ├── capture.md # Screenshot/video capture
│ ├── visual-qa.md # Gemini vision QA pipeline
│ ├── dynamic_prompt.md # Dynamic VQA prompt
│ ├── static_prompt.md # Static VQA prompt
│ └── doc_api/ # 862 Godot class API references
│ ├── _common.md # Index of ~128 common classes
│ ├── _other.md # Index of ~732 remaining classes
│ └── {ClassName}.md # Per-class API docs
└── scripts/
├── visual_qa.py # Gemini vision QA script
├── class_list.py # Godot class listing
├── ensure_doc_api.sh # Doc bootstrapper
└── godot_api_converter.py # API doc converter
- OpenClaw installed and configured
- Godot 4 (headless or editor) on PATH
- Python 3 with pip
- API keys as environment variables:
GOOGLE_API_KEY— Gemini (image generation + visual QA)TRIPO3D_API_KEY— Tripo3D (image-to-3D, only needed for 3D games)
Copy both skill directories into your OpenClaw workspace skills folder:
cp -r godogen godot-task ~/.openclaw/workspace/skills/Then tell your OpenClaw agent to "make a game" and describe what you want.
Same as the original: MIT