Everything Game Dev Code

A universal scaffold for AI-assisted game development.

Multi-engine. Multi-harness. One coordinated workflow.

Unity · Unreal Engine · Godot · HTML/JS — 2D and 3D, strict engine isolation, shared standards.

---

Not just a prompt collection — a structured operating system for game projects that combines:

• Rules for policy and standards
• Agents for role specialization
• Commands for repeatable entry points
• Skills for reusable execution patterns
• Contexts for phase-specific behavior
• Hooks for workflow automation
• Harness adapters for Claude, Codex, Cursor, OpenCode, Kiro, Grok Build, and Kilo (see docs/harness-support.md for what each one supports)

Quickstart

git clone https://github.com/MRCalderon3D/everything-game-dev-code.git
cd everything-game-dev-code
npm run setup:hooks

npm run setup:hooks wires the repo's pre-commit checks (structure sync + validation) so committed artifacts never drift; it needs Node.js 18+ and takes one second. Run npm run doctor at any time to diagnose your installation (environment, hooks, active engine profile, generated-artifact drift) with remediation hints.

Open the folder in your AI coding assistant (Claude Code, Cursor, Codex, OpenCode, Kiro, Grok Build, or Kilo). The scaffold is loaded from the assistant's adapter plus the shared AGENTS.md, rules/, commands/, agents/, and skills/ layers.

Then type commands in the chat:

Command	What it does
/plan	Outline your project before coding
/gdd	Generate a Game Design Document
/tdd	Apply test-driven development to the current task
/scene-bootstrap	Scaffold a new scene
/unity-setup	Bootstrap a Unity project with conventions
/unity-build-fix	Diagnose and fix Unity build errors
/godot-setup	Bootstrap a Godot project
/unreal-setup	Bootstrap an Unreal project
/web-setup	Bootstrap a web (HTML5) project, 2D canvas or 3D WebGL
/generate-assets	Generate real assets — images, skyboxes, 3D models, SFX, music, voice, video — from text prompts
/full-game	Orchestrate an entire game from scratch (experimental)

You don't have to follow a specific order. Pick whatever command fits your current need — start a new project, generate a GDD for an existing one, run a QA review, or fix a build error.

Step-by-step guide

The guides/Dash & Collect/ folder contains a full tutorial that walks through building a game using the scaffold's commands, agents, skills, and contexts across all project phases.

Example projects

The samples/ folder contains thirteen complete HTML games built with the /full-game command in a single pass each (PirateInvaders, Tetris2DMutation, LosRenacidos, pacmanAI, harness-comparison variants, PrismDefense3D — the first 3D sample, built on Three.js against the web 3D rules, plus its CC0-textured Visuals variant and its AssetsGen variant where every model, texture, skybox, sound, and the intro cinematic were generated through the /generate-assets layer, and NebulaLance — a 2.5D R-Type-style shoot-'em-up whose ship, enemies, boss, nebula, audio, and image-to-video menu cinematic are all AI-generated), plus a writeup comparing how different AI harnesses performed on the same brief. For real projects, we recommend going step by step.

Goals

• Keep shared game-development standards engine-neutral.
• Let Unity, Unreal, Godot, and the web layer each extend the base cleanly without contaminating one another.
• Support real production work across design, engineering, content, QA, release, and live ops.
• Turn repeated solutions into reusable skills and structured workflows.
• Make the repository portable across multiple coding assistants and harnesses.

Repository Model

This scaffold is organized in layers:

• rules/ — what good looks like
• agents/ — who does the work
• commands/ — how work starts
• skills/ — how work is executed well
• contexts/ — how priorities shift by phase
• hooks/ — how workflow safeguards are enforced
• manifests/ — how subsets are installed by profile, plus the engine layer registry (engines.json), the harness adapter registry (harnesses.json), and the generative asset provider registry (asset-providers.json)
• schemas/ — JSON validation for manifests, hooks, and plugins
• scripts/ — generators, validators, and diagnostics (new:engine, doctor, sync:*)
• docs/templates/ — structured templates for GDD, TDD, QA plans, and other deliverables
• docs/orchestration/ — agent routing, role handoffs, and workflow sequences
• tests/ — how the scaffold verifies itself
• harness adapters — how different AI clients consume the same source of truth

Engine Isolation Policy

The repository is intentionally split into:

• rules/common/
• rules/unity/
• rules/unreal/
• rules/godot/
• rules/web/

And equivalent skill / command / review layers where needed.

Shared documents should describe intent, ownership, and quality bars. Engine-specific files should describe implementation conventions inside that engine only.

AI Asset Generation (optional)

The scaffold ships an optional engine-neutral asset generation layer. It is an upgrade, not a dependency: the default asset workflow is unchanged, and AI generation only activates when a provider API key is present in the environment.

Without a key (default). Use the scaffold's built-in tooling exactly as always: the engine placeholder commands (/unity-placeholders, /godot-placeholders, /web-placeholders) and the procedural / Canvas / WebAudio pipelines. A game built entirely on placeholders is a complete, valid result — nothing about the workflow requires an API.

With a key (FAL_KEY for the default fal.ai provider). /generate-assets becomes available to upgrade placeholders to real content from text prompts:

1. Generate placeholders first so every asset has a stable name and path.
2. Run /generate-assets to produce images and textures, equirectangular skyboxes, 3D models (GLB/OBJ), sound effects, music, voice lines, and intro/cinematic video — dropped onto those same names and paths (zero code changes).
3. Import through the active engine layer and review with the matching pass command (/art-2d-pass, /art-3d-pass, /audio-pass).

Resolution order, quality, and cost. For images and skyboxes the scaffold prefers a free harness-native generator (e.g. Codex $imagegen) when the running harness has one, then the paid API, then placeholders; 3D/audio/video only have the API or placeholder paths. Quality vs price is the user's dial: every capability has budget / balanced / premium tiers (--quality, default balanced) — a full asset pass runs roughly $2-3 / $5-7 / $10+ respectively. Because the API costs money, generation also asks before spending: every run prints a cost estimate, and scripts/generate-assets.js refuses runs at or above a configurable threshold (confirmOverUsd) without explicit confirmation. Autonomous flows like /full-game keep placeholders as the default and ask once — tier and spend — before generating.

Capability-to-model routing lives in manifests/asset-providers.json (default provider: fal.ai — one pay-per-use key covers every modality). scripts/generate-assets.js performs the generation and writes a .provenance.json sidecar (provider, model, prompt, seed, request id) per run; assets without provenance are treated as unlicensed content. The key is read only from the FAL_KEY environment variable and is never committed (.env files are gitignored). Governance rules — the optional-capability, resolution-order, and cost-confirmation policy — live in rules/common/asset-pipeline.md. For a worked end-to-end example see the samples/PrismDefense3DAssetsGen variant.

Intended Use Cases

• New game project setup
• Multi-engine studio workflows
• Internal AI workflow standardization
• GDD and technical design maintenance
• QA and release readiness reviews
• Plugin / content / tooling governance
• Cross-discipline planning and orchestration

Supported Harnesses

• Claude Code
• Codex
• Cursor
• OpenCode
• Kiro
• Grok Build
• Kilo

Each harness adapter points back to the same shared scaffold rather than becoming a second source of truth.

MCP servers

The scaffold curates a set of MCP servers (browser playtesting, AI asset generation, Blender/DCC control, and more) in mcp-configs/mcp-servers.json, with ready-to-use, harness-native configs generated into mcp-configs/generated/. MCP is vendor-neutral, so they work with any MCP-capable client. See docs/mcp-setup.md for per-harness install and a Blender end-to-end walkthrough.

Codex usage

Codex should start from AGENTS.md and .codex/README.md. Slash-style commands such as /plan, /gdd, and /unity-review map to commands/<name>.md; if the Codex client does not execute slash commands natively, type the command name in chat and ask Codex to run the matching scaffold command. For raster-first asset work in Codex, use $imagegen alongside the shared art pipeline skills. Good fits include concept sheets, sprite source art, UI mockups, marketing key art, painted backgrounds, and bitmap edits; keep vector/code-native asset work in the normal repo workflows. If a generated image becomes a real project asset, move the selected file from $CODEX_HOME/generated_images/... into the workspace and keep the scaffold's naming and folder conventions.

Current Status

The scaffold is intentionally modular. Different blocks may be added or replaced over time, but the repository should always preserve:

• flat agent and command structures
• layered rules
• grouped skills
• engine isolation
• harness portability

Principles

• Design before implementation
• Explicit ownership over implicit assumptions
• Testability over cleverness
• Documentation that supports execution
• Measured performance and release readiness
• Accessibility, QA, and compliance as first-class requirements

License

This repository is provided under the MIT License unless you replace it with your studio’s internal licensing policy.