Unity AI Workflow

A comprehensive, agent-first toolkit for AI-assisted Unity development. Works with Claude Code (primary) and Google Antigravity IDE. This repository provides the "brain" (rules, agents, skills, and workflows) to turn your AI assistant into an expert Unity team.

No Unity project inside — this repo is pure workflow infrastructure (rules, agents, skills, docs, templates). You copy it into your game's workspace and the AI reads it as its operating manual.

🎮 Dev Modes

Choose how much the AI does vs. how much you do:

Mode	Who Builds	Best For
Assistant	You build. AI documents, advises, and explains.	Learning, full creative control
Mix (default)	Collaborative. AI suggests; you confirm key decisions.	Most projects
Automatic	AI builds after a short onboarding Q&A.	Rapid prototyping, jam games

Set your mode in docs/ProjectConfig.yaml → ai_mode. You can change it at any time.

Automatic mode success story: "I told the AI my game idea, answered a few setup questions, and it created the scripts, integrated assets, and built a working prototype — minimal manual setup."

---

🧃 Core Philosophy: Game Feel is Not Optional Later

Every feature is built complete using /implement-feature:

interrogation → GDD/GFD docs → TDD implementation → game feel integration → commit

The AI asks about VFX, SFX, camera reactions, and haptics before writing a single line of code. Polish is iterative, not a phase. A feature isn't "done" when it compiles — it's done when it plays well.

---

📂 Recommended Workspace Structure

Important

Don't open the Unity project directly. Instead, create a parent workspace folder containing both your Unity project and docs. This lets the AI read design documents, templates, AND Unity code simultaneously.

MyGame/                          ← Open THIS folder in your AI tool
├── CLAUDE.md                    ← Claude Code instructions (auto-loaded)
├── .claude/commands/            ← Slash commands (Claude Code)
├── .agent/                      ← AI brain (Antigravity + shared skills)
│   ├── rules/
│   ├── agents/
│   ├── skills/                  ← Shared knowledge base (both tools)
│   └── workflows/               ← Antigravity workflows
├── docs/                        ← Living documents (filled during pre-prod)
│   ├── ProjectConfig.yaml
│   ├── GDD.md
│   ├── TDD.md
│   ├── GFD.md
│   └── SprintPlan.md
├── templates/                   ← Pristine starters (for reference)
└── MyGameUnity/                 ← Unity project (created via Unity Hub)
    ├── Assets/
    ├── Packages/
    └── ProjectSettings/

---

🚀 Quick Start

With Claude Code

1. Create your workspace: Create a parent folder for your game (e.g. MyGame/).
2. Copy the toolkit: Copy everything from this repo into your workspace root.
3. Create your Unity project: Inside the workspace via Unity Hub (e.g. MyGame/MyGameUnity/).
4. Open the workspace in your terminal or VS Code with Claude Code.
5. CLAUDE.md is auto-loaded — rules, routing, and workflow suggestions are built in.
6. Run /setup-project: The AI walks you through dev mode selection and full project initialization.

With Antigravity

1. Create your workspace and copy the toolkit (same as above).
2. Create your Unity project inside the workspace via Unity Hub.
3. Open the parent folder (not the Unity folder) in Antigravity.
4. Run /setup-project: Same workflow, powered by .agent/workflows/.
5. Connect MCPs (Agent panel > ... > MCP Servers):
   • Unity MCP: For in-editor scene/prefab/asset management
   • GitHub MCP: For automated git operations
   • Linear/Notion MCP: For task and doc syncing

Both tools share the same skills (.agent/skills/), docs, and templates. You can switch between them freely.

---

🧠 How it Works

• Global Constitution: CLAUDE.md (Claude Code) and RULES.md (Antigravity) enforce Unity 6.2+ best practices, performance standards, and thread safety.
• Auto-Routing: Agent routing table automatically loads the right personas and skills based on your task and ai_mode.
• TCREI Prompting: Agents use the Task-Context-References-Evaluate-Iterate methodology for structured, high-quality outputs.
• Verification System: Every AI recommendation is marked [VERIFIED], [SYNTHESIZED], or [UNVERIFIED] — no silent hallucinations.
• Expert Skills: Dedicated skills for UI Toolkit data binding, ScriptableObject architecture, Netcode, game feel, testing, debugging, and more.
• Integrated Feature Loop: /implement-feature combines code + game feel in one pass, with deep interrogation before any coding starts.
• Session Handoff: SessionState template ensures context survives across AI sessions.
• Guided Workflows: Slash commands for every phase of development.

---

📅 Project Phases

Follow the phase guides in docs/phases/ for a standardized development journey:

1. 00: Ideation — From rough idea to GDD + GFD.
2. 01: Pre-Production — Technical choices, stack definition, naming conventions.
3. 02: Technical Design — Architecture, Assembly Definitions, patterns.
4. 03: Project Setup — Automated folder scaffolding and package installation.
5. 04: Production — Feature-by-feature loop (interrogate → implement → feel → commit).
6. 05: Polish — Final tuning pass, visual refinement, performance profiling.

---

🛠️ Configuration

The AI reads docs/ProjectConfig.yaml to stay in sync with your versions, packages, and architectural choices. Key settings:

ai_mode: "mix"          # assistant | mix | automatic
prototype_mode: false   # Relaxes architecture rules for rapid iteration
auto_confirm: false     # Skip confirmation prompts (overridden by automatic mode)

---

📋 Roadmap

See ROADMAP.md for the full version history and what's planned next.

---

🙏 Resources & Inspiration

This workflow was built with the help of these resources. Ratings reflect how much each influenced the final result.

GitHub Repos

Resource	Rating	Influence
anthropics/skills	⭐⭐⭐⭐⭐	Official skill format — canonical reference for skill structure
sickn33/antigravity-awesome-skills	⭐⭐⭐⭐⭐	Core inspiration for Antigravity-native skill patterns
Common-ka/ai-agent-unity-rules	⭐⭐⭐⭐	Unity-specific AI rules — shaped our rules/agents structure
vudovn/antigravity-kit	⭐⭐⭐⭐	Workflow/skill structure reference for .agent/ layout
obra/superpowers	⭐⭐⭐⭐	Agent persona / superpowers philosophy — influenced agent routing
gsd-build/get-shit-done	⭐⭐⭐⭐	Pragmatic workflow philosophy
CoplayDev/unity-mcp	⭐⭐⭐⭐	Unity MCP integration — directly referenced in MCP setup
stillwwater/UnityStyleGuide	⭐⭐⭐⭐	Naming conventions baseline
justinwasilenko/Unity-Style-Guide	⭐⭐⭐⭐	Folder structure + conventions
nextlevelbuilder/ui-ux-pro-max-skill	⭐⭐⭐	UI skill patterns (adapted for UI Toolkit)
affaan-m/everything-claude-code	⭐⭐⭐	Claude Code ecosystem overview
thedotmack/claude-mem	⭐⭐⭐	Memory/session persistence patterns
hesreallyhim/awesome-claude-code	⭐⭐⭐	Curated list — helped discover other repos
automazeio/ccpm	⭐⭐⭐	Claude Code project management patterns

YouTube Videos

Video	Creator	Rating	Influence
Advanced Project - Simple Game (Unity)	PracticAPI	⭐⭐⭐⭐⭐	Production-quality Unity 6 architecture — SEP, MVC+Commands, DI, state machines, pooling, Addressables. Directly influenced DI skill and class taxonomy
Essential AI Skills For 2026	Tina Huang	⭐⭐⭐⭐⭐	Introduced TCREI methodology and key tools we implemented
You're Using Claude Code Wrong — 10 Tips	Simon Scrapes | AI Automation	⭐⭐⭐⭐½	Inspired the verification/hallucination-checking system
How I Use Claude Code (Meta Staff Engineer Tips)	John Kim	⭐⭐⭐½	Helpful Claude Code usage patterns
Google's AI Prompt Engineering Course in 20 Min	Tina Huang	⭐⭐⭐	Good prompting fundamentals

Web Resources

Resource	Rating	Influence
PracticAPI — The Perfect Unity Project	⭐⭐⭐⭐⭐	Reference architecture for production Unity 6 — MVC, Commands, DI, state machines, 14 documented techniques
Unity — Organizing Your Project	⭐⭐⭐⭐⭐	Official guidance — directly shaped recommended folder structure
TCREI Prompt Engineering Framework	⭐⭐⭐⭐⭐	TCREI methodology is baked into AGENTS.md
Loic Jacob — Game Feel Theory	⭐⭐⭐⭐⭐	ADSR envelope, Three Pillars, Sensation Vocabulary
skills.sh	⭐⭐⭐⭐	Skill discovery catalog
Unity AI Features	⭐⭐⭐	Context for Unity's AI direction
Unity Sentis Documentation	⭐⭐⭐	Reference for in-engine AI inference
Unity 6 AI Manual	⭐⭐⭐	Future-facing Unity AI reference

Plus various AI-assisted research sessions (Perplexity, Google AI Studio) for game feel theory, shader patterns, and asset pipeline research.

---

Created by David-GD13. Targeted for Unity 6.2+ and modern AI workflows.