everything-claude-unity
Health Warn
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 5 GitHub stars
Code Pass
- Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
- Permissions — No dangerous permissions requested
This toolkit provides a plug-and-play MCP server and a collection of AI agents, slash commands, and safety hooks designed to give Claude deep expertise in Unity game development. It allows developers to manage Unity scenes, write C# code, and trigger mobile builds using natural language.
Security Assessment
Overall Risk: Low. The automated code scan of 12 files found no dangerous patterns, no hardcoded secrets, and the tool does not request any dangerous permissions. However, because it is written in Shell and explicitly designed to orchestrate tasks like executing builds and managing project files, it inherently runs local commands. Users should always review installation scripts like `install.sh` before executing them to ensure they understand exactly what is being dropped into their development environment.
Quality Assessment
The project is highly maintained, with its last push occurring today. It is openly licensed under the standard MIT license, making it free for personal and commercial use. The documentation is excellent and highly detailed, indicating a professional and well-thought-out project. The only notable drawback is its extremely low community visibility. With only 5 GitHub stars, it is either very new or has a small user base, meaning it has not yet undergone the extensive peer review that comes with widespread community adoption.
Verdict
Safe to use, but review the shell installation script first given the low community adoption and early stage of the project.
The ultimate Claude Code toolkit for Unity — 20 AI agents, 22 slash commands, 41 skills, and safety hooks for AI-assisted Unity game development. Drop-in, production-ready.
everything-claude-unity
English | 中文 | 日本語 | 한국어 | Español | Português | Deutsch | Français | Türkçe
The ultimate Claude Code toolkit for Unity game development.
A production-ready, plug-and-play system that gives Claude Code deep Unity expertise — from writing performant C# to building scenes, profiling performance, and triggering iOS/Android builds — all through natural language.
Built for solo indie mobile game developers. Drop it into any Unity project and it works.
Why this works?
- Drop-in, zero-config — one
install.shand Claude Code knows Unity inside out - 20 specialized agents across haiku/sonnet/opus tiers with automatic complexity routing
- 22 safety hooks that prevent scene/meta corruption, silent data loss, and GC spikes before they happen
- MVS architecture built-in — VContainer + MessagePipe + UniTask patterns enforced by rules and templates
- MCP-native — scene building, profiling, and iOS/Android builds triggered from chat
- 60+ automated tests and reproducible benchmarks so the toolkit itself stays honest
What You Get
| Component | Count | Purpose |
|---|---|---|
| Agents | 20 | Specialized sub-agents across 3 model tiers (haiku/sonnet/opus) |
| Commands | 22 | Slash commands like /unity-workflow, /unity-ralph, /unity-team, /unity-skillify |
| Skills | 41 | Knowledge modules for Unity systems, gameplay patterns, and mobile genres |
| Hooks | 22 | Safety net, quality gates, notifications, session persistence, auto-learning |
| Rules | 5 | C# coding standards, performance rules, MVS architecture patterns |
| Scripts | 8 | Validation tools for meta files, code quality, serialization, architecture |
| Templates | 10 | C# templates for MVS pattern (Model, View, System, LifetimeScope, Message) |
| Tests | 60+ | Automated test suite for hooks, state, cross-validation, templates, skills |
| Benchmarks | 4 | Quality evaluation scenarios with structural comparison |
Highlights
/unity-workflow — Full Development Pipeline
A structured 4-phase pipeline for any feature: Clarify requirements, Plan the implementation, Execute with specialized agents, Verify with automated review + fix loop.
/unity-workflow "add a combo scoring system with multipliers and visual feedback"
/unity-prototype — One Prompt to Playable
Describe a mechanic, and Claude writes the C# scripts, builds the scene via MCP, sets up physics layers, configures the camera, and verifies it compiles.
/unity-prototype "2D platformer with wall jumping and dash"
/unity-ralph — Relentless Verify-Fix Loop
Runs the verify-fix loop persistently — refuses to stop until the project is clean or hits the safety limit. Up to 30 effective verification passes with stall detection.
/unity-ralph --max-iterations 10
/unity-team — Parallel Agent Orchestration
Spawn multiple agents simultaneously — coder + tester + reviewer working in parallel for faster development.
/unity-team --build "add health system with damage and healing"
/unity-team --security "audit the save system"
/unity-team --build --quick "add a basic score counter"
What's New in v1.3.0
- Smart Model Routing — complexity heuristics auto-select haiku/sonnet/opus agents based on task signals
- Persistent State — session data survives reboots in
.claude/state/with structured schema - Plan Critic —
unity-criticagent challenges your plans before execution (built into/unity-workflow) - Security Auditing —
unity-security-reviewercatches PlayerPrefs secrets, hardcoded keys, insecure saves - Benchmarking — measure agent quality with reproducible evaluation scenarios
- Skillify —
/unity-skillifyauto-generates skills from your accumulated session learnings - Multi-channel Notifications — Discord, Slack, OS-native with event filtering and rate limiting
- Hardened CI — cross-validation, tool restriction checks, template and skill quality gates
Verify-Fix Loop
The unity-verifier agent automatically reviews your code changes, fixes safe issues (missing [FormerlySerializedAs], uncached GetComponent, ?. on Unity objects), and re-verifies — up to 3 iterations until clean. Built into /unity-workflow and available as an optional step in /unity-feature and /unity-prototype.
Hook Profiles
Hooks are organized into three profiles. Set UNITY_HOOK_PROFILE to control which hooks run:
| Profile | What's Active | Best For |
|---|---|---|
minimal |
Safety hooks only (block scene/meta corruption, editor guards, pre-compact) | CI pipelines, experienced developers |
standard |
Safety + quality warnings + session persistence + stop validation (default) | Daily development |
strict |
Everything: GateGuard, cost tracking, auto-learning, build analysis | New projects, learning, auditing |
UNITY_HOOK_PROFILE=strict # Enable all hooks including GateGuard
UNITY_HOOK_PROFILE=minimal # Only critical safety hooks
DISABLE_UNITY_HOOKS=1 # Bypass all hooks entirely
UNITY_HOOK_MODE=warn # Downgrade blocks to warnings
DISABLE_HOOK_BLOCK_SCENE_EDIT=1 # Disable a specific hook
Quick Start
Prerequisites
- Claude Code installed
- Unity 2021.3 LTS or later
- unity-mcp (optional but recommended for full pipeline)
Install
# From your Unity project root:
git clone https://github.com/XeldarAlz/everything-claude-unity.git /tmp/ecu
/tmp/ecu/install.sh --project-dir .
rm -rf /tmp/ecu
Or manually:
git clone https://github.com/XeldarAlz/everything-claude-unity.git
cp -r everything-claude-unity/.claude your-unity-project/.claude
chmod +x your-unity-project/.claude/hooks/*.sh
Upgrade / Uninstall
# Upgrade to latest (preserves your customizations, creates backup)
./upgrade.sh --project-dir .
# Preview changes before upgrading
./upgrade.sh --project-dir . --dry-run
# Clean removal (with backup)
./uninstall.sh --project-dir .
Setup Unity MCP (Recommended)
The MCP bridge gives Claude direct control over the Unity Editor — scene building, profiling, builds, and more.
- In Unity:
Window > Package Manager > + > Add package from git URL - Paste:
https://github.com/CoplayDev/unity-mcp.git?path=/MCPForUnity#main - Open
Window > MCP for Unityand click Start Server - Claude Code auto-connects via
.claude/settings.json
First Run
cd your-unity-project
claude
# Verify installation:
/unity-doctor # Check MCP, hooks, project structure
# Start working:
/unity-audit # Full project health check
/unity-workflow # Full pipeline: clarify → plan → execute → verify
/unity-prototype # Rapid prototype a game mechanic
Agents
Code Agents
| Agent | Model | What It Does |
|---|---|---|
unity-coder |
opus | Implements features with Unity subsystem awareness, loads relevant skills |
unity-coder-lite |
sonnet | Lightweight variant for simple additions (fields, methods, straightforward components) |
unity-fixer |
opus | Diagnoses bugs using Unity-specific patterns (missing refs, execution order, coroutine lifecycle) |
unity-fixer-lite |
sonnet | Quick fixes for obvious issues (typos, missing imports, simple errors) |
unity-reviewer |
sonnet | Code review checking serialization safety, GC in hot paths, lifecycle ordering |
unity-shader-dev |
opus | HLSL/ShaderGraph development optimized for mobile GPUs, live testing via MCP |
Orchestration Agents
| Agent | Model | What It Does |
|---|---|---|
unity-verifier |
opus | Verify-fix loop: reviews changes, auto-fixes safe issues, re-verifies (max 3 iterations) |
unity-prototyper |
opus | End-to-end prototyping: writes code + builds scene + physics + camera |
MCP-Powered Agents
| Agent | Model | What It Does | Key MCP Tools |
|---|---|---|---|
unity-scene-builder |
opus | Builds scenes from descriptions | manage_scene, batch_execute |
unity-test-runner |
sonnet | Writes + runs tests, reports results | run_tests, read_console |
unity-build-runner |
sonnet | Configures and triggers builds | manage_build, manage_packages |
unity-optimizer |
opus | Profiles and fixes performance issues | manage_profiler, manage_graphics |
Hybrid Agents
| Agent | Model | What It Does |
|---|---|---|
unity-ui-builder |
opus | Builds UI screens with code + visual setup via MCP |
unity-network-dev |
opus | Implements multiplayer with Netcode/Mirror/Photon/Fish-Net |
unity-migrator |
sonnet | Unity version and render pipeline migration |
Commands support --quick (routes to sonnet lite agent) and --thorough (routes to opus) flags. See docs/MODEL-ROUTING.md for the full routing table.
Commands
Full Pipeline
/unity-workflow <description> Clarify → Plan → Execute → Verify (the recommended workflow)
Daily Workflow
/unity-feature <description> Plan + implement a feature (--quick for simple tasks)
/unity-fix <bug or error> Diagnose and fix a bug (--quick for obvious fixes)
/unity-prototype <mechanic> One prompt to playable prototype
/unity-scene <description> Build a scene via MCP
/unity-shader <description> Create shaders with live preview
/unity-ui <screen description> Build UI with visual setup
/unity-network <framework> Set up multiplayer
Quality Gates
/unity-review [scope] Code review (--thorough for deep analysis)
/unity-optimize Profile via MCP + fix bottlenecks
/unity-test Write + run tests via MCP
/unity-audit Full project health check
/unity-profile Deep profiling session
Orchestration
/unity-ralph [options] Persistent verify-fix loop (refuses to stop until clean)
/unity-team <--preset|--custom> Parallel agents (coder + tester + reviewer simultaneously)
/unity-interview <topic> Deep Socratic requirements interview before coding
/unity-learn [subcommand] Session analytics: review, extract patterns, draft skills
Project Lifecycle
/unity-init Scan project + generate CLAUDE.md
/unity-build Configure + trigger builds
/unity-migrate Plan version/pipeline migration
/unity-doctor Diagnostic health check (MCP, hooks, project structure)
Hooks
22 hooks across 5 lifecycle events, organized by profile level.
Blocking Hooks — PreToolUse (minimal profile)
| Hook | What It Prevents |
|---|---|
block-scene-edit |
Direct text editing of .unity/.prefab YAML (corrupts references) |
block-meta-edit |
Editing .meta files (breaks asset GUIDs) |
block-projectsettings |
Staging ProjectSettings/ via git (use MCP instead) |
guard-editor-runtime |
UnityEditor namespace in runtime code without #if UNITY_EDITOR |
guard-project-config |
Weakening code quality rules (.editorconfig, analyzer settings, .csproj NoWarn) |
GateGuard — PreToolUse (strict profile)
| Hook | What It Does |
|---|---|
gateguard |
Blocks Edit/Write on C# files until the agent has Read them first. Prevents hallucinated changes. For MVS files, suggests reading Model/System counterparts. |
Quality Hooks — PostToolUse (standard profile)
| Hook | What It Catches |
|---|---|
warn-serialization |
Field renamed without [FormerlySerializedAs] (silent data loss) |
warn-filename |
C# file name doesn't match class name (script won't attach) |
warn-platform-defines |
#if UNITY_ANDROID without #else fallback |
quality-gate |
GetComponent in Update, LINQ in gameplay, ?. on Unity objects, uncached Camera.main, SendMessage |
validate-commit |
Missing .meta files, code quality issues on commit |
suggest-verify |
Suggests /unity-review after 5+ C# files modified |
build-analyze |
Post-build: shader variant counts, size, stripping issues, deprecated APIs |
Tracking Hooks — PostToolUse (standard/strict profile)
| Hook | What It Records |
|---|---|
track-edits |
Files modified during session (standard) |
track-reads |
Files read during session — feeds GateGuard (strict) |
cost-tracker |
Every tool call with timestamp for session metrics (strict) |
Session Hooks — SessionStart / Stop
| Hook | Lifecycle | What It Does |
|---|---|---|
session-restore |
SessionStart | Restores prior branch, workflow phase, modified files list |
session-save |
Stop | Saves session state for next conversation (branch, edits, duration) |
stop-validate |
Stop | Runs full-file validation on all C# files modified during session |
auto-learn |
Stop | Captures session patterns (MVS breakdown, tool usage, category) to learnings log |
notify |
Stop | Sends webhook notification (Discord/Slack) when session exceeds minimum duration |
Advisory Hooks — PreCompact
| Hook | What It Does |
|---|---|
pre-compact |
Saves git state before context compaction |
All hooks support kill switches via environment variables. See Hook Profiles above.
MVS Architecture Templates
Templates for the Model-View-System pattern with VContainer, MessagePipe, and UniTask:
| Template | Purpose |
|---|---|
Model.cs.template |
Pure C# data class with ReactiveProperty<T> — no Unity dependencies |
System.cs.template |
Plain C# class with VContainer constructor injection, IDisposable |
View.cs.template |
MonoBehaviour observing Model via Subscribe(), method injection |
LifetimeScope.cs.template |
VContainer composition root with Model/System/View/MessagePipe registration |
Message.cs.template |
readonly struct for MessagePipe — zero heap allocation |
Plus the original templates: MonoBehaviour.cs, ScriptableObject.cs, EditModeTest.cs, PlayModeTest.cs, AssemblyDefinition.asmdef.
Skills
Always-On Core (8)
- serialization-safety —
[FormerlySerializedAs],[SerializeField], Unity null checks - scriptable-objects — SO event channels, variable refs, runtime sets, factory pattern
- event-systems — C# events, UnityEvent, SO channels, zero-alloc EventBus
- object-pooling —
ObjectPool<T>, warm-up, return-to-pool lifecycle - assembly-definitions — When to split, reference rules, Editor/Runtime separation
- unity-mcp-patterns — How to use MCP tools effectively (
batch_execute,read_console) - learner — Post-debugging knowledge extraction with quality gates and confidence scoring
- hud-statusline — Claude Code statusline integration showing workflow phase and session metrics
Unity Systems (10)
URP pipeline, Input System, Addressables, Cinemachine, Animation, Audio, Physics, NavMesh, UI Toolkit, ShaderGraph
Gameplay Patterns (6)
Character controller (2D/3D), inventory system, dialogue system, save system, state machine, procedural generation
Genre Blueprints (8) — Mobile-Focused
Hyper-casual, Match-3, Idle/Clicker, Endless Runner, Puzzle, RPG, 2D Platformer, Top-down
Third-Party (5)
DOTween, UniTask, VContainer, TextMeshPro, Odin Inspector
Platform (1)
Mobile optimization (iOS + Android) — touch input, safe areas, ASTC textures, thermal throttling, battery management
Coding Rules
The toolkit enforces Unity best practices through 5 always-loaded rule files:
- csharp-unity —
[SerializeField] privatewithm_prefix, sealed by default, explicit types - performance — Zero-alloc Update, cache GetComponent, pool objects, no LINQ in gameplay
- serialization —
[FormerlySerializedAs]on renames,obj == nullnotobj?. - architecture — MVS pattern, VContainer for DI, MessagePipe for events, UniTask for async
- unity-specifics — Editor/Runtime separation, threading, coroutine lifecycle,
?.danger
Validation Scripts
Run these to check project health:
./scripts/validate-meta-integrity.sh --all # Missing/orphaned .meta files, duplicate GUIDs
./scripts/validate-code-quality.sh # Performance pitfalls in C# code
./scripts/validate-asmdefs.sh # Circular assembly definition dependencies
./scripts/detect-missing-refs.sh # Broken references in scenes/prefabs
./scripts/analyze-build-size.sh # Build size analysis from Editor.log
./scripts/validate-serialization.sh # Field renames missing FormerlySerializedAs
./scripts/validate-architecture.sh # MVS pattern compliance checks
./scripts/generate-claude-md.sh > CLAUDE.md # Auto-generate project CLAUDE.md
Example CLAUDE.md Files
Pre-built configurations for mobile game types:
examples/CLAUDE.md.hyper-casual— One-tap controls, minimal visuals, ad monetizationexamples/CLAUDE.md.match3— Grid system, cascades, special tiles, lives/energyexamples/CLAUDE.md.idle-clicker— Big numbers, offline progress, prestige systemexamples/CLAUDE.md.mobile-casual— Touch input, small build, ad integrationexamples/CLAUDE.md.2d-platformer— Tilemap, virtual joystick, mobile-optimizedexamples/CLAUDE.md.rpg— Stats, inventory, dialogue, touch controls
Architecture
Workflow Pipeline
/unity-workflow "add combo scoring"
│
├─ Phase 1: Clarify ── interview about requirements, constraints, platform
├─ Phase 2: Plan ── scan project, choose agents, present implementation plan
├─ Phase 3: Execute ── route to unity-coder / unity-prototyper / unity-ui-builder
└─ Phase 4: Verify ── unity-verifier runs review → auto-fix → re-verify loop
Agent Interaction
User prompt
│
▼
Command (orchestrates the workflow)
│
├──▶ Code Agent (writes C# scripts, loads relevant skills)
│ │
│ └──▶ MCP Tools (creates GameObjects, configures components)
│
├──▶ Verifier Agent (reviews changes, auto-fixes, re-verifies)
│
├──▶ Test Agent (writes + runs tests via MCP)
│
└──▶ Optimizer Agent (profiles via MCP, fixes bottlenecks)
Hook Safety Net
Claude attempts to edit PlayerView.cs
│
├──▶ _lib.sh: check profile level, kill switches
├──▶ PreToolUse: guard-editor-runtime.sh — UnityEditor guard
├──▶ PreToolUse: gateguard.sh — was this file Read first? [strict]
│ suggest reading PlayerModel.cs too
│
├──▶ [Edit happens]
│
├──▶ PostToolUse: warn-serialization.sh — field rename check
│ quality-gate.sh — GetComponent in Update? LINQ? ?.?
│ track-edits.sh — record for session metrics
│
└──▶ [Session ends]
├──▶ stop-validate.sh — full-file checks on all modified C#
├──▶ session-save.sh — persist state for next conversation
└──▶ auto-learn.sh — log session patterns
Session Lifecycle
SessionStart
└──▶ session-restore.sh — load prior state (branch, phase, files)
[... work happens, tracked by hooks ...]
Stop
├──▶ stop-validate.sh — batch validation on all modified files
├──▶ session-save.sh — save state to /tmp/unity-claude-hooks/
└──▶ auto-learn.sh — append session metrics to learnings.jsonl
Documentation
| Guide | Purpose |
|---|---|
| Getting Started | Installation, first run, troubleshooting |
| Architecture | Design philosophy, component overview, hook system, workflow pipeline |
| Agent Guide | All 15 agents, when to use each, customization |
| Model Routing | Agent model assignments, --quick/--thorough flags, cost trade-offs |
| MCP Setup | unity-mcp installation, verification, troubleshooting |
Contributing
See CONTRIBUTING.md for guidelines.
Key areas where contributions are welcome:
- New mobile genre skills (tower defense, racing, card/gacha, simulation)
- New system skills (ProBuilder, Spline, 2D Animation)
- Mobile platform skills (ARKit/ARCore, notifications, deep links)
- Networking framework skills for mobile (FishNet, Dark Rift)
- Bug reports and hook improvements
License
MIT License. See LICENSE for details.
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found