Unity-Skills stands out by providing a REST API-based automation engine that lets AI agents directly control Unity Editor scenes through a rich library of skills. This repo tackles the challenge of safely automating complex Unity workflows by aligning server-side permission models with AI agent capabilities and ensuring transactional integrity with automatic rollback on failures. It’s a practical, enterprise-grade approach to integrating AI control into a traditionally manual, stateful environment.
What Unity-Skills does and how it’s built
At its core, Unity-Skills is an AI-driven automation framework for the Unity Editor. It exposes 750 distinct skills organized into 51 functional modules, accessible via a REST API. These skills empower AI agents—such as Claude Code, Antigravity, Codex, and Cursor—to perform direct scene manipulations and editor operations programmatically.
The architecture is client-server: the Unity Editor runs a server component that listens for REST API calls representing skill invocations. This server manages multi-instance concurrency through automatic port discovery and maintains ultra-stable long-lived HTTP connections with configurable timeouts (defaulting to 15 minutes). It also handles Unity’s Domain Reload events gracefully, automatically restoring state and connections.
A key structural feature is the three-tier permission system on the server side, with modes named Approval, Auto, and Bypass. These modes correspond to Claude Code’s permission framework and govern how AI-initiated commands are authorized before execution, adding a layer of enterprise-grade safety. The system enforces transactional atomicity: if a multi-step operation fails partway, it automatically rolls back all changes to keep the editor state consistent and avoid partial corruption.
Under the hood, this project is a deep refactoring of the earlier unity-mcp project, targeting Unity 2022.3 or later. It supports integration with multiple AI IDEs and terminals, providing a one-click installer for each supported environment.
What sets Unity-Skills apart: permission modes and transactional safety
What distinguishes Unity-Skills is how it blends AI agent control with strict safety guarantees suitable for production workflows.
The server-side permission design is aligned explicitly with Claude Code’s permission modes, breaking down into three tiers:
- Approval: Commands require explicit human approval before execution.
- Auto: Commands are automatically approved under predefined safe conditions.
- Bypass: Commands execute without approval, used in trusted or test environments.
This design mitigates risks of unintended or harmful AI commands altering critical project assets or scenes, a common concern when delegating control to AI.
The transactional atomicity mechanism is another standout feature. Unity-Skills wraps multi-step skill executions in transactions, ensuring that if any part fails or triggers an error, the system rolls back all changes made during that operation. This rollback prevents the editor from ending up in a partially modified or inconsistent state, which can be notoriously difficult to recover from in Unity’s complex scene environment.
Supporting multiple AI IDEs natively—Claude Code, Antigravity, Codex, and Cursor—with tailored skill invocation methods demonstrates a focus on developer experience. For example, Codex supports both explicit skill invocation via $skill commands and implicit intent recognition, while Cursor auto-discovers skill directories and provides UI integration for skill management.
The codebase is designed to handle real-world Unity Editor edge cases, such as recompilation triggers, asset reimports, and domain reload events that temporarily disrupt REST availability. The system’s ability to recover automatically from these interruptions adds to its robustness.
Quick start with Unity-Skills
The repository provides a streamlined installation and setup process optimized for the four supported AI terminals and IDEs.
1. Install the Unity plugin
Add the UnitySkills plugin via the Unity Package Manager using the Git URL. You can choose the stable, beta, or specific version:
# Stable version (main branch)
https://github.com/Besty0728/Unity-Skills.git?path=/SkillsForUnity
# Beta version (beta branch)
https://github.com/Besty0728/Unity-Skills.git?path=/SkillsForUnity#beta
# Specific version example (v1.6.0)
https://github.com/Besty0728/Unity-Skills.git?path=/SkillsForUnity#v1.6.0
All versions are available on the GitHub Releases page.
2. Start the UnitySkills server
Inside Unity, start the server by navigating to the menu:
Window > UnitySkills > Start Server
Note that certain editor operations like script recompilation, package changes, or asset reimports will temporarily make the REST API unavailable. The server handles these domain reloads transparently, so just wait a moment and retry.
3. Configure AI skills with one click
The plugin provides a skill installer UI:
Window > UnitySkills > Skill Installer
From here, select the AI terminal or IDE you want to configure skills for (Antigravity, Claude Code, Codex, Cursor). This sets up the corresponding skill directories and configuration automatically.
This tight integration simplifies the developer experience drastically, enabling AI agents to start using the skills right away.
verdict
Unity-Skills targets developers and teams who want to integrate AI agents deeply into their Unity Editor workflows with safety and robustness. Its three-tier permission system and transactional atomicity offer enterprise-grade safeguards rarely seen in AI-driven automation for game engines.
The tradeoff is complexity and a dependency on Unity 2022.3 or later, which may limit adoption in projects on older versions. Also, while the permission system mitigates risks, the overall approach requires careful configuration and understanding of AI behavior to avoid unintended effects.
If you’re experimenting with AI automation in Unity or building tools that require AI agents to operate with fine-grained control and safety, Unity-Skills provides a solid foundation. It’s especially relevant for teams valuing production stability and multi-agent support.
The codebase and architecture reflect a mature approach to integrating AI with a traditionally manual and stateful environment, balancing flexibility with safety. Worth exploring for anyone working at the intersection of game development and AI-driven automation.
Related Articles
- Softaworks Agent Toolkit: A modular plugin marketplace for AI coding agents — Softaworks Agent Toolkit offers 40+ modular AI skills and plugins for coding agents like Claude Code, enabling composabl
- dotnet/skills: a curated marketplace of AI agent skills for the .NET ecosystem — dotnet/skills offers 11 AI coding assistant plugins covering the full .NET stack, following the agentskills.io standard
- SkillHub Desktop: Unifying AI coding assistant skills with a Rust + React Tauri app — SkillHub Desktop tackles fragmentation in AI coding assistants by unifying skill management across tools using a Rust ba
- Inside Claude Code: A detailed reconstruction of Anthropic’s AI safety and architecture — A deep dive into Claude Code’s 512K lines of TypeScript reveals a layered YOLO safety classifier, multi-agent IPC, and t
- SkillUI: Automating design system extraction into Claude Code skills — SkillUI reverse-engineers design systems from URLs or repos into Claude Code skill packages using static and visual anal
→ GitHub Repo: Besty0728/Unity-Skills ⭐ 1,073 · C#