About Unity's plugin
Learn about Unity's plugin for third-party agents, its bundled skills, and best practices for working with your AI agent.
읽는 시간 2분최근 업데이트: 18시간 전
Unity's plugin for Claude Code, Codex, and Grok adds a set of Unity skills to your AI agent. These skills describe common tasks around Unity feature sets, such as 2D and tilemaps, UI Toolkit, localization, multiplayer, live services, LevelPlay, and In-App Purchasing. Generally, to use these skills, ask your agent in plain language or invoke a skill directly.
참고
This isn't a Unity Asset Store plugin or package and you can't install it through the Package Manager.Prerequisites
Unity's plugin works with Unity 6.0 or later.
Install the plugin
For detailed instructions on how to install Unity's plugin for your agents, refer to the following:
Use the plugin
After you install the plugin, your agent uses skills to load relevant context to accomplish tasks you give it in your Unity project. For example, if you ask an agent to do the following, the plugin loads skills that correspond to these tasks:
Add in-app purchases so players can buy a coin packI want to build a settings screenMy pixel art looks blurry and jitters when the camera movesShow rewarded video ads so players can earn coinsCreate a hexagonal tile palette for my levelChinese characters show up as empty boxes in my TextMeshPro labelsReview my ScriptableRendererFeature for Render Graph problems
In Claude Code, Codex, and Grok, you can also invoke skills explicitly in context slash menus instead of relying on your agent to choose one to invoke on your behalf.
For a detailed list of skills, refer to the unity-agent-plugin GitHub repository.
Best practices
The following best practices can help you work effectively with Unity's plugin:
- Name the UI framework in UI-related requests: Unity includes several UI systems. Unity recommends UI Toolkit for new projects, so specify UI Toolkit when you start one. For existing projects that already use Unity UI (uGUI), specify uGUI. Either way, naming the framework at the beginning of your queries is more efficient than letting the agent guess.
- Ask the agent to read the skill's reference files if a result looks wrong: Each skill stores detailed guidance in a folder. If the agent skips that step, the output is often incorrect. Ask it to read the reference files and try again.
references/ - Expect the agent to ask clarifying questions: With the plugin installed, the agent asks questions before making changes rather than assuming. This is helpful in interactive sessions, but can block automated pipelines. If you're running the agent in a pipeline, write prompts that include enough detail that no follow-up questions are needed.
- Name the skill explicitly if it doesn't trigger automatically: Some agents might not make the link between the skill you want it to use and the natural phrasing you use in your instruction. Explicitly name the skill in your request as a reliable fallback regardless of which agent you're using.