← All posts

gameLightning Games

Unity's Official AI Agent Plugin: A Practical Guide for Mobile Teams

Unity's Official AI Agent Plugin: A Practical Guide for Mobile Teams

Ask a general-purpose coding agent to add in-app purchases to a Unity game and you often get code written for an API version Unity retired years ago. Since September, Unity has been shipping its own answer to that problem: an official plugin that loads Unity-written "skills" into Claude Code, Codex, Grok Build and other agents. This guide explains what is in it, how to install it, where it actually saves time on a mobile project, and where you still need to check every line yourself.

▶ Watch the video version (about 4 minutes):

What Unity shipped, and why

At Unite Seoul, Unity launched the Unity CLI so that coding agents could install editors, open projects and run builds from a terminal (we covered that in our Unity CLI and Pipeline 0.9 guide). The plugin is the next layer. The CLI lets an agent operate Unity; the plugin tells the agent how Unity wants each job done.

Unity's argument is simple. A general agent learns Unity from forum posts and old tutorials, so it produces code that compiles but targets the wrong API or the wrong pipeline. The Decoder summed it up as a plugin meant "to stop AI agents from using outdated tutorials". In Unity's words, each skill teaches the agent "where a Unity project keeps its assets, how an atlas is supposed to be built, what a URP renderer feature has to satisfy before it will run at all" (Unity blog, Claude Code plugin).

The timeline so far:

  • September 2026: the plugin launched for Claude Code with 29 skills, followed by Codex with 31.
  • Later: Grok Build became the third agent. Unity says the skills are identical across agents and maintained from a single repository.
  • Now: the GitHub repository lists Claude Code, Codex, Grok, Google Antigravity and GitHub Copilot CLI, and its skills folder held 34 skills after a batch published on November 6, 2026.

It is still labelled beta. Unity's documentation says features "might change", and the plugin works with Unity 6.0 or later. It is not an Asset Store package, so you will not find it in the Package Manager.

The business context matters too. On its Q3 2026 earnings call, Unity said its daily active users rose 25% since early August, after it opened the editor to coding agents. Unity is clearly betting on this path, so expect the skill list to keep growing.

What is inside: the skills that matter for mobile

A skill is a folder with a SKILL.md workflow plus references/ files and sometimes ready-made C# under resources/. The agent loads the relevant one when your request matches, or you call it by name. Here is how the current set maps to a typical mobile game:

AreaSkills (folder names)Mobile use
Monetizationimplement-in-app-purchases, levelplay-unity-integrationIAP v5, rewarded and interstitial ads
2D and UImanage-sprite-atlas, 2d-pixel-perfect, ui-uitk, ui-ugui, sprite-editorAtlases, crisp pixel art, menus
Performanceoptimize-text-mesh-pro, optimize-audio, urp-postprocessingMemory, load times, GPU cost
Renderingmigrate-birp-to-urp, validate-urp-render-graph-renderer-feature, shader-graph-create-custom-nodePipeline migration, custom passes
Live gamelocalize-unity-project, setup-multiplayer-services, build-live-gameLocalization, online features
Toolingunity-cli, unity-pipeline, new-unity-project, unity-package-managementSetup, packages, builds

Two details stand out when you read the skill files. First, they are strict. The IAP skill tells the agent to read a pre-check reference before touching anything, to "establish a baseline of the current behavior" and to "create a restore point" before changes (implement-in-app-purchases). Second, they are opinionated about Unity's own products. The LevelPlay skill covers only LevelPlay mediation and tells the agent not to characterise competing SDKs (levelplay-unity-integration). If your game uses AppLovin MAX or AdMob mediation, this skill will not help.

Installing it in five minutes

You need Unity 6.0+ and one supported agent. The commands below come from the repository README and Unity's Claude Code page.

Claude Code (from a terminal):

claude plugin marketplace add Unity-Technologies/unity-agent-plugin
claude plugin install unity@unity-agent-plugin

Inside a running session you can type the same steps as /plugin marketplace add ... and /plugin install .... Claude Code asks for a scope: user (all your projects), project (shared with everyone in the repository) or local (only you, only this repository). For a studio, project scope is the useful one, because every developer gets the same skills.

Codex:

codex plugin marketplace add Unity-Technologies/unity-agent-plugin
codex plugin add unity@unity-agent-plugin

Grok Build: grok plugin install Unity-Technologies/unity-agent-plugin --trust. GitHub Copilot CLI uses copilot plugin marketplace add and copilot plugin install with the same names.

To check it worked, type /unity: in Claude Code to see the skill list, or run codex plugin list and look for unity@unity-agent-plugin. One catch: in Claude Code, auto-update is off by default for third-party marketplaces, so run /plugin marketplace update unity-agent-plugin now and then, or new skills will not arrive.

Where it saves real time on a mobile project

From the skill contents, these are the jobs where a small Unity mobile team is most likely to see a difference:

  1. IAP upgrades. Moving from IAP v4 to v5, or off RevenueCat or native billing, is tedious and easy to get subtly wrong. The skill maps the old calls to the new ones, checks subscription logic (for example, comparing IsSubscribed() with Result.True instead of casting to bool) and ends with a report of follow-up work in App Store Connect and Google Play Console.
  2. Ad SDK build failures. The LevelPlay skill has a dedicated scenario for Android builds that break because of Maven Central dependencies, plus a migration path from older IronSource APIs.
  3. Sprite atlases. The atlas skill ships working C# for Sprite Atlas V2 and warns that writing atlas code from memory is "the single most common cause of an atlas that imports cleanly and then does nothing at runtime".
  4. Text and fonts. TextMeshPro font stacks, fallback atlases and CJK characters are a common memory problem in games localised for Asia. A prompt like "Chinese characters show up as empty boxes" maps directly to a skill.
  5. Built-In to URP. With the Built-In Render Pipeline deprecated in Unity 6.5, the migrate-birp-to-urp skill covers pink materials, lighting, lightmaps and post-migration bugs. It pairs well with our URP migration guide.

Limits and risks to plan for

Unity is fairly honest about the limits. Its Claude Code documentation says that for simple uGUI work the plugin "makes little difference", and that on a strong model it mainly adds correctness rather than new abilities. Treat it as a way to get the current API, not as a way to skip design.

Plan for these points:

  • It asks questions. With the plugin, the agent asks before it changes things. That is good at your desk but can stall an automated pipeline. Unity's best practices say to write prompts detailed enough that no follow-up questions are needed.
  • Skills do not always trigger. If the agent ignores a skill, name it in the prompt. If the output looks wrong, ask the agent to read the skill's references/ files and try again.
  • Vendor bias. Monetization skills cover Unity's products. That is fine if you already use them, but it is not neutral advice.
  • Beta and licence. The plugin is under the Unity Companion License and may change. Pin a version for release branches and read the diff when you update.
  • Your code still needs review. A correct API call can still be the wrong design for your game. Keep version control, small commits and device testing.

A safe first week for a small studio

If you want to try it without risking a release:

  1. Install it with project scope on a branch, not on main.
  2. Pick one boring, well-defined job: an atlas cleanup, an IAP v5 upgrade in a prototype, or a TextMeshPro fallback setup.
  3. Name the UI framework in every UI prompt ("using uGUI" or "using UI Toolkit"); Unity recommends this.
  4. Ask the agent for its final report: files changed, how it verified the change, and follow-up steps.
  5. Build to a real Android phone and an iPhone, and run a store sandbox purchase before you merge.
  6. Compare the time spent against doing it by hand. Keep the plugin for the jobs where it clearly won.

Our take

At Lightning Games we build Unity mobile games for English-first markets, and most of our week is not new gameplay. It is SDK upgrades, store rules, atlases, fonts and build fixes. That is exactly where outdated agent answers hurt most and where Unity's own guidance helps. Our plan is to start with IAP and atlas work in a branch, with every change reviewed and tested on devices. We would not let it touch ad mediation in a live game without a full test pass, and we would not expect it to replace a developer who understands the project. Used that way, it is a useful upgrade to the coding agents many teams already run.

Sources

Want to see what we build with Unity? Visit Lightning Games.


Cover photo: Macro laptop coding (Unsplash).jpg) by Marc Mueller, CC0 1.0, via Wikimedia Commons (cropped).

Facebook