Openclaw Unity Skill

Unity Plugin Skill

Control Unity Editor through ~100 built-in tools. Works in both Editor and Play mode.

Safety and permissions

What this skill can change: anything the Unity Editor can — GameObjects and components, scenes, prefabs, materials, shaders, textures and ScriptableObjects, files under Assets/, installed packages, Play mode, and arbitrary C# run inside the Editor process (script.execute, plus reflection-based calls).

What runs without asking: read-only tools only — get*, list, find, script.read, debug.hierarchy, debug.screenshot, console.getLogs. Their behaviour is unchanged.

What is gated at runtime: every project-changing tool — create, delete, destroy, save, set*, component.add/remove, asset.* writes, prefab.*, material.*, texture.*, editor.refresh/recompile, Play-mode control, input simulation, test.run, script.execute — and any batch.execute that carries one of them. The gateway extension refuses these by default.

Custom tools are gated by default (1.8.0): a name that is not in this skill's built-in catalogue is a project-registered custom tool — the Editor add-on's OpenClawCustomTools.Register lets a project publish a tool under any name — so its verb proves nothing. mygame.getScore is treated exactly like asset.delete: refused without the operator opt-in and confirm: true. Before 1.8.0 such a name passed ungated whenever it started with get/list/find, which is what made the gate too weak to approve. To exempt one you have read:

# exact names, comma separated, on the gateway process
OPENCLAW_UNITY_READONLY_CUSTOM_TOOLS="mygame.getScore,mygame.listEnemies" \
  openclaw gateway restart

script.execute, execute_code, execute_custom_tool, manage_tools and anything whose name contains a destructive verb can never be declared read-only: the allowlist ignores those entries. The in-package allowlist (READ_ONLY_CUSTOM_TOOLS in extension/index.ts) is empty on purpose.

How to enable them — both steps are required:

1. The operator starts the gateway with the opt-in environment variable. It is read from the gateway process, so the model cannot set it:

   OPENCLAW_EDITOR_ALLOW_DESTRUCTIVE=1 openclaw gateway restart
   # or, scoped to this skill only:
   OPENCLAW_UNITY_ALLOW_DESTRUCTIVE=1 openclaw gateway restart
   

2. The caller passes confirm: true on each project-changing call, after the user has approved that specific change:

   unity_execute: asset.delete {path: "Assets/Old/Item.prefab"}, confirm: true
   

Missing either one is a refusal with an explanation of what was blocked and how to allow it. Nothing reaches the Editor in the meantime.

Preview first: dryRun: true reports how a call is classified and what would be sent, and sends nothing:

unity_execute: gameobject.destroy {name: "Player"}, dryRun: true
→ {risk: "project-changing", requiresConfirmation: true, wouldRun: false, executed: false}

Check the current state: openclaw unity status prints whether project-changing tools are enabled, and GET /unity/status reports the same as destructiveOperations: "enabled" | "blocked".

Bridge authentication (1.8.0): the /unity/* endpoints this extension serves are the Editor add-on's half of the bridge. They now require a per-launch secret, and they answer loopback peers only.

1. When the gateway loads this extension it generates a 32-byte token and writes it to ~/.openclaw/unity-bridge.token with mode 0600 ($OPENCLAW_CONFIG_DIR/$OPENCLAW_HOME override the directory). The token is never logged — only its path is.
2. The Unity add-on reads that file (or OPENCLAW_BRIDGE_TOKEN, or the API Token field of OpenClawConfig) and sends it as X-OpenClaw-Bridge-Token on POST /unity/register. Without it: 401, and no session exists.
3. register answers with a per-session token. poll, heartbeat and result must carry it as X-OpenClaw-Session, and it must belong to the sessionId in the request. Otherwise: 401.
4. Every queued command carries a nonce. POST /unity/result is accepted only for a tool call that is actually in flight for that session and only with that call's nonce; anything else is dropped with 409 and never reaches the model. That is what stops a third local process from answering in the Editor's place.
5. Requests from a non-loopback peer are refused with 403, as is any request carrying an Origin or Referer header — no browser may drive the bridge. No CORS headers are sent (before 1.8.0 the bridge replied Access-Control-Allow-Origin: *, which let any open web page reach it).

Legacy behaviour is off by default. OPENCLAW_UNITY_ALLOW_LEGACY_UNAUTHENTICATED=1 (or OPENCLAW_EDITOR_ALLOW_LEGACY_UNAUTHENTICATED=1) restores the pre-1.8.0 unauthenticated bridge for an old add-on and logs a warning saying so; openclaw unity status and GET /unity/status report auth: "legacy-unauthenticated" while it is on. Update the add-on instead.

What the token does and does not buy: the file is 0600, so another user on the machine cannot read it. A process running as you can, and that is the same boundary that protects your SSH keys and the gateway's own config. The bridge is not a sandbox against your own account.

Session isolation (1.8.0): each connected Editor gets its own session id, session token, command queue and in-flight table. One session cannot poll another's queue or answer another's tool call. When more than one Editor is connected, unity_execute refuses to guess and asks for an explicit sessionId — it will not silently drive the wrong project. Session tokens are never reported by unity_sessions or GET /unity/status.

Scope of the gate: it lives in the gateway extension that ships with this skill (extension/index.ts), so it covers every call routed through the OpenClaw gateway — Telegram, Discord and the other channels. The Unity Editor add-on's local MCP bridge (port 27182, Mode 2 below) is a separate install from the plugin repository; from plugin 1.7.0 it has its own per-launch token (~/.openclaw/unity-mcp-bridge.token, mode 0600), refuses browser-originated requests and binds 127.0.0.1 only, but it is not covered by this confirmation gate: an MCP client that holds that token talks to the Editor directly.

Connection Modes

1. OpenClaw Gateway (Remote)

For Telegram, Discord, and other OpenClaw channels:

• Auto-connects when Unity opens
• Configure in: Window → OpenClaw Plugin → Settings

2. MCP Bridge (Local)

For Claude Code, Cursor, and local AI tools:

• Start: Window → OpenClaw Plugin → MCP Bridge → Start
• Default port: 27182
• Add to Claude Code: claude mcp add unity -- node <path>/MCP~/index.js

First-Time Setup

If unity_execute tool is not available, install the gateway extension:

# From skill directory
./scripts/install-extension.sh

# Restart gateway
openclaw gateway restart

The extension files are in extension/ directory.

What install-extension.sh Does

# 1. Copies extension files from skill to gateway
#    Source: <skill>/extension/
#    Destination: ~/.openclaw/extensions/unity/

# 2. Files installed:
#    - index.ts     # Extension entry point (HTTP handlers, tools)
#    - package.json # Extension metadata

# After installation, restart gateway to load the extension.

🔐 Security: Model Invocation Setting

When publishing to ClawHub, disableModelInvocation controls who may start the skill:

Recommendation: true

Reason: this skill can execute arbitrary C# inside the Editor (script.execute) and call Editor APIs by reflection. That is the whole point of it, and it is also why it should never start on its own inference. As of v1.7.0 this skill ships with disableModelInvocation: true — it runs only on explicit user request, and every project-changing call is gated on top of that (see Safety and permissions).

The full capability disclosure — arbitrary code execution, destructive operations, network surface — is in Security & Privacy Disclosure at the end of this document. Read it before enabling the skill.

Quick Reference

Core Tools

Common Workflows

1. Scene Inspection

unity_execute: debug.hierarchy {depth: 2}
unity_execute: scene.getActive

2. Find & Modify Objects

unity_execute: gameobject.find {name: "Player"}
unity_execute: component.get {name: "Player", componentType: "Transform"}
unity_execute: transform.setPosition {name: "Player", x: 0, y: 5, z: 0}

3. UI Testing

unity_execute: input.clickUI {name: "PlayButton"}
unity_execute: input.type {text: "TestUser", elementName: "UsernameInput"}
unity_execute: debug.screenshot

4. Play Mode Control

unity_execute: editor.play              # Enter Play mode
unity_execute: editor.stop              # Exit Play mode
unity_execute: editor.getState          # Check current state
unity_execute: editor.pause             # Pause
unity_execute: editor.unpause           # Resume

5. Material Creation

unity_execute: material.create {name: "RedMetal", color: "#FF0000", metallic: 0.8}
unity_execute: material.assign {gameObjectName: "Player", materialPath: "Assets/Materials/RedMetal.mat"}
unity_execute: material.modify {path: "Assets/Materials/RedMetal.mat", metallic: 1.0, emission: "#FF4444"}

6. Prefab Workflow

unity_execute: prefab.create {gameObjectName: "Player", path: "Assets/Prefabs/Player.prefab"}
unity_execute: prefab.instantiate {prefabPath: "Assets/Prefabs/Player.prefab", x: 0, y: 1, z: 0}
unity_execute: prefab.open {path: "Assets/Prefabs/Player.prefab"}
unity_execute: prefab.save
unity_execute: prefab.close

7. Asset Management

unity_execute: asset.find {query: "Player", type: "Prefab"}
unity_execute: asset.copy {sourcePath: "Assets/Prefabs/Player.prefab", destPath: "Assets/Backup/Player.prefab"}
unity_execute: asset.move {sourcePath: "Assets/Old/Item.prefab", destPath: "Assets/New/Item.prefab"}

8. Package Management

unity_execute: package.list
unity_execute: package.search {query: "TextMeshPro"}
unity_execute: package.add {packageName: "com.unity.textmeshpro"}
unity_execute: package.add {gitUrl: "https://github.com/example/package.git"}

9. Test Running

unity_execute: test.list {testMode: "EditMode"}
unity_execute: test.run {testMode: "EditMode", filter: "PlayerTests"}
unity_execute: test.getResults

10. Script Execution (Enhanced)

script.execute is project-changing: it needs the operator opt-in and confirm: true on every call (see Safety and permissions).

# Debug logging
unity_execute: script.execute {code: "Debug.Log('Hello')"}, confirm: true

# Time manipulation
unity_execute: script.execute {code: "Time.timeScale = 0.5"}

# PlayerPrefs
unity_execute: script.execute {code: "PlayerPrefs.SetInt('score', 100)"}

# Reflection-based method calls
unity_execute: script.execute {code: "MyClass.MyMethod()"}
unity_execute: script.execute {code: "MyClass.MyStaticMethod('param1', 123)"}

Tool Categories (~100 tools)

Console (3 tools)

• console.getLogs - Get logs with optional type filter (Log/Warning/Error)
• console.getErrors - Get error/exception logs (with optional warnings)
• console.clear - Clear captured logs

Scene (7 tools)

• scene.list - List scenes in build settings
• scene.getActive - Get active scene info
• scene.getData - Get full hierarchy data
• scene.load - Load scene by name (Play mode)
• scene.open - Open scene in Editor mode
• scene.save - Save active scene (Editor mode)
• scene.saveAll - Save all open scenes (Editor mode)

GameObject (8 tools)

• gameobject.find - Find by name, tag, or component
• gameobject.getAll - Get all GameObjects with filtering
• gameobject.create - Create object or primitive (Cube, Sphere, etc.)
• gameobject.destroy - Destroy object
• gameobject.delete - Delete object (alias for destroy)
• gameobject.getData - Get detailed data
• gameobject.setActive - Enable/disable
• gameobject.setParent - Change hierarchy

Transform (6 tools)

• transform.getPosition - Get world position {x, y, z}
• transform.getRotation - Get Euler rotation {x, y, z}
• transform.getScale - Get local scale {x, y, z}
• transform.setPosition - Set world position {x, y, z}
• transform.setRotation - Set Euler rotation
• transform.setScale - Set local scale

Component (5 tools)

• component.add - Add component by type name
• component.remove - Remove component
• component.get - Get component data/properties
• component.set - Set field/property value
• component.list - List available component types

Script (3 tools)

• script.execute - Execute code: Debug.Log, Time, PlayerPrefs, reflection calls
• script.read - Read script file
• script.list - List project scripts

Application (4 tools)

• app.getState - Get play mode, FPS, time
• app.play - Enter/exit Play mode
• app.pause - Toggle pause
• app.stop - Stop Play mode

Debug (3 tools)

• debug.log - Write to console
• debug.screenshot - Capture screenshot
• debug.hierarchy - Text hierarchy view

Editor (9 tools)

• editor.refresh - Refresh AssetDatabase (triggers recompile)
• editor.recompile - Request script recompilation
• editor.domainReload - Force domain reload
• editor.focusWindow - Focus window (game/scene/console/hierarchy/project/inspector)
• editor.listWindows - List open windows
• editor.getState - Get editor state
• editor.play - Enter Play mode
• editor.stop - Exit Play mode
• editor.pause / editor.unpause - Pause control

Input Simulation (10 tools)

• input.keyPress - Press and release key
• input.keyDown / input.keyUp - Hold/release key
• input.type - Type text into field
• input.mouseMove - Move cursor
• input.mouseClick - Click at position
• input.mouseDrag - Drag operation
• input.mouseScroll - Scroll wheel
• input.getMousePosition - Get cursor position
• input.clickUI - Click UI element by name

Material (5 tools) - NEW in v1.5.0

• material.create - Create material with shader, color, metallic, smoothness
• material.assign - Assign material to GameObject
• material.modify - Modify material properties (color, metallic, emission)
• material.getInfo - Get detailed material info with all shader properties
• material.list - List materials in project with filtering

Prefab (5 tools) - NEW in v1.5.0

• prefab.create - Create prefab from scene GameObject
• prefab.instantiate - Instantiate prefab in scene with position
• prefab.open - Open prefab for editing
• prefab.close - Close prefab editing mode
• prefab.save - Save currently edited prefab

Asset (7 tools) - NEW in v1.5.0

• asset.find - Search assets by query, type, folder
• asset.copy - Copy asset to new path
• asset.move - Move/rename asset
• asset.delete - Delete asset (with trash option)
• asset.refresh - Refresh AssetDatabase
• asset.import - Import/reimport specific asset
• asset.getPath - Get asset path by name

Package Manager (4 tools) - NEW in v1.5.0

• package.add - Install package by name or git URL
• package.remove - Remove installed package
• package.list - List installed packages
• package.search - Search Unity package registry

Test Runner (3 tools) - NEW in v1.5.0

• test.run - Run EditMode/PlayMode tests with filtering
• test.list - List available tests
• test.getResults - Get last test run results

Batch Execution (1 tool) - NEW in v1.6.0

• batch.execute - Execute multiple tools in one call (10-100x performance)
  • commands: Array of {tool, params} objects
  • stopOnError: Stop on first error (default: false)

Session (1 tool) - NEW in v1.6.0

• session.getInfo - Get session info (project, processId, machineName, sessionId)

ScriptableObject (6 tools) - NEW in v1.6.0

• scriptableobject.create - Create new ScriptableObject asset
• scriptableobject.load - Load and inspect ScriptableObject fields
• scriptableobject.save - Save ScriptableObject changes
• scriptableobject.getField - Get specific field value
• scriptableobject.setField - Set field value with auto-save
• scriptableobject.list - List ScriptableObjects in project

Shader (3 tools) - NEW in v1.6.0

• shader.list - List shaders in project
• shader.getInfo - Get shader properties and info
• shader.getKeywords - Get shader keywords

Texture (5 tools) - NEW in v1.6.0

• texture.create - Create new texture with color fill
• texture.getInfo - Get texture info (size, format, import settings)
• texture.setPixels - Fill region with color
• texture.resize - Resize texture via import settings
• texture.list - List textures in project

Custom Tools API - v1.6.0

Register project-specific tools:

OpenClawCustomTools.Register(
    "mygame.getScore",
    "Get current score",
    (args) => new { success = true, score = GameManager.Score }
);

MCP Resources - v1.6.0

Access Unity data via MCP resource URIs:

Tips

Screenshot Modes

• Play mode: ScreenCapture - includes all UI overlays
• Editor mode: Camera.main.Render() - no overlay UI
• Use {method: "camera"} for camera-only capture

Finding Objects

gameobject.find {name: "Player"}           # By exact name
gameobject.find {tag: "Enemy"}             # By tag
gameobject.find {componentType: "Camera"}  # By component
gameobject.getAll {activeOnly: true}       # All active objects

Script Recompilation

Unity may not auto-recompile after code changes. Use:

editor.refresh    # Full asset refresh + recompile

Play Mode Transitions

• Plugin survives Play mode transitions via SessionState
• If connection lost, wait for auto-reconnect or use Window → OpenClaw Plugin → Settings → Connect

MCP Bridge Usage

For Claude Code / Cursor integration:

1. Start: Window → OpenClaw Plugin → MCP Bridge → Start
2. Register: claude mcp add unity -- node /path/to/MCP~/index.js
3. Verify: curl http://127.0.0.1:27182/status

Input Simulation Limitation

Keyboard/mouse simulation works for UI interactions but NOT for Input.GetKey(). For gameplay testing:

• Use transform.setPosition to move objects directly
• Or migrate to Unity's new Input System

Troubleshooting

Security & Privacy Disclosure

This skill drives a live Unity Editor — treat it like giving a collaborator editor access. Full disclosure of capabilities:

• Arbitrary code execution (by design): script.execute compiles and runs C# inside the Unity process, and several tools use reflection to reach editor internals. This is the core of editor automation — it also means the skill can do anything the editor can. Only use in trusted, version-controlled projects. Ask the user to review C# snippets before running code they didn't write.
• Destructive operations are gated at runtime (v1.7.0+): deleting GameObjects/assets, saving scenes/projects, installing packages, simulating keyboard/mouse input and running script.execute are refused by the gateway extension unless it was started with OPENCLAW_EDITOR_ALLOW_DESTRUCTIVE=1 and the call carries confirm: true. Confirm the change with the user before setting it — see Safety and permissions.
• Custom tools are gated too (v1.8.0+): any tool name outside the built-in catalogue — including tools a project registers through OpenClawCustomTools — is classified project-changing whatever its verb reads like, and execute_code/manage_tools/execute_custom_tool/script.execute can never be exempted.
• The bridge is authenticated (v1.8.0+): /unity/* needs a per-launch token from ~/.openclaw/unity-bridge.token (mode 0600) to register, a per-session token on every later request, and the issued nonce on every result. Non-loopback peers and browser-originated requests get 403. A process running as your user can read the token file — that is the boundary.
• Package installation: Git-based package installs import external, unvetted code into the project. Verify the source URL with the user first.
• Metadata: the connection handshake includes machine name and process ID (used to route messages to the right editor instance). No other host information is collected or transmitted.
• Network surface: the gateway serves /unity/* to loopback peers only and requires the bridge token; the Editor add-on's MCP bridge listens on 127.0.0.1:27182 and (from plugin 1.7.0) requires its own token. Do not port-forward or reverse-proxy either one.
• Trigger scope: routine-sounding requests ("clean up the scene", "save everything", "just try it") map to state-changing editor operations — confirm once before the first state-changing call in a session.
• Safety defaults: disableModelInvocation: true is set — the model cannot auto-invoke this skill; it runs only on explicit user request. Project-changing tools default to refused, custom tools count as project-changing, the bridge requires a token, and dryRun: true previews any call without sending it. Keep project backups / source control current before automation sessions.

Links

• Skill Repository: https://github.com/TomLeeLive/openclaw-unity-skill
• Plugin Repository: https://github.com/TomLeeLive/openclaw-unity-plugin
• OpenClaw Docs: https://docs.openclaw.ai
• MCP Setup Guide: See Plugin Repository → Documentation~/SETUP_GUIDE.md

License

Apache-2.0 — See LICENSE.md