Getting Started with Skills

Prerequisites

This lesson uses 4 skills. No prior experience with AgentUX skills is needed. This is where you start. Every skill and the full GraphRAG verification loop run on Free. Pro extends Free with cross-project asset analytics and Cypher access for graph queries the JSON-RPC layer cannot easily express, and Team Studio extends Pro with shared studio infrastructure for teams.

What You'll Build

By the end of this lesson, you will have a simple scene with a positioned camera, properly configured lens, and a key light illuminating a prop, all set up by telling Claude what you want. More importantly, you will understand the three ways skills activate and how to chain them together naturally.

Step-by-Step Workflow

Step 1: Invoke your first skill

Open Claude Code in your Unreal Engine project and type:

/skill camera-placement

This loads the Camera Placement skill into your session. Claude now has deep expertise in camera positioning, composition rules, and UE5 camera types.

Now tell Claude what you want:

"Place a CineCamera in my level looking at the origin from about 5 meters away at a 30-degree downward angle."

What to expect: Claude will use the AgentUX plugin to spawn a CineCameraActor in your level, position it at the coordinates you described, and aim it at the origin. You will see the camera appear in your viewport.

Step 2: Understand auto-detection

You do not always need to explicitly invoke a skill. AgentUX can auto-detect which skill is needed based on your prompt. Try asking Claude:

"Make the camera use a 35mm lens with a wide aperture for shallow depth of field."

What happens: AgentUX recognizes that your request involves lens and optics configuration. It automatically activates the lens-optics skill and applies the settings to your camera. You did not need to type /skill lens-optics. The system detected it.

When to invoke explicitly: Auto-detection works for most requests. Use explicit /skill <name> when you want to ensure a specific skill is active, or when your request could match multiple skills.

How it works behind the scenes

When Claude acts on your request, the skill guides it to verify UE API references against the GraphRAG knowledge base before executing. GraphRAG contains 887,000+ nodes of indexed Unreal Engine source code across all 5.x releases. This means Claude looks up actual property names, function signatures, and class hierarchies rather than guessing from training data. The result: correct property names, fewer silent failures, and awareness of version-specific changes like UE 5.7.4's float to real display name rename.

You do not need to interact with GraphRAG directly. Skills handle the verification automatically. If you want to query it yourself, use tools like get_class_details("UCameraComponent") or search_product_commands("camera").

Step 3: Chain skills together

Real workflows combine multiple skills in sequence. Now add a light to your scene:

/skill lighting-fundamentals

Then tell Claude:

"Add a key light, a warm directional light coming from camera-left at about 45 degrees. I want soft shadows."

What to expect: Claude places a directional light with the angle and warmth you described, configures shadow softness, and positions it relative to your camera. The scene now has shape and depth from the lighting.

Step 4: Compose a simple shot

Now bring it together. Tell Claude:

"Place a chair at the origin as my subject, then adjust the camera for a classic medium shot with the chair in the right third of frame."

What happens: AgentUX chains set-blocking (to place the chair) and camera-placement (to reframe) together. Skills work together when the request spans multiple domains. Claude coordinates them automatically.

Tips & Best Practices

• Be specific about what you want. "Place a camera" is vague. "Place a CineCamera 3 meters from the desk, eye-level, looking slightly left" gets precise results.
• Use film terminology. Skills understand terms like "medium shot," "Dutch angle," "rim light," and "bounce light." Speak the language of the craft.
• Iterate, do not restart. If the camera is not quite right, say "move it 1 meter to the right" instead of starting over. Skills maintain context within your session.
• Check the viewport. After each skill action, look at your Unreal Engine viewport to confirm the result matches your intent. Adjust as needed.

Next Steps

Now that you understand how to invoke, auto-detect, and chain skills, try these next: