Skills System
The Skills System is a structured knowledge base that gets injected into the AI system prompt based on the detected engine and bridge connection state. It gives the AI deep, accurate knowledge about Unity and Godot without relying on general training data.
How it works
When you start a conversation, the IDE detects your game engine and assembles a skills prompt block. This block is injected into the AI system prompt on the first message of each conversation, giving the AI:
- Engine-specific coding knowledge and best practices
- Bridge command format and usage patterns
- Critical rules and common pitfalls
- Resolution and UI layout awareness
Skill categories
Engine skills
Comprehensive knowledge about the game engine, injected regardless of bridge connection. Covers:
Core architecture
- Scene graph
- Node/Component systems
- Lifecycle methods
- Resource management
Scripting
- GDScript patterns
- C# conventions
- Signal/event systems
- Coroutines and async
Physics
- Collision layers/masks
- Raycasting
- Body types
- Shape configuration
Rendering
- Materials and shaders
- Lighting setup
- Camera configuration
- 2D and 3D rendering
UI systems
- Control nodes / Canvas
- Layout containers
- Input handling
- Responsive design
Animation
- AnimationPlayer/Animator
- Tweening
- State machines
- Blend trees
Audio
- AudioStreamPlayer
- Spatial audio
- Bus routing
- Music/SFX patterns
3D
- Mesh rendering
- Environment setup
- Lighting and shadows
- Navigation meshes
Performance
- Object pooling
- LOD systems
- Draw call optimization
- Memory management
Common pitfalls
- Engine-specific gotchas
- Antipatterns
- Version-specific issues
- Platform differences
Bridge skills
Instructions for how the AI should use bridge commands. These are only injected in Agent mode and adapt based on connection state:
- Connected: Instructions tell the AI to execute bridge commands for scene construction in real-time
- Disconnected: Instructions tell the AI to still output bridge commands, marked as pending for later execution
Unity bridge skills
When a Unity project is detected and Agent mode is active, the AI receives these specific instructions:
- Use the bridge for scene construction; write C# only for runtime behavior
- Before building UI, call
editor.getScreenInfoandscene.getHierarchyto understand canvas dimensions - CanvasScaler is auto-configured when added, so don't set it manually
- Size UI elements relative to the reference resolution returned in bridge responses
- Follow the UI hierarchy: Canvas → CanvasScaler → Panel → Button → Text
- Always wire up serialized reference fields after creating objects using
component.setProperty parentPathis required when creating child objects
Godot bridge skills
When a Godot project is detected and Agent mode is active, the AI receives these specific instructions:
- Use the bridge for scene construction; write GDScript only for runtime behavior
- New scene workflow: Write scripts first, then call
editor.getScreenInfo+scene.create, then create all nodes in one batch - Existing scene workflow: Call
scene.getTreefirst, review the tree, then modify - Godot uses a node tree, not a GameObject+Component model. Node types are built-in (Sprite2D, CharacterBody2D, Label)
- Scripts attach directly to nodes
- Critical physics rules: CollisionShape2D must have a shape resource assigned, use full dimensions (not half-extents), set CapsuleShape2D height before radius
- Every physics body needs a CollisionShape2D child with a shape assigned
- Must call
scene.saveafter making changes
Smart context injection
Project context (file lists, scene analysis, bridge state) is only sent on the first API call of each conversation. This keeps subsequent messages fast and within token limits while still giving the AI full project awareness.
The skills prompt is cached per engine and regenerated only when the engine type or bridge connection state changes.