Agent Skills

A library of Unreal-specific procedures the agent reads before it acts, so it follows a known-good recipe for your project instead of guessing its way through the editor.

Nine hundred callable tools do not add up to competence. An agent that can call SetParameterByIndex still has to know that a material change does not show in the level until the material is applied, that a ProgressBar fill is driven by its Percent property and not by its size, or that saving while inside a Multi-User session writes into a Concert sandbox. Agent Skills are the layer that carries that knowledge: short, executable procedures the agent reads before it touches the editor, stored in the project itself so they can describe your project rather than Unreal in general.

What a skill actually is

A skill is a real Unreal asset: a Blueprint whose parent class is UAgentSkill, creatable from the Content Browser under Artificial Intelligence, Generative AI, or authored by the agent itself. It carries two authored fields. Description is the one-line summary of what the skill covers and when to reach for it. Instructions is the multi-line body holding the actual procedure and its gotchas. Because a skill is an asset and not a loose file, it sits in Content, travels with the project, and goes through source control and review like everything else your team ships.

How the agent finds the right one

• ListSkills returns only a map of skill path to short description, so surveying the entire library costs almost nothing in context.
• GetSkills takes an explicit list of paths and returns the full instruction body for just those skills, which is the second half of the same progressive-disclosure trick the toolset meta-tools use.
• The instructions block the MCP server returns during the initialize handshake tells the agent to call ListSkills first whenever it is asked what it can do, then read the relevant skill before starting the work.
• CreateSkill and UpdateSkill let the agent write the library back. Both are documented as calls it may only make after explicit direction from you.
• ClaudusBridge also exposes CreateAgentSkill on its own Asset toolset, taking a folder, name, description and instructions body, and returning the generated class path of the new skill.

What ships in the box

Version 1.2.0 ships twenty-five skills inside the plugin, mounted under the ClaudusBridge Skills folder, covering the editor domains where agents most reliably get stuck.

• Editor and scene: EditorNavigation, SceneAndActors, LevelsAndLighting, NonIntrusiveMultiLevel, ObjectProperties, HeadlessCapture
• Blueprints and gameplay: Blueprints, BlueprintGraphDSL, GameFramework, CharacterMovement, EnhancedInput, GameplayTagsAndData, SaveGame, Animation
• Rendering and effects: Materials, CustomHLSL, Niagara, UMG
• Collaboration and workflow: MultiUserEditing, MultiAgentOrchestration, TerminalGuide, ToolLibraryAuthoring, SkillGrowth
• Showcase: ChessGameplay, MultiUserChess

Growing the library on your own project

The shipped set is a starting point, not the destination. The SkillGrowth skill sets the protocol for adding to it: when you finish a durable, reusable workflow, or hit a non-obvious project gotcha that no existing skill covers, capture it. Deduplicate against ListSkills first, confirm with the user before writing, and keep the result executable rather than descriptive. That is what turns the library from generic Unreal advice into your studio's naming conventions, your water system's quirks, and the three steps your build always needs in the right order.

Against the 5.7 generation

This is one entry on the Galidar roadmap. For what already shipped in each release, see the Changelog.