Mentions & Commands

UND's input area supports two special syntaxes that let you pull context into the conversation or trigger workflow actions without leaving the chat. Type @ to reference files, folders, URLs, and editor state. Type / to execute commands that manage the conversation and task flow.

Both trigger an autocomplete popup that filters results as you type, so you rarely need to remember exact paths or command names.

@Mentions

Mentions inject external context directly into your message before it reaches the AI. This is how you point the agent at specific files, show it your project structure, or give it visual information from the viewport.

@/path/to/file.cpp -- File Content

Reference any file in your project by its path. The file's full contents are injected into the conversation context, so the AI can read and reason about it without needing to call read_file first.

@Source/MyGame/Character/MyCharacter.h Can you add a health component to this class?

This is faster than waiting for the AI to discover and read the file itself, and it ensures the agent is looking at exactly what you want.

@/path/to/folder/ -- Directory Listing

Reference a directory to inject its listing into the context. The AI sees the folder structure and file names, which helps it understand project organization before diving in.

@Source/MyGame/Abilities/ What ability classes do I have?

@https://url -- Web Page Content

Paste a URL with the @ prefix to fetch the page content and inject it as plain text. Useful for referencing documentation, API references, or forum posts.

@https://dev.epicgames.com/documentation/en-us/unreal-engine/gameplay-ability-system-for-unreal-engine How do I set up GAS in my project?

@problems -- Compiler Errors

Injects the current list of compiler errors and warnings from the editor. Use this when you want the AI to diagnose build failures without manually copying error messages.

@problems Fix these compilation errors

@gitchanges -- Git Diff

Injects the output of git diff HEAD, showing all uncommitted changes in the project. Useful for code review or asking the AI to summarize what changed.

@gitchanges Summarize what I changed and suggest a commit message

@asset:Game/BP_Foo -- UAsset Metadata

Reference an Unreal asset by its content path to inject its metadata (class, parent class, properties) into the context.

@asset:Game/Blueprints/BP_PlayerCharacter What components does this Blueprint have?

@screenshot -- Viewport Capture

Captures the current editor viewport as an image and sends it to the AI as a vision input. The AI can analyze what it sees in the viewport and respond accordingly.

@screenshot The lighting in this scene looks flat. How can I improve it?

This requires a model that supports image inputs (most current models from Anthropic, OpenAI, and Google do).

/Slash Commands

Slash commands trigger actions that manage the conversation, context window, or task state. They execute immediately and do not consume an agent turn.

/newtask

Start a fresh conversation task. This clears the current conversation context and begins a new task with a clean slate. The previous conversation is preserved in the task history and can be revisited from the History view.

/newtask

/clear

Clear the current conversation display. This removes all messages from the chat view but does not affect the underlying task state or history.

/clear

/condense

Trigger AI-powered conversation condensing. The system takes all messages except the most recent ones, summarizes them into a compact system message using the AI provider, and replaces the original messages. This significantly reduces context window usage while preserving the essential information from earlier in the conversation.

Use this when you are deep into a long task and notice the token usage bar filling up.

/condense

/compact

An alias for /condense. Both commands do exactly the same thing.

/compact

/help

Display the list of available slash commands with brief descriptions.

/help

/rules

Load a .undrules file from your project root. This file contains project-specific instructions that are injected into the system prompt for every conversation. Use it to define coding standards, naming conventions, architectural constraints, or any other rules the AI should follow when working on your project.

/rules

Tips

• Combine mentions with natural language. You can use multiple mentions in a single message: @Source/MyGame/Player.h @Source/MyGame/Enemy.h Compare these two classes and suggest a common base class.

• Use @problems after compilation. When a build fails, type @problems to instantly give the AI all the error context it needs to diagnose and fix issues.

• Use /condense proactively. Do not wait until you hit the context limit. If a conversation has gone through 20+ tool calls, condensing early keeps the agent responsive and accurate.

• @screenshot works with any viewport. It captures whatever the active editor viewport is showing -- level geometry, material previews, Blueprint graphs, or any other editor content.