The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the discovery and loading of 'Agent Skills' from the Unreal project directory via the AgentSkillToolset.GetSkills tool. These instructions are designed to override the agent's default behavior, creating a potential surface for indirect prompt injection if a project contains malicious instruction files that the agent is directed to follow.
Ingestion points: AgentSkillToolset.GetSkills (SKILL.md)
Boundary markers: None mentioned; project-specific instructions are explicitly stated to take precedence over defaults (SKILL.md)
Capability inventory: Extensive capabilities including actor manipulation, Blueprint/Material editing, and C++ compilation via LiveCodingToolset.CompileLiveCoding (SKILL.md)
Sanitization: No sanitization or validation of the loaded instructions is described.
[COMMAND_EXECUTION]: The documentation provides instructions for executing local binaries to install a proxy server (unreal_mcp_proxy.exe or unreal_mcp_proxy) and uses Unreal Editor console commands (ModelContextProtocol.StartServer, ModelContextProtocol.GenerateClientConfig, etc.) to manage the server and client configurations.
[PERSISTENCE]: The skill setup instructions guide the user to enable bAutoStartServer=True in the EditorPerProjectUserSettings.ini file. This configuration ensures the MCP server, which provides the agent with control over the Unreal Editor, is automatically launched whenever the project is opened.
[DYNAMIC_EXECUTION]: The skill can invoke the LiveCodingToolset.CompileLiveCoding tool, which triggers a C++ compilation process within the running Unreal Editor. This allows the agent to dynamically rebuild and execute C++ code changes.

unreal-mcp