ue-code-authoring
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project files such as C++ source code and build configurations. The agent has extensive capabilities to write files, execute shell commands, and build code, but the skill lacks explicit boundary markers or sanitization protocols to prevent the agent from following instructions embedded in the data it analyzes. * Ingestion points: Reads source code and configuration files within the project workspace. * Boundary markers: None mentioned. * Capability inventory: File writing, Bash execution, solution building, and Python execution in the editor. * Sanitization: None present.
- [DYNAMIC_EXECUTION]: The skill facilitates solution builds and the execution of Python code directly within the Unreal Engine editor via the Rider MCP tools. These features allow for the execution of arbitrary logic within the development environment.
- [COMMAND_EXECUTION]: The agent is instructed to use Bash for git and toolchain operations, providing a direct shell interface for system interaction.
- [DATA_EXFILTRATION]: The read_file tool is explicitly documented to support path traversal (../), which allows the agent to access files outside the project root. While intended for reading engine sources, this capability could be exploited to read sensitive system files if directed by a malicious prompt.
Audit Metadata