game-automation
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides the ability to read and modify the Windows Registry via the
um win regcommand (e.g.,um win reg get "HKCU\Software\..."). Accessing or altering the registry is a sensitive system operation that can expose private application data or modify the behavior of the operating system. - [COMMAND_EXECUTION]: The skill allows the agent to terminate arbitrary processes using
um win kill <pid>. While intended for cleaning up crashed game components or reporters, this capability could be used to terminate security software or other critical user applications if the agent is misdirected. - [COMMAND_EXECUTION]: The skill implements automated user input simulation, including keystrokes, mouse clicks, and typing (
um win drive). Although the documentation includes warnings about avoiding focus on online games while a human is active, the tool itself can send input to any windowed process, potentially performing unintended actions in sensitive applications like browsers or terminals. - [INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by having the agent ingest and process external, untrusted data to drive its logic.
- Ingestion points: The agent reads game logs and processes screenshots captured via
um win shotto decide its next actions. - Boundary markers: The instructions do not specify any delimiters or sanitization steps to distinguish between legitimate log data and potentially malicious instructions embedded within those logs.
- Capability inventory: The agent has access to powerful capabilities including input simulation (
um win drive), process termination (um win kill), and registry manipulation (um win reg). - Sanitization: There is no evidence of filtering or validation of the content read from the game environment before it influences the agent's decision-making loop.
Audit Metadata