device-interaction
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading and acting upon data from an external application's UI, which serves as an untrusted input vector.\n
- Ingestion points: The
DeviceEventSynthesizetool captures and parses UI hierarchy files and screenshots to locate elements (SKILL.md).\n - Boundary markers: No specific delimiters or instructions are provided to help the agent ignore or isolate potentially malicious text within the UI components (e.g., instructions hidden in labels or button names).\n
- Capability inventory: The agent can synthesize touch events, hardware button presses, and keyboard input, and it can configure app launch parameters via
DeviceInteractionInstallAndRun(SKILL.md).\n - Sanitization: There is no mention of sanitizing or validating UI text before the agent interprets it to make functional decisions or reports results.\n- [COMMAND_EXECUTION]: The
DeviceInteractionInstallAndRuntool allows the agent to specifycommandLineArgumentsandenvironmentVariables. This capability is essential for the skill's purpose of testing different application states but provides a mechanism to influence the execution of the target binary on the host or simulator.
Audit Metadata