The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates extensive control over mobile devices through the agent-device CLI, allowing for app lifecycle management, UI interaction via coordinates and selectors, and modification of system permissions.
[EXTERNAL_DOWNLOADS]: The install-from-source command allows for the automated download and installation of application binaries from remote URLs provided during runtime.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its data ingestion patterns.
Ingestion points: Data is ingested from external mobile applications via snapshot -i (UI element labels and text), logs path (application log files), and network dump (captured HTTP traffic).
Boundary markers: The provided instructions do not specify the use of delimiters or 'ignore' instructions to isolate the agent from potentially malicious text embedded within the mobile application's UI or log output.
Capability inventory: The agent has high-privilege capabilities including installing software, entering text, and clicking elements, which could be triggered by instructions hidden in the data it processes.
Sanitization: The skill does not mention validation or sanitization of strings extracted from the mobile environment before they are added to the agent's context.

agent-device