The Agent Skills Directory

[COMMAND_EXECUTION]: The exec command allows for raw command strings to be passed to the emulator bridge, providing flexible but potentially unsafe execution of simulator controls if driven by unvalidated agent logic.
[DATA_EXFILTRATION]: Accessing the accessibility tree via the ax command exposes the internal UI structure of simulated apps, which may contain sensitive user or application data visible to the agent.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface where untrusted data from a simulated application could influence agent behavior.
Ingestion points: UI metadata and the accessibility tree are ingested via orca emulator ax (SKILL.md).
Boundary markers: There are no specific delimiters or instructions defined to isolate data retrieved from the simulator from the agent's core instructions (Absent).
Capability inventory: The skill provides a wide range of interaction tools, including tap, type, gesture, and exec, which can be used to perform actions based on potentially injected instructions (SKILL.md).
Sanitization: The skill does not describe any mechanism for sanitizing or validating the UI content before it is returned to the agent (Absent).
[EXTERNAL_DOWNLOADS]: The skill references and integrates the serve-sim open-source tool for emulator streaming and control.

orca-emulator