The Agent Skills Directory

[COMMAND_EXECUTION]: The script executes local system commands including adb, emulator, flutter, and nc. These operations are essential for its primary function of managing and interacting with an Android emulator. The script implements a require_int validation function that strictly checks numeric arguments (coordinates, durations, line counts) before they are used in shell arithmetic or passed to subprocesses, effectively preventing command injection attacks.
[PROMPT_INJECTION]: The skill handles data from external sources (the Android accessibility tree and app logs) which may contain content from an untrusted app. To mitigate indirect prompt injection, the script wraps this data in <untrusted-ui-data> or <untrusted-ui-xml> tags and includes explicit warnings to the agent to treat the content as data only. Additionally, UI labels are processed using Python's repr() function to escape control characters and prevent terminal injection.
[DATA_EXFILTRATION]: No evidence of remote data exfiltration was found. Network activity is confined to local communication with the emulator's qemu console on localhost via nc (netcat).
[CREDENTIALS_UNSAFE]: The skill does not contain hardcoded secrets. It accesses the local emulator authentication token (~/.emulator_console_auth_token) using standard methods required for console interaction.

android-emulator