The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs high-privilege operations using adb shell through several scripts, including system-level app management (pm clear, am force-stop) and root-level interactions with the device environment.
[CREDENTIALS_UNSAFE]: The adb-karrot-test-login.py script accepts email and password as plaintext command-line parameters. This exposes sensitive user credentials in system process lists, shell history, and logs.
[DATA_EXFILTRATION]: The skill monitors system logs (logcat) via adb-karrot-liapp-monitor.py for specific security keywords and captures screenshots for AI analysis. These capabilities provide a surface for harvesting sensitive system and application data beyond the intended automation scope.
[EXTERNAL_DOWNLOADS]: The karrot-bypass-playintegrity.toon workflow automates the installation of external binary Magisk modules (e.g., PlayIntegrityFork.zip). These modules modify core system behavior to bypass security SDKs and represent unverifiable third-party code with system-level permissions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of untrusted screen data and system logs. Ingestion points: karrot_ai_vision.py and adb-karrot-liapp-monitor.py. Boundary markers: Absent. Capability inventory: High-privilege adb shell execution and file writes. Sanitization: Absent.

adb-karrot