analyze-mobile-app

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill extensively uses adb shell commands to interact with Android devices. This includes simulating physical inputs (tap, swipe, keyevent, text typing) and extracting system-level information (wm size, dumpsys window).
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it parses untrusted UI data to drive its navigation logic.
  • Ingestion points: UI hierarchy data is retrieved from the device via adb shell cat /sdcard/ui.xml and processed by the agent.
  • Boundary markers: No boundary markers or instructions are provided to the agent to help it distinguish between mobile app content and operational commands.
  • Capability inventory: The skill possesses high-privilege capabilities on the connected device via adb shell, including the ability to enter text, click buttons, and navigate the file system.
  • Sanitization: There is no evidence of sanitization or filtering of the UI data before it is used to calculate coordinates for interaction or to generate reports.
  • [DATA_EXFILTRATION]: While no network-based exfiltration is detected, the skill extracts sensitive UI information and screenshots from a mobile device to a local host environment, which may include user data if private applications are active during the analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 06:23 PM