android-device-automation
Pass
Audited by Gen Agent Trust Hub on Apr 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes
adb shellvia therunadbshellcommand to facilitate deep interaction with Android devices, including app management and system-level control. - [EXTERNAL_DOWNLOADS]: The skill fetches the
@midscene/androidpackage from the npm registry usingnpxto perform its automation tasks. - [REMOTE_CODE_EXECUTION]: Use of
npx -y @midscene/android@1triggers the execution of remote code in the local environment to drive the automation engine. - [PROMPT_INJECTION]: The skill's vision-driven automation creates a surface for indirect prompt injection, where malicious instructions embedded in a target application's UI or a website could influence the agent's behavior.
- Ingestion points: Visual data captured through
take_screenshotand processed duringactsequences (SKILL.md). - Boundary markers: Absent; the skill does not provide explicit delimiters or instructions for the vision model to ignore commands embedded in UI elements.
- Capability inventory: The agent can execute arbitrary shell commands on the device via
runadbshelland perform complex UI interactions viaact(SKILL.md). - Sanitization: No filtering or sanitization is performed on visual content before it is analyzed by the multimodal model.
Audit Metadata