agent-device

Pass

Audited by Gen Agent Trust Hub on May 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill's primary function is to orchestrate the agent-device CLI tool to perform actions like opening apps, tapping UI elements, and managing sessions. This includes a batch mode for executing sequences of commands from a JSON file.
  • [EXTERNAL_DOWNLOADS]: The bootstrap-install.md file indicates that the install and reinstall commands can accept URLs as artifact paths for downloading and installing application binaries (e.g., .apk, .ipa).
  • [DATA_EXFILTRATION]: The remote-tenancy.md reference describes how to interact with remote macOS hosts using curl to perform RPC calls, which involves transmitting authentication tokens and lease identifiers over the network to the daemon endpoint.
  • [REMOTE_CODE_EXECUTION]: The skill supports a replay command that executes automation scripts from local files, which serves as a mechanism for running predefined instruction sets on target devices.
  • [INDIRECT_PROMPT_INJECTION]: By design, the skill reads and processes UI content via the snapshot and get text commands. This creates an attack surface where malicious content within a third-party application's UI could potentially influence the agent's subsequent logic or actions.
Audit Metadata
Risk Level
SAFE
Analyzed
May 12, 2026, 08:13 AM