argent-device-interact

Pass

Audited by Gen Agent Trust Hub on May 14, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from mobile user interfaces, creating a surface for indirect prompt injection.
  • Ingestion points: Discovery tools including describe, debugger-component-tree, and screenshot (SKILL.md) ingest content directly from the device screen, which may include text or identifiers from untrusted third-party apps.
  • Boundary markers: The skill lacks explicit boundary markers or instructions to the agent to disregard instructions potentially embedded within retrieved UI data.
  • Capability inventory: The agent has access to powerful interaction capabilities such as keyboard for text and special key injection, gesture-tap for UI manipulation, and open-url for navigating to web pages or deep links (SKILL.md).
  • Sanitization: No sanitization or validation logic is defined for the UI content captured by the discovery tools before it is used to determine subsequent agent actions.
Audit Metadata
Risk Level
SAFE
Analyzed
May 14, 2026, 10:26 PM