pokeclaw-android-ai-agent

Pass

Audited by Gen Agent Trust Hub on Apr 11, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: Implements automation via the Android Accessibility Service, allowing the agent to programmatically perform gestures such as tapping, swiping, and typing text into UI elements across the system.
  • [EXTERNAL_DOWNLOADS]: Fetches a Gemma 4 LiteRT model from Google's official HuggingFace repository for on-device inference.
  • [DATA_EXFILTRATION]: Accesses and parses the active window's UI tree and system notifications to provide the agent with environmental context. This involves reading sensitive on-screen data to enable the AI's decision-making process.
  • [PROMPT_INJECTION]: The skill processes untrusted screen data and notification text through the LLM. While it contains a capability inventory for device control, this behavior is central to its stated purpose as an accessibility-based automation agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 11, 2026, 05:10 AM