android-agent

• Analyzed: May 16, 2026, 4:07 AM
• Findings: 1
• Version: 1.1.1
• Hash: de3d89a192cba2085bf8e7483a527e429617f050b24edb416a42545db25323b0
• Source: palm
• External report: View on VirusTotal

Findings (1)

This skill is classified as suspicious due to its inherent high-risk capabilities and privacy implications, despite extensive documentation and warnings. The primary concern is the explicit design feature of sending screenshots of the connected Android device's screen content to OpenAI's GPT-4o vision model for processing, as detailed in `SKILL.md`. While this is a documented function necessary for the AI agent's operation, it constitutes data exfiltration of potentially sensitive on-screen information to a third-party service. Additionally, the skill leverages ADB and Android accessibility services, granting powerful, near-root level control over the device, and handles the `ANDROID_PIN` via environment variables, which are significant security risks, even though they are clearly disclosed in the `SKILL.md`'s 'Security' and '⚠️ Security Notes' sections. There is no evidence of *intentional* hidden malicious behavior, obfuscation, or unauthorized persistence.