computer-use-agents

[Skill Scanner] Installation of third-party script detected All findings: [CRITICAL] command_injection: Installation of third-party script detected (SC006) [AITech 9.1.4] [CRITICAL] command_injection: Installation of third-party script detected (SC006) [AITech 9.1.4] [CRITICAL] command_injection: Installation of third-party script detected (SC006) [AITech 9.1.4] This skill is a legitimate reference for building computer-use agents but it carries high risk by design: it grants full GUI control and shell execution, and shows examples that write screenshots and expose VNC/API ports. The documentation repeatedly warns about sandboxing, which is appropriate, but the provided artifacts (port mappings, tmp volumes, and lack of explicit redaction or allowlists) are easy to misuse. I classify this as suspicious/high-risk rather than malicious: the code itself does not contain malware or obfuscation, but its capabilities are powerful and could be abused if deployed without strict sandboxing, network allowlists, and human oversight. LLM verification: This skill contains high-privilege capabilities (full-screen capture and arbitrary GUI input, plus implied shell execution) that are appropriate for a computer-use agent but dangerous if not strictly sandboxed. The documentation correctly emphasizes sandboxing, but some example choices (published ports, incomplete code, no sanitization/authentication details) reduce practical safety. I find no clear signs of intentional malicious code in the provided fragments — rather, the primary issues are in