use-windows-vm-computer-use

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill contains explicit instructions to disable sandboxing (--yolo), run arbitrary agent prompts in an interactive user session via scheduled tasks, and recommends Base64-encoded PowerShell transport and silent installs — patterns that enable remote code execution and obfuscation and could be abused as a backdoor.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). SKILL.md describes running a VM-local Codex computer-use backend by placing a user-provided “” prompt into a scheduled task argument (and then executing codex --yolo exec ... ' -o resultPath ' + $prompt), so outsider-authored free text in that prompt is directly ingested into the LLM at runtime.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs disabling sandbox/approval protections with --yolo, registering temporary scheduled tasks (including optionally running at highest/elevated runlevel), and installing software in the VM—actions that bypass security and modify system state.