computer-use-agents

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill deliberately implements full-screen screenshot capture and base64 encoding combined with cloud model/tool integration (Anthropic), plus exposed VNC/API ports and remote bash/text-editor tooling, which enable intentional data exfiltration of sensitive screen contents and remote command execution/backdoor-style control.

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

• Third-party content exposure detected (high risk: 0.85). The required runtime workflow feeds the agent’s LLM context with outsider-authored free text from the current desktop UI via the screenshot/perception step (the screenshot can contain arbitrary text from websites/apps the user didn’t author), which the vision-language model then reads as part of its prompt context.