vercel-sandbox

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

• Third-party content exposure detected (high risk: 0.85). The required runtime workflow uses chrome-use open <url> and then reads back page-derived text (e.g., get title --json and snapshot stdout) into the agent/LLM context, which can include outsider-authored free text from arbitrary web pages.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill runs remote package install/execute commands at runtime—specifically "npm install -g chrome-use" and "npx chrome-use install" which fetch and execute code from the npm registry when no snapshot is provided—so it relies on remote code execution during runtime (npm registry via npm/npx).

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

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly runs sudo package-management commands (e.g., "sudo dnf install" and "sudo ldconfig") when bootstrapping or creating snapshots, which requests elevated privileges and modifies the VM's system state.