computer-use-browser-desktop
Fail
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructions include a prerequisite command that downloads a script from an external URL (https://box.ascii.dev/install) and pipes it directly into the shell (sh). This 'curl | sh' pattern is a high-risk operation that allows a remote server to execute arbitrary code on the local machine without prior user inspection of the script.
- [COMMAND_EXECUTION]: The skill utilizes the 'box' and 'lux' CLI tools to execute various shell commands for managing virtual machines, establishing SSH connections, and controlling cloud-based desktop environments.
- [PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it passes natural language goals to the 'Lux' automation agent. 1. Ingestion points: The 'lux start' command in SKILL.md. 2. Boundary markers: No markers or warnings are used to delimit the goal string. 3. Capability inventory: The skill can execute arbitrary commands on remote VMs and control browser sessions. 4. Sanitization: No sanitization or validation of the input goals is specified.
Recommendations
- HIGH: Downloads and executes remote code from: https://box.ascii.dev/install - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata