server-security

Fail

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: HIGHCOMMAND_EXECUTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is vulnerable to command injection through the $ARGUMENTS variable. User input is assigned to SSH_TARGET and then interpolated directly into multiple shell commands (e.g., ssh $SSH_TARGET "...") without sanitization. An attacker could provide a crafted input string to execute arbitrary commands on the host machine running the AI agent.
  • [DATA_EXFILTRATION]: The skill accesses highly sensitive files on the target server to perform its audit, including /etc/passwd, /etc/ssh/sshd_config, and various credential files like .env, .key, and .pem files. While this is the intended purpose of the skill, the retrieval of these files into the agent's context constitutes significant data exposure.
  • [REMOTE_CODE_EXECUTION]: The skill executes a wide range of commands on remote systems via SSH, including installing new packages (apt-get install fail2ban), modifying system configuration files (/etc/nginx/nginx.conf), and changing firewall rules (ufw).
  • [DATA_EXFILTRATION]: The skill explicitly searches for and reads the contents of sensitive files such as .env, .key, .pem, and database files (.db, .sqlite) across multiple directories (/opt, /var/www, /home).
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 19, 2026, 03:58 PM
Security Audit — agent-trust-hub — server-security