Command Injection in PentestAgent

Executive Summary

This vulnerability is an OS command injection flaw in GH05TCREW PentestAgent up to a specific version. It exists in the LocalRuntime.execute_command function of runtime/runtime.py. The issue allows remote attackers to execute arbitrary shell commands on the operator's machine by manipulating inputs like HTTP responses or nmap results that contain prompt-injection content. The attack leverages the fact that LocalRuntime uses asyncio.create_subprocess_shell without command validation or sandboxing.

Useful 0 Not Useful 0

Detection Guidance

Check for unauthorized command execution by monitoring shell history and process logs for unexpected commands. Look for commands containing 'curl', 'bash', 'nc', 'socat', or '/dev/tcp' in logs or network traffic. Inspect files like ~/.ssh/authorized_keys, /etc/cron*, /etc/passwd, /etc/shadow, or ~/.bashrc for modifications. Use 'ps aux' or 'top' to detect suspicious running processes.

Useful 0 Not Useful 0

Impact Analysis

If you use PentestAgent with LocalRuntime, an attacker could trick the tool into executing malicious commands on your system. This could lead to full system compromise, data theft, or installation of malware. The default runtime is LocalRuntime, so most users are exposed unless they explicitly use DockerRuntime for isolation. Attackers could exploit this via web responses, SSH banners, or DNS records.

Useful 0 Not Useful 0

Mitigation Strategies

Immediately switch to DockerRuntime by using the --docker flag to isolate command execution. Update to the latest version with the command validator fix. Block known malicious patterns in network traffic and disable LLM-generated shell commands until patched. Review and remove any unauthorized files or processes detected.

Useful 0 Not Useful 0