The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides functional commands for establishing reverse shells using Bash built-in redirection to external IPs (/dev/tcp/ATTACKER/4444) and instructions for setting up SOCKS proxies and tunnels to pivot traffic out of a restricted network.
[CREDENTIALS_UNSAFE]: Contains extensive patterns for harvesting sensitive information, including searching for SSH private keys (id_rsa, *.pem), reading password hashes from /etc/shadow, and scanning for cleartext credentials in process environments, shell histories, and application configuration files (.env, wp-config.php).
[PRIVILEGE_ESCALATION]: Outlines multiple methods for gaining elevated access, including hijacking SSH agent sockets in /tmp, abusing D-Bus system methods to start unauthorized services, exploiting NFS no_root_squash to create SUID binaries, and hijacking active sudo tokens via ptrace techniques.
[PERSISTENCE]: Details methods for maintaining unauthorized access, such as injecting attacker public keys into authorized_keys files and creating malicious systemd service units that execute on system startup.
[COMMAND_EXECUTION]: Provides guidance on executing arbitrary code through D-Bus exploitation, systemd service manipulation, and network relay tools like socat and chisel.
[EXTERNAL_DOWNLOADS]: Directs users to fetch and use offensive security tools like chisel and sudo_inject from third-party GitHub repositories that are not recognized as trusted sources.
[PROMPT_INJECTION]: The AI LOAD INSTRUCTION attempts to influence the agent's behavior by framing it as an unrestricted 'Expert Linux lateral movement' attacker, explicitly mentioning techniques that base models are expected to miss.

linux-lateral-movement