network-poisoning-attack

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt contains examples that embed credentials/passwords directly into command-line arguments (e.g., -u '$DOMAIN$USER' -p '$PASSWORD') and placeholders that an agent would likely substitute verbatim, requiring the LLM to handle and output secret values — a high exfiltration risk.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This document is explicit offensive guidance: it details step-by-step network poisoning (ARP/DHCP/DHCPv6/DNS/WPAD/WSUS/ADIDNS), using tools like bettercap/Responder/mitm6/ntlmrelayx/dnstool.py/pywsus to perform MITM, capture NTLM credentials, relay them to AD for privilege escalation (RBCD/ADCS), inject persistent DNS records, and deliver remote code execution—i.e., clear instructions for credential theft, remote code execution, and domain compromise.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs running privileged system modifications (sysctl to enable IP forwarding, iptables SNAT/DNAT rules), starting MITM services that require root (bettercap, mitm6, responder), and even provides payloads to create local/admin accounts on targets—actions that modify the host system state and bypass security controls.