conducting-full-scope-red-team-engagement

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). This prompt includes multiple examples that embed plaintext credentials/secrets directly into commands (e.g., domain/user:password in wmiexec/psexec, secretsdump.py, and --secret=s3cr3t), which would require the LLM to generate or handle secret values verbatim and thus creates an exfiltration risk.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The content explicitly prescribes offensive TTPs—C2 setup, payload generation (msfvenom), spearphishing, LSASS dumps, DCSync, Kerberoasting, lateral movement and exfiltration techniques—any of which constitute clear backdoor/credential-theft and data-exfiltration behaviors that can be directly abused for malicious compromise outside an authorized engagement.

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 offensive tooling and techniques that perform privilege escalation (UAC bypass), credential harvesting (Mimikatz/DCSync), payload/C2 deployment and exploitation — actions that alter system state and encourage bypassing security controls on the host.