windows-credential-harvesting

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly instructs locating, decrypting, and returning plaintext credentials, passwords, hashes, and key material (including embedding passwords/NTLM hashes/hex keys into commands and saving discovered secrets), which requires the LLM to handle and output secret values verbatim.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This document contains explicit, actionable instructions for harvesting Windows credentials (files, registry, DPAPI, LSASS, SAM), decrypting and exfiltrating secrets (mimikatz, SharpDPAPI, shadow copy extraction, copying to attacker shares), and techniques that enable credential theft, remote payload execution, and data exfiltration — indicating deliberate malicious intent or dual-use offensive capabilities requiring strict authorization.

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 the agent to perform local privileged actions (privilege escalation/UAC bypass, reg save, mklink shadowcopy mounts, copying SAM/SYSTEM, dumping LSASS/mimikatz, runas /savecred, and writing evidence files) that modify system state and seek to obtain or use elevated credentials, so it actively pushes compromise of the host.