windows-token-impersonation

Fail

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill's primary function is to escalate privileges to NT AUTHORITY\SYSTEM by exploiting various Windows token privileges (e.g., SeImpersonate, SeDebug, SeBackup).
  • It provides instructions for abusing system tools like schtasks and icacls to bypass security controls.
  • It facilitates the use of "Potato" exploits to impersonate service accounts.
  • [REMOTE_CODE_EXECUTION]: The skill explicitly instructs the agent to download and execute arbitrary binaries and scripts from remote, attacker-controlled servers.
  • Examples include downloading GodPotato-NET4.exe, nc.exe, and shell.ps1 via PowerShell iwr.
  • It also includes the deployment of ASPX webshells for command execution on IIS servers.
  • [DATA_EXFILTRATION]: The skill targets highly sensitive system files and credential stores for extraction.
  • It includes commands to save and exfiltrate the SAM, SYSTEM, and SECURITY registry hives using reg save.
  • It provides methods for reading raw disk sectors to bypass NTFS ACLs and access files like ntds.dit.
  • [PERSISTENCE]: The skill describes techniques for maintaining access, such as replacing the utilman.exe system binary with cmd.exe to create a backdoor at the login screen.
  • [DYNAMIC_EXECUTION]: It utilizes PowerShell to load .NET assemblies directly into memory ([System.Reflection.Assembly]::Load), a technique often used to evade file-based detection.
  • [INDIRECT_PROMPT_INJECTION]: The skill consumes data from an external state management server (get_state_summary), creating a potential surface where untrusted data from a target environment could influence agent behavior.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 6, 2026, 09:15 AM
Security Audit — agent-trust-hub — windows-token-impersonation