privesc-windows
Fail
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: CRITICALPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill implements multiple techniques to bypass security controls, including registry-based UAC bypasses (fodhelper, eventvwr) and token impersonation via the Potato exploit family. The script scripts/uac_bypass.ps1 was explicitly flagged as a Trojan by antivirus and reputation scanners.\n- [COMMAND_EXECUTION]: Several PowerShell and C scripts are included to manipulate system services, registry keys, and process tokens to execute commands with SYSTEM-level integrity.\n- [DYNAMIC_EXECUTION]: The skill includes a C harness (scripts/byovd_loader.c) for loading kernel drivers and a script (scripts/service_hijack_audit.ps1) that generates C source code for proxy DLLs at runtime, facilitating customized exploitation.\n- [CREDENTIALS_UNSAFE]: The documentation in references/credential-harvesting.md provides specific command-line examples for dumping LSASS memory and SAM/SYSTEM hives to harvest plaintext passwords and NTLM hashes.\n- [EXTERNAL_DOWNLOADS]: The instructions encourage the download of various external hacktools (e.g., winPEAS, Seatbelt, GodPotato) from remote sources.\n- [REMOTE_CODE_EXECUTION]: The skill provides recipes for executing remote code directly in memory using PowerShell's IEX and [Reflection.Assembly]::Load methods, fetching payloads from remote network locations.\n- [DATA_EXFILTRATION]: The skill details how to acquire and stage sensitive system databases (SAM, SYSTEM) for offline credential cracking, which is a standard step in data exfiltration and lateral movement.\n- [PERSISTENCE]: Documentation explains how to establish persistence on a host by creating malicious scheduled tasks designed to run as the SYSTEM user.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted system fingerprints and outputs (such as whoami results) and processes them to recommend attack strategies without boundary markers or sanitization, making it vulnerable to malicious data inputs influencing agent behavior.
Recommendations
- CRITICAL: 1 infected file(s) detected - DO NOT USE
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
- AI detected serious security threats
Audit Metadata