windows-lateral-movement
Fail
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: HIGHCOMMAND_EXECUTIONDATA_EXFILTRATIONPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides extensive command-line templates for executing code on remote systems using SMB (PsExec), WMI, WinRM, and DCOM. These commands are intended for unauthorized lateral movement across a network.
- [DATA_EXFILTRATION]: Documents comprehensive methods for dumping sensitive information, including LSASS memory, registry hives (SAM, SYSTEM, SECURITY), and the Active Directory database (NTDS.dit), which facilitates the theft of domain credentials and hashes.
- [PRIVILEGE_ESCALATION]: Includes techniques for bypassing Protected Process Light (PPL) for LSASS and acquiring SYSTEM-level privileges using built-in Windows utilities such as vssadmin and reg.exe.
- [PERSISTENCE]: Explains how to maintain long-term access to a compromised host through the use of WMI event subscriptions.
- [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The skill instructions themselves provide the agent with a suite of offensive capabilities that can be triggered by external prompts.
- Boundary markers: None identified; there are no instructions for the agent to verify the authorization of targets or enforce boundaries.
- Capability inventory: Includes remote shell execution, sensitive file extraction, and persistence mechanisms across all files.
- Sanitization: There is no documentation or instruction for validating or escaping input such as IP addresses, domains, or credentials provided to the commands.
Recommendations
- AI detected serious security threats
Audit Metadata