The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides specific command sequences for executing remote PowerShell scripts using the Invoke-Command cmdlet in conjunction with Net.WebClient and iex (Invoke-Expression). This is a well-known dropper pattern for executing remote payloads directly in the target's memory.
[COMMAND_EXECUTION]: Documentation includes extensive use of offensive security tools such as Evil-WinRM, crackmapexec, and NetExec (nxc) to perform remote command execution and system enumeration.
[EXTERNAL_DOWNLOADS]: The skill instructions link to various external tools and repositories on GitHub, including evil-winrm, pypsrp, and SharpWSManWinRM. It also references a third-party Docker container quickbreach/powershell-ntlm as a recommended execution environment.
[DYNAMIC_EXECUTION]: The skill describes using Evil-WinRM to perform in-memory execution of .NET binaries via the Invoke-Binary feature, which circumvents traditional file-based detection mechanisms by loading executable code directly into the process space.
[PRIVILEGE_ESCALATION]: The methodology covers the exploitation of CVE-2021-38647 (OMIGOD), a vulnerability in the Azure Open Management Infrastructure (OMI) that allows unauthenticated remote code execution with root privileges.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection.
Ingestion points: The skill ingests data from external network discovery tools such as Nmap and Test-WSMan (SKILL.md).
Boundary markers: No boundary markers or instructions to ignore embedded instructions in the ingested data are present.
Capability inventory: The skill possesses extensive capabilities for remote command execution and system modification via tools like Evil-WinRM, PowerShell Remoting, and crackmapexec (SKILL.md, references/winrm-techniques.md).
Sanitization: No sanitization or validation of the input received from external tools is implemented before it is used in subsequent operations.

winrm-pentesting