The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill is entirely dedicated to escalating user privileges from standard user to Administrator or SYSTEM.
Provides instructions for exploiting Service Binary Paths using sc config to redirect service execution to attacker-controlled shells.
Includes methodologies for 'Potato' attacks (JuicyPotato, PrintSpoofer, RoguePotato, GodPotato) to abuse SeImpersonatePrivilege and SeAssignPrimaryTokenPrivilege.
Detailed exploitation of the AlwaysInstallElevated registry setting to install malicious MSI packages as SYSTEM.
Instructions for exploiting unquoted service paths by placing malicious executables in writable parent directories.
[CREDENTIALS_UNSAFE]: The skill facilitates the harvesting of sensitive authentication data.
Commands to locate and extract the SAM and SYSTEM registry hives which contain password hashes.
Specific exploit workflow for CVE-2021-36934 (HiveNightmare) using Mimikatz to read shadow copies of credential hives.
Systematic search patterns for passwords in the registry, configuration files, and unattend.xml files.
Extraction of cleartext WiFi passwords using netsh wlan show profile.
Instructions to read PowerShell history files (ConsoleHost_history.txt) which often contain sensitive commands and passwords.
[REMOTE_CODE_EXECUTION]: The skill describes methods to execute arbitrary code provided by an attacker.
Workflow for generating and installing malicious MSI files using msfvenom and msiexec.
Use of nc.exe (Netcat) to establish reverse shells from the target system to an external IP address.
Guidance on transferring and running third-party offensive binaries such as WinPEAS, Seatbelt, and various Potato exploits.
[COMMAND_EXECUTION]: Extensive use of shell commands (PowerShell and CMD) to probe system architecture, user privileges, network configurations, and active connections.
[DATA_EXFILTRATION]: Provides the capability to extract sensitive system information and credentials, which can then be sent to an external attacker-controlled listener via the reverse shell workflows described.
[DYNAMIC_EXECUTION]: Instructions for runtime compilation of malicious DLLs using Mingw-w64 (x86_64-w64-mingw32-gcc) for DLL hijacking attacks.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests data from local system files (findstr on .xml, .ini, .txt, .config), registry values, and command outputs (e.g., systeminfo passed to wes.py).
Boundary markers: Includes a 'Mandatory confirmation gate' instructing the agent to wait for user authorization before executing dangerous commands.
Capability inventory: Includes service configuration modification, MSI installation, registry manipulation, and network reverse shells.
Sanitization: No explicit sanitization or validation of the ingested file content is mentioned before processing.

windows-privilege-escalation