The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes 'AI LOAD INSTRUCTION' blocks that attempt to force the agent into a specialized 'Expert Attack' persona. This pattern is designed to override default safety guardrails by instructing the model to disregard perceived constraints in favor of offensive security expertise.
[COMMAND_EXECUTION]: The skill provides a comprehensive list of shell commands (CMD and PowerShell) for exploiting service misconfigurations, unquoted paths, and writable registry keys to achieve unauthorized privilege elevation.
[REMOTE_CODE_EXECUTION]: The playbook instructs the agent to download and run third-party attack binaries (e.g., GodPotato, PrintSpoofer) and describes techniques like CMSTP abuse to execute remote scripts (e.g., .sct files) from external servers.
[DATA_EXFILTRATION]: Includes instructions for dumping LSASS memory and sensitive registry hives (SAM, SYSTEM, SECURITY) to extract user credentials for further exploitation.
[CREDENTIALS_UNSAFE]: The skill targets the discovery of stored secrets by searching for private keys, API tokens, and credentials in the Windows Registry and filesystem using 'cmdkey' and manual path traversal.
[EXTERNAL_DOWNLOADS]: Directs the agent to fetch various exploitation tools and scripts from public repositories and attacker-controlled infrastructure.

windows-privilege-escalation