The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions provide specific commands for executing high-impact security tools including Mimikatz, Impacket (psexec, wmiexec), and CrackMapExec. These tools enable remote code execution and administrative control over target systems within a network environment.
[PRIVILEGE_ESCALATION]: The workflow details a systematic process for escalating privileges from a standard domain user to a Domain Administrator. This includes techniques like Kerberoasting, AS-REP Roasting, and abusing Active Directory Certificate Services (AD CS) templates to impersonate privileged users.
[CREDENTIALS_UNSAFE]: The included Python script scripts/agent.py is designed to accept a domain user password as a plaintext command-line argument (--password). This method of credential handling is insecure as passwords can be exposed in process lists, shell history, and system logs.
[DATA_EXFILTRATION]: The skill describes methods for harvesting sensitive domain data, such as dumping the entire Active Directory database using the DCSync technique (impacket-secretsdump). This allows an attacker to extract NTLM hashes for every account in the domain.
[INDIRECT_PROMPT_INJECTION]: The agent.py script retrieves and processes untrusted data from an Active Directory environment via LDAP queries without sufficient sanitization or boundary markers.
Ingestion points: User attributes and group memberships are fetched from a Domain Controller in scripts/agent.py.
Boundary markers: The script lacks delimiters or instructions to prevent the agent from interpreting retrieved LDAP strings as instructions.
Capability inventory: The skill possesses capabilities for network operations, file writing, and instructions for administrative command execution.
Sanitization: While the script parses UAC flags, it does not sanitize or validate string-based user attributes before including them in the final JSON report.

executing-active-directory-attack-simulation