The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell commands for offensive security tools such as Impacket, NetExec, and BloodHound to perform network exploitation and credential harvesting.
[DATA_EXFILTRATION]: Instructions facilitate the extraction and local storage of sensitive Active Directory data, including NTDS.dit database dumps, SAM hashes, and LSASS memory contents.
[PRIVILEGE_ESCALATION]: The skill requires the use of sudo to synchronize the system clock, which is necessary for Kerberos authentication but represents an escalation of privileges.
[PERSISTENCE_MECHANISMS]: The skill documents several techniques for establishing domain persistence, such as forging Golden Tickets, backdooring the AdminSDHolder object, and installing Skeleton Keys.
[DYNAMIC_EXECUTION]: A custom Python wrapper script (nxc_kerberos_wrapper.py) is used to dynamically modify the execution environment by shimming the socket and datetime libraries at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from Active Directory objects (e.g., user descriptions and file share contents), creating a surface where malicious input could influence agent behavior. (1) Ingestion points: LDAP enumeration, SMB share spidering, and user metadata harvesting. (2) Boundary markers: No explicit markers are used to isolate untrusted directory content. (3) Capability inventory: The agent has access to powerful shell commands and file system operations via the penetration testing toolset. (4) Sanitization: There is no evidence of sanitization or validation of the data retrieved from the target environment.

active-directory-attack