acl-abuse

Warn

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill provides detailed instructions and commands for escalating privileges within an Active Directory environment. Key techniques include modifying group memberships (e.g., Domain Admins), granting DCSync rights for credential harvesting, and taking ownership of directory objects.
  • [PERSISTENCE_MECHANISMS]: The skill implements a persistence mechanism via the AdminSDHolder template object. By adding a GenericAll ACE for an attacker to this object, the permissions are automatically propagated to all privileged accounts in the domain every 60 minutes.
  • [COMMAND_EXECUTION]: The skill relies on executing various powerful command-line tools such as bloodyAD, Impacket (secretsdump, getTGT, GetUserSPNs), and Certipy to perform sensitive operations like password resets and ACL modifications.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill generates and handles sensitive credential materials, including Kerberos ccache files, PFX certificates, and password hashes, saving them to the local filesystem (e.g., $TMPDIR and the engagement/evidence directory).
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from Active Directory attributes and uses it to drive subsequent high-privilege actions without explicit boundary markers or output sanitization.
  • Ingestion points: Data enters the agent context via output from enumeration tools like bloodyAD (get writable/object), PowerView (Invoke-ACLScanner), and BloodHound Cypher queries.
  • Boundary markers: None. The skill does not instruct the agent to use delimiters or ignore instructions embedded in the directory attributes it reads.
  • Capability inventory: The skill has extensive write access to Active Directory, the ability to reset passwords, and the ability to configure logon scripts (RCE surface) through tools like bloodyAD and dacledit.py.
  • Sanitization: None. The skill does not perform any validation or filtering of the attributes retrieved from the domain before acting upon them.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 6, 2026, 09:16 AM
Security Audit — agent-trust-hub — acl-abuse