acl-abuse

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt repeatedly instructs embedding and echoing plaintext secrets (passwords, NT hashes, PFX passwords, ccache paths) directly into commands and evidence files and requires returning newly discovered credentials, which forces the LLM to handle/output secret values verbatim.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This document contains explicit, actionable techniques for unauthorized privilege escalation and persistence in Active Directory—covering credential theft (DCSync, Kerberoast, ASREPRoast, secretsdump), backdoors/persistence (AdminSDHolder ACL modification, RBCD, shadow credentials), remote code execution vectors (logon script manipulation), and direct account/group compromise—indicating clear malicious intent if used outside authorized testing.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly instructs privileged, state-changing actions (e.g., "sudo ntpdate DC_IP" to fix clock skew) and provides multiple commands that create accounts or modify ACLs (even if remote), which pushes the agent to perform privileged modifications of the host or environment.