hunt-ntlm-info
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse NTLMSSP Type-2 challenges from external, untrusted network servers, which creates a potential surface for indirect influence over the agent's context.
- Ingestion points: The
WWW-AuthenticateHTTP header is captured from a target host and processed by the Python script provided inSKILL.md. - Boundary markers: The skill does not implement boundary markers or instructions for the agent to ignore malicious or unexpected data within the parsed AV_PAIRS structure.
- Capability inventory: The skill utilizes network socket operations (
socket,ssl) and binary parsing (struct) within the agent's execution environment. - Sanitization: Input is decoded using Base64 and parsed with
struct.unpack_from. While this enforces a specific binary format, it does not validate the integrity or intent of the metadata values extracted from the remote host. - [COMMAND_EXECUTION]: The skill provides ready-to-execute shell commands and a Python script for the agent to use against external targets.
- A
curlcommand is provided for probing HTTP response headers for NTLM availability. - A Python script is included that performs raw socket communication and SSL wrapping to capture and decode NTLM challenges.
- The script explicitly disables SSL certificate verification (
ssl.CERT_NONE), which is a security weakness that could allow machine-in-the-middle attacks during the reconnaissance phase.
Audit Metadata