network-protocol-attacks
Fail
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous commands for active network disruption and traffic redirection, such as ARP spoofing via
arpspoofandbettercap, and STP root bridge manipulation usingyersinia. It specifically details how to execute remote commands on SQL servers usingntlmrelayx.pywith thexp_cmdshellprocedure. - [DATA_EXFILTRATION]: The skill focuses on capturing sensitive network authentication data. It documents workflows for harvesting NTLMv1 and NTLMv2 hashes using
Responderandmitm6, which are then stored in local logs and can be exfiltrated or used in relay attacks against other services. - [EXTERNAL_DOWNLOADS]: The skill recommends installing third-party packages using
pip install mitm6and utilizes external scripts likefrogger.shfor VLAN hopping, without providing source URLs or integrity verification mechanisms. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted network data, creating a surface for indirect prompt injection.
- Ingestion points: Network traffic responses (DNS, LLMNR, DHCPv6) and authentication blobs captured by security tools.
- Boundary markers: None identified; captured data is processed without delimiters to distinguish it from instructions.
- Capability inventory: The skill has broad capabilities to execute shell commands, perform network operations, and write to the local file system.
- Sanitization: There is no documentation for sanitizing captured network data before it is parsed or used by the agent.
- [PRIVILEGE_ESCALATION]: Many of the described techniques, including interface manipulation (
ip_forward) and raw socket access for tools likearpspoofandettercap, require root or administrative privileges to execute. - [DYNAMIC_EXECUTION]: The skill includes embedded Python code snippets utilizing the
scapylibrary for runtime packet crafting and transmission, such as generating double-tagged VLAN frames.
Recommendations
- AI detected serious security threats
Audit Metadata