network-sniffing-mitm

Warn

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: MEDIUMPERSISTENCEPRIVILEGE_ESCALATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [PERSISTENCE]: The skill documents techniques for maintaining access and capture capabilities on a network. Specifically, in SKILL.md, the 'Practical Steps' section suggests establishing persistence by dropping a 'passive backdoor' (such as a pcapd cron job) to facilitate long-duration traffic interception.
  • [PRIVILEGE_ESCALATION]: Numerous commands and guides within the skill require administrative or root-level access. Operations such as active network poisoning, raw packet capture, and system-level configuration changes (e.g., enabling IP forwarding via /proc/sys/net/ipv4/ip_forward) are typically executed using sudo or require elevated privileges.
  • [DYNAMIC_EXECUTION]: The skill makes extensive use of scriptable frameworks for automation and traffic modification. This includes Bettercap 'caplets' and custom Python scripts for mitmproxy (e.g., log_creds.py and replace_downloads.py) which are executed at runtime to process and manipulate intercepted network data.
  • [INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and analyzing untrusted network traffic from external sources. Because the agent processes this external data and has access to powerful tools like Bash and Write, there is a potential attack surface where malicious network responses could be used to influence the agent's actions or logic.
  • [DATA_EXFILTRATION]: The skill's primary focus is the interception of sensitive information, including NTLM hashes, session cookies, and cleartext credentials from protocols like FTP and HTTP. While intended for security testing, these techniques demonstrate the ability to harvest and document sensitive data from target environments.
  • [COMMAND_EXECUTION]: The skill provides numerous complex shell commands and scripts for network manipulation. These include commands that interact with system interfaces and network stacks, which could be misused if the agent is directed to execute them without proper oversight.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 19, 2026, 04:36 PM