External Network Penetration Testing

Warn

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The instructions direct the agent to execute multiple third-party scripts that are not provided within the skill package, including 'linkedin2username.py', 'dnscan.py', 'ct-exposer.py', 'cloud_enum.py', 'CloudUnflare.py', 'credmaster.py', and 'MSOLSpray.py'.
  • [COMMAND_EXECUTION]: The skill provides a wide array of shell commands for reconnaissance and exploitation, including network scanning with Nmap and Masscan, OSINT gathering, and multi-service password spraying.
  • [EXTERNAL_DOWNLOADS]: The skill methodology points to external code repositories for additional tooling, such as 'carlospolop/Pastos' and 'ustayready/fireprox' on GitHub.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, untrusted sources such as DNS records, certificate transparency logs, and web archives, creating an attack surface where an attacker could influence the agent's behavior.
  • Ingestion points: External data from crt.sh, Shodan, Censys, and Wayback Machine enters the agent context.
  • Boundary markers: None identified in the workflow instructions.
  • Capability inventory: Subprocess execution for numerous scanning, reconnaissance, and exploitation tools.
  • Sanitization: No explicit sanitization or validation of external input before processing or command interpolation is described.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 6, 2026, 03:30 PM