Skills
skills/mukul975/anthropic-cybersecurity-skills/implementing-vulnerability-management-with-greenbone/Gen Agent Trust Hub

implementing-vulnerability-management-with-greenbone

Fail

Audited by Gen Agent Trust Hub on Mar 16, 2026

Risk Level: HIGHCREDENTIALS_UNSAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [CREDENTIALS_UNSAFE]: The script scripts/agent.py contains hardcoded default credentials. \n
  • Evidence: Default parameters username='admin' and password='admin' are defined in the connect_gvm function and the argparse CLI configuration.\n- [PROMPT_INJECTION]: The skill processes untrusted scan data, creating a surface for indirect prompt injection. \n
  • Ingestion points: The skill ingests external scan results via the parse_report_xml and analyze_offline_report functions in scripts/agent.py. \n
  • Boundary markers: No delimiters or specific instructions are provided to the agent to treat scan data as untrusted content. \n
  • Capability inventory: The skill has the capability to initiate network connections using the python-gvm library and write report files to the local filesystem. \n
  • Sanitization: There is no evidence of sanitization or validation of the ingested vulnerability data before it is processed or summarized. \n- [EXTERNAL_DOWNLOADS]: The skill utilizes the python-gvm library, which is an external dependency from a well-known source.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 16, 2026, 03:46 PM