zabbix

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). The SKILL.md's "Proxy requests" and "Running actions" workflows instruct the agent to call arbitrary endpoints via "membrane request CONNECTION_ID /path/to/endpoint" and to consume action output results, meaning the agent will fetch and interpret content from the connected Zabbix instance (third‑party API/web content) as part of its decision-making.