security-operations

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow ingests and analyzes incident reports and “reported email” content (e.g., phishing headers/body/attachments) from external users via the “Phishing Email Response Playbook” trigger (“User reports suspicious email via … or to security@company.com”), which can include outsider-authored free text/HTML/headers that the agent would read to triage and document.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill contains explicit privileged and destructive commands (iptables rules, usermod/Disable-ADAccount, apt/yum upgrades, reg delete, scheduled-task deletion, EDR containment API calls, mail purge actions, etc.) that modify system and network state and require elevated/sudo or administrative access, so it pushes the agent to change the host environment.