netops

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

• Third-party content exposure detected (high risk: 0.72). At runtime, the skill’s packet-sensor workflow (e.g., SPAN/RSPAN/ERSPAN to a Linux/Windows sensor) ingests readable network traffic from other parties into the agent’s LLM context via the “Flight Recorder / export bundle” and any “Insights”/LLM analysis step, which can include outsider-authored free text (e.g., HTTP payloads, DNS query names, TLS-decrypted content, or other application-layer strings) captured from non-user sources.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). This skill explicitly instructs modifying system/network state (persistent br0 Linux bridge via netplan/NetworkManager, turning hosts into L3 gateways, ethtool tuning, scripts that require root) and runs many sudo-level commands, so it pushes privileged, state-changing operations that can compromise the host or network.