fleet-hunting-with-velociraptor

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

• Third-party content exposure detected (medium risk: 0.65). The required workflow runs velociraptor query / artifacts collect / hunts, which ingest and return endpoint-derived free-text fields like CommandLine from the monitored clients’ process state (outsider-authored by other users/processes on those endpoints), and that returned text is streamed into the agent/LLM context via the CLI output/results.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs using sudo to install the Velociraptor binary into /usr/local/bin and to install/run agents as services and enroll clients, which modify system files and create persistent, privileged services on hosts.