Threat model

Rookhold assumes job code can be malicious. The strong deployment keeps the service credential outside the model, denies network inside each workload, uses a private root filesystem, applies cgroup and rlimit ceilings, removes capabilities, bounds output, and records the controls that became effective.

The outer service remains privileged and belongs on a dedicated VM. Read the complete security boundary and SECURITY.md.