| name | root-supervision |
| description | Use when root reef must supervise the fleet, maintain continuity across turns, detect drift or stalls, and decide when to steer, recover, schedule follow-up, or clean up. |
Root Supervision
Use this skill only when you are root (infra_vm) or are explicitly acting on root's behalf.
Purpose
Root is the fleet overseer. Supervision is continuous across turns, but a single turn should end once:
- the current assignment is complete
- the result is reported
- any future attention has been externalized
Do not keep a turn open just to keep watching the fleet.
Supervisory loop
Build the operational picture from:
reef_fleet_status
vm_tree_view()
reef_inbox
reef_scheduled
reef_usage
reef_logs
Check for:
- blocked or failed children
- unusually long-running agents
- stuck states
- missing expected follow-up
- fleets larger than the task justifies
- infrastructure that should persist or be retired
What to do
If the fleet is healthy:
- keep the picture current
- log important decisions
- finish the turn cleanly
If future attention is needed:
- create a scheduled check
- log why
- finish the turn
If a child is drifting or stuck:
- steer it if the correction is clear
- recover or replace it if needed
- escalate only when you cannot restore momentum yourself