Launched, claimed and unmanaged work
Fleet Console distinguishes work the controller launched from work it merely knows about. A row in the registry is not sufficient evidence of signalling authority.
Three management classes
| Class | How the record arises | Controller control |
|---|---|---|
| Controller-launched | A new process starts through fleetctl launch | Verified service-control actions and lease cleanup |
| Claimed | An existing process is recorded with fleetctl claim | Observation and attribution, not signalling |
| Unmanaged | Work was started outside controller supervision | No controller stop, restart, pause or kill |
A claim adds an owner, purpose, lease and event timeline. It does not transform an already-running process into one the controller started. An expired claim is reported and left alone rather than reaped. Output from a claimed process still goes where it was originally launched; claiming cannot retroactively capture its stdout or stderr.
Inspect without changing ownership
With the bundled CLI on PATH, inspect the current inventory and its recorded grouping:
fleetctl list --json
fleetctl agents --json
Review output locally before sharing it: commands, paths, ports and agent names can identify private work. An agent label reports who requested or was assigned the work; it does not prove that the human owner approved a disruptive action.
Why PID identity matters
A PID can be recycled after a process exits. Before signalling controller-launched work, the controller verifies the PID, recorded start time and process group. A refusal is not an invitation to bypass the guard with a raw signal. Reinspect the current process and service record instead.
The CLI documentation explains why claiming buys visibility without authority. Use service logs to review what the controller actually recorded.