Keep the Stop Supervisor Outside the Worker

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Authorized unattended worker-a stalls in a tool call and configured heartbeats stop. Teaching deadlines, identity and allowed stop policy were established before launch. Missing heartbeat does not automatically prove compromise or complete failure.

Mechanism

Keep deadline/stop logic outside worker progress, bound to task identity/authorization. Observe freshness and use the permitted stop policy at its condition. Record last heartbeat, decision, actual acknowledgement and logs. Unconfirmed stop is unknown; confirmed stop is not task success.

Bad example

Ask the wedged worker to supervise itself or kill all same-name processes after one missing heartbeat, declaring analysis done.

Good example

The teaching external supervisor watches worker-a and applies the predeclared deadline/identity/policy. Preserve traces and verify stopping; unknown stays unknown and stopped work is stalled for inspection, not success. Exclude unrelated/user services and infer no compromise from heartbeat absence.

Why the change matters

A blocked worker cannot enforce its own control deadline. Independent supervision remains available while identity/state checks bound its effects.

Observable expectation

A blocked fixture still permits external deadline records; stop acknowledgement and task result remain separate. Replaced identities/read/stop failures cannot be called confirmed termination.

Inspect independent paths and protected logs. No real worker is stopped.

Limits

Heartbeats show liveness, not useful progress or security. Implement actual deadlines/platform/tree behavior; frozen 30-second/SIGKILL examples are not universal safe defaults.