monitor-agentslisted

# Monitor agents A native subagent has no runtime state to poll: a foreground call returns its result, and a background call notifies the session when it completes. Tracking a worker means waiting for that return, then reading the evidence it left behind. ## When to use - After delegating a task, to wait for the worker. - When the developer asks what a worker is doing, or a worker seems stalled. ## When not to use - Once a worker has returned and you already have its report: move to `review-work`. ## How a subagent settles | Outcome | Meaning | | --- | --- | | returned | The subagent tool call completed. **Ready, not correct.** | | returned a question | The worker is blocked on a decision; escalate. | | errored | The call failed; inspect, then re-delegate or escalate. | | no completion | A background subagent that never notifies, or an abandoned foreground call. Unresolved. | There is no `working`/`idle`/`done`/`blocked`/`unknown` state to poll: the harness reports completion by returning or notifying, and the worker's own `.teammate-report.md` is the evidence. A returned subagent is ready for review, not proof of success. ## Wait without going silent A long foreground call blocks the primary and leaves the developer without a point of contact, which defeats the reason Team Mate stays free. A worker run is minutes to tens of minutes, so never sit inside one. - **Let the harness wait.** A foreground subagent call returns when the worker finishes; a background su