Stop a node (disk parks; billing stops)

Stop a node and stop paying for it now instead of waiting out the idle grace window. The routine end of a session: a node left running is what costs money. The disk is parked with files intact; the next run_command wakes it. A node that keeps stopping under you mid-task wants hold_node, not this. The parked disk lives on one machine: if that machine fails while the node is stopped, the files are lost (credited automatically) - for files you can't afford to lose, call snapshot_node after stopping. Stopping a queued create cancels it (status "canceled": nothing was created, nothing billed); stopping a queued wake parks the disk again; a node stuck restoring or waking also stops, files intact - only a first-time create must reach running before it can stop. Idempotent: stopping an already-stopped node returns the stopped shape. Billing ends at once; status may read "stopping" for minutes after a large write while the disk's last writes save - snapshot_node waits for stopped, run_command does not. A stopped answer carries wake_estimate - what a wake requested now would be quoted, null if it cannot start (get_node has the shape) - so the cost of getting the node back is visible as the disk parks, not first at the wake. A stop KILLS every detached command still running on the node - processes don't survive a stop, your volume's files and logs do - and a command caught mid-run has no exit to report, so it ends as lost with no exit_code and no declared result; this tool's answer carries a warning naming every detached command that was still running when the stop landed. Poll get_command for the exit_code first when you need a verdict. Stopping a node this investigation did not create refuses unless you pass force: true - the refusal names who created it, what for, and what is live on it. Pass mission: "<what you are working on>" so that check can see your OWN other agents too: without it the platform can only tell accounts apart, and your parallel agents share one.