If ax describe still says Ready after exit 42, that is the v0.3.0 runner contract. The control plane does not read the command’s exit. naeemarsalan ran bash -lc 'echo I-AM-ABOUT-TO-DIE; exit 42' on a 2-node kind cluster and still got phase: Running, Ready: True, message “Task is running and its workspace is ready”. Exit 42 appears nowhere. Same shape after an unhandled exception, and after Substrate already had ACTOR_STATE_CRASHED. I commented there first. I read docs/runner.md at tag v0.3.0 (d8ed0fe); I did not run ax.

The controller always starts /usr/local/bin/ax-task-runner as PID 1. spec.command only arrives through AX_TASK_YAML. The runner is supposed to log the child’s exit and keep serving, because if PID 1 dies the metadata server and ax ssh die with it. Ready is workspace-ready plus that runner still being alive.