Template skills not importing? Explore setup prompts and check the result before running a task.See why

Why doesn't my Grok Bot wake up when a background Shell command finishes?

Last checked October 29, 2026 · Grok Bot changes often

Short answer

On October 29, 2026 staff said not to count on a Local Execution background command to wake a Bot after its turn ends; it is a tracked issue. For short jobs, make the Bot wait in the same turn with a block_until_ms that covers the job, or an explicit wait on the command. For long jobs, keep your own check or message the Bot when the job finishes and it will pick up the result.

What was reported

A user ran long commands through Local Execution Shell with block_until_ms: 0, so the command runs in the background, then ended the Bot's turn. The command often exited successfully, but the Bot's chat was not woken: no toast and no automatic continue. They measured roughly 40% silent misses on a busy day, with late wakes 6 to 16 minutes later. Cloud agent wake worked. An older thread from July 2026 described a related "error resuming chat" case.

What staff said

  • Do not count on a Local Execution background command to wake the Bot after its turn ends.
  • Short jobs: keep the Bot in the same turn with a block_until_ms that covers the job, or an explicit wait on the command. Staff suggested adding this to the Bot's instructions.
  • Long jobs: keep a reconcile check, or message the Bot when the job finishes.
  • The setup was fine; the problem is tracked, and a completion-hook idea was passed on.

A simple pattern

Put one line in the Bot's instructions: "For any command I need the result of, wait for it in this turn; do not end the turn while a background command is running." For long work, use a scheduled routine that checks status instead of waiting; see Why didn't my Grok Bot routine run on time?.

Long waits also burn context and usage, so see Why does Grok Bot use up my weekly usage so fast?.

Not tested by Boterly

We have not run background commands on a live Bot. The 40% figure is one user's own measurement, not ours or staff's.

Sources