Grok Bot: status on Auto-review errors and related platform issues (Oct 29–31)

Support Bug Reports
, , ,
1

Where does the bug appear (feature/product)?

Grok Bot

Describe the Bug

Hi, I run several Grok Bots (UTC+7) for monitoring and alerting, and we’re hitting the platform issues below. Could you share the current status, a fix timeline, or a recommended workaround for each?

  1. Auto-review classifier errors block actions. Scheduled routine scripts fail with “An error occurred while classifying this action. Please review manually.” No approval card appears, and per support, retrying won’t help. Allow rules don’t apply either, because the check itself fails. We saw this during the Oct 29 04:15–06:13 outage and the Oct 30 06:44–08:20 Auto Review degradation, and others report it continuing after those were marked resolved. Is there an ETA? And when the check errors on a scheduled run, can the action go to an approval card instead of being dropped?
  2. Scheduled routine runs fail server-side or are missed. A routine run failed server-side on Oct 30 at 23:38, outside any status-page incident, and the following runs were missed. Missed runs aren’t replayed. Is a way to retry or replay failed runs, or to get missed-run alerts, on the roadmap?
  3. A routine run can’t update a different routine. We get “routine runs can only update themselves”, so a run can’t reschedule a companion timeout routine. Is that intended, and is there a supported way to do it?
  4. Outbound connections other than port 443 are unreliable from the cloud computer. A Tailscale mesh client on the box falls back to relays, its connections drop, and it often needs a restart after the box sleeps or resets. Is there a fix coming for UDP and non-443 traffic, or a supported way to reach a home machine?
  5. The cloud computer sleeps and doesn’t wake for its own scripts. We understand that’s by design. Is an always-on option or a scheduled wake planned?
  6. Storage mounts late after a reset. After a reset, the bot’s persistent store mounted late, so startup checks failed. Is that a known issue?

Thanks. Happy to share timestamps or ticket details privately.

Steps to Reproduce

See the description above; this is a status request covering several ongoing platform issues rather than a single reproducible bug.

Operating System

MacOS

Version Information

Not collected

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

5

Hey @Vu_Le1, thanks for the clear list. In order:

  1. Auto-review errors: Both windows were incidents on our side, both are resolved, and these errors are rare again. When the check itself errors, the step is blocked with no card, and Allow rules can’t apply. Showing an approval card instead is something we’re tracking, with no timeline yet. During an outage, turning Auto-review off gets actions through, but it removes the check for all your Bots, so only do it briefly.
  2. Failed and missed runs: Failed runs aren’t retried, missed times aren’t replayed, and there’s no missed-run alert yet (it’s a request we’re tracking). Rerun with Test in the routine’s details, and for alerting, have each run ping an external heartbeat check over HTTPS. If you DM me the routine name and what you saw at 23:38, I’ll take a look.
  3. Updating another routine: That’s intentional. Ask the Bot in its main chat to change the companion routine, or give the timeout routine a fixed schedule that checks a timestamp the main routine writes.
  4. Non-443 traffic: It’s an issue we’re tracking, with no timeline, and I can’t promise UDP. A Tailscale client on the computer will keep falling back to relays and dropping, so for your home machine, use what you already have: the Grok Bot desktop app on your Mac mini, open and allowed to run tasks (Approvals, security, and privacy | SpaceXSI Docs)
  5. Sleep: There’s no always-on option today; it’s a request we’re tracking. Routines are the supported scheduled wake (as often as every 5 minutes), so move periodic checks there and keep intervals as long as you can, since each run counts toward usage.
  6. Storage after a reset: That’s an issue we’re tracking. Fully quit and reopen the desktop app (or open it on your phone) to bring the store back, have startup checks retry instead of failing, and skip Reset, since it won’t help.