- Persist PROVISIONING before enqueueing, so a fast worker run cannot be
clobbered back to PROVISIONING and stick (the job dedupes, so nothing retried).
- Mark the row FAILED when the worker refuses to provision, instead of leaving
it PROVISIONING forever with no error and no retry.
- Classify provisioning failures and rethrow retryable ones, so the configured
maxAttempts actually applies. A missing owner or org stays permanent.
- Keep the row ARCHIVED when unarchiving fails, so the unarchive is retried
rather than falling through to a path that invites into an archived channel.
- Recheck the entitlement in the worker; an org can lose it while queued.
- Read connected_team_ids from conversations.info — users.conversations does
not return it, so the domain matching hint was always empty.
- Show the Slack Connect invite while INVITED; the deep link 404s until the
owner joins. Drop the duplicated channel name.
- Surface action errors, which were returned but never rendered.
isshaddad
6be17a34ff