How to do it
- Hit the kill switch first Pause all agents and hold every queued post so no marketing content goes out during the incident.
- Alert to draft Your incident tool sends a signed webhook; an agent drafts a short update with what is affected and when the next update will be.
- On-call approves The on-call communicator approves from the phone swipe stack, editing times and scope as needed.
- Update and resolve Follow-up updates and the resolution post go through the same path, each recorded in the audit log.
- Resume carefully After the incident, review held posts before resuming, and reschedule anything that would feel tone-deaf.
We are investigating failed logins for some users in the EU region since 14:05 UTC. Data is not affected. Next update by 14:45 UTC or sooner. Live status: link to status page.
Incident communication is a trust exercise
Customers forgive outages much more easily than silence or confusion. Good incident updates are short, factual, time-stamped, and consistent across channels. They promise the next update and keep that promise. An agent can draft updates in seconds from alert data; a person must approve, because the wrong scope or cause in public is hard to take back.
Before the next incident
- Connect your company accounts on X, Bluesky, and Mastodon, where customers look for live updates.
- Make sure on-call staff know how to use the kill switch and can approve from the installable PWA.
- Send incident events with a signed webhook from your incident tool, or trigger drafts from an n8n flow.
- Put incident rules in brand memory: always use UTC, never speculate on root cause, always link the status page. Use the incident update template and the maintenance notice template.
Permissions
The incident agent gets draft only, and stays draft-only on the trust ladder. Speed comes from a fast approver, not from skipping approval. Every update is recorded in the append-only audit log, which helps during the postmortem.
After the incident
Review which updates went out and when from the audit log. Post a resolution note and, if appropriate, a bug fix announcement. Before resuming the kill switch, go through held posts and reschedule anything that would land badly the same day.
Pitfalls
- Guessing the cause. Say what you know and when you will update.
- Different stories on different channels. Draft all networks from the same facts.
- Forgetting to resume. Held posts stay held until someone resumes.