An On-Call Handoff Prompt That Turns 48h of Alerts Into a Brief

The worst on-call handoffs are the ones that don't happen. The outgoing engineer types "all quiet, nothing major" into the channel, logs off, and the incoming engineer inherits two snoozed alerts about to un-snooze and a half-finished mitigation nobody wrote down. An on-call handoff prompt fixes the laziest part of the shift: it turns the last 48 hours of alert history into a structured brief so the handoff is the same every time, even when the outgoing engineer is exhausted.

Search this and you'll find best-practice guides telling you what a good handoff contains, or a tool that generates a brief if you live inside its platform. The guides are right about the content. They just leave you to assemble it by hand. A reusable prompt does the assembly from a pasted alert dump, in any model, on any rotation.

Why the handoff keeps failing

A handoff brief has a predictable shape: what's on fire, what's muted, what's mid-flight, and what to keep an eye on. The reason it fails isn't that engineers don't know the shape. It's that writing it well takes 20 minutes at the end of a draining shift, and "all quiet" takes ten seconds.

A prompt collapses that 20 minutes to one. Paste the alert export and your rough notes, and the structure comes back filled. The discipline moves from the tired human to the contract.

The opinionated bit: the most valuable section of a handoff isn't the open incidents, it's the muted alerts. Open incidents are visible by definition. A muted alert is an invisible landmine, and it's exactly what gets forgotten verbally. A handoff prompt that surfaces every snooze and its expiry time earns its place on that section alone.

What you can do with this prompt

• Convert a 48-hour alert export into a ranked list of what actually mattered
• Surface every muted or snoozed alert with its expiry time
• Carry forward open incidents with their current status and next step
• Flag work in progress so the next engineer doesn't undo a mitigation
• Build a short watch list of things that looked shaky but didn't page
• Produce a copy-paste brief for the handoff channel in a consistent format

Anatomy of the handoff prompt

Variables → Prompt → Output

{{alert_history}}      48h of alerts: fired, resolved, snoozed
{{open_incidents}}     anything still active at shift end
{{shift_notes}}        free-text from the outgoing engineer
{{shift_window}}       the time range this brief covers

→ Triage + dedupe + structure instructions →

→ Output contract:
   ## Open incidents (status | next step)
   ## Muted / snoozed alerts (alert | expires)
   ## Work in progress
   ## Watch list
   ## Quiet, no action

This is a template-intent job, so the contract is the product. The five sections are non-negotiable. The model fills each one or writes "none" rather than silently dropping a heading, which is how the muted-alert section quietly disappears in a hand-written brief.

Step-by-step usage

1. Gather inputs

Export the alert history for your shift window. Grab the list of anything still open. Dump your rough notes into {{shift_notes}} without polishing. The prompt polishes.

2. Fill variables

Set {{shift_window}} to the actual range so the brief is dated. Paste raw alert data into {{alert_history}}. Don't pre-filter the noise; the prompt collapses repeats.

3. Run the prompt

Low temperature keeps the same shift producing the same brief structure. You want consistency across handoffs so the incoming engineer reads the same layout every time and knows where to look.

4. Post-process

Check the muted-alerts section against your own memory. If you snoozed something that isn't in the export, the prompt can't surface it, so add it to {{shift_notes}} and rerun. This is the one section worth a manual double-check.

5. Iterate

A brief that's all noise means the alert export had too much low-signal chatter. Tighten what you paste, or add a "collapse alerts that fired and auto-resolved under five minutes" rule to the prompt.

Prompt-craft patterns that matter here

The snooze-expiry extraction. Make the prompt pull every muted alert's un-snooze time into its own column. A snooze that expires during the next shift is the single most useful thing to hand off, and it's the easiest to forget.

For each snoozed or muted alert, extract the alert name and the exact
un-snooze time. Sort by soonest expiry. Flag any expiring within the
next shift window with "EXPIRES THIS SHIFT".

The collapse rule. A noisy alert that fired 40 times becomes one row with a count, not 40 rows. Without this, the brief is unreadable and the signal drowns. State the collapse explicitly so the dedupe is deterministic.

Variables you'll set

Getting started

1. Copy the prompt into your model and set a low temperature.
2. Paste the shift's alert export into {{alert_history}}.
3. Set {{shift_window}} so the brief is dated correctly.
4. Add rough notes to {{shift_notes}}; don't polish them.
5. Run it and verify the muted-alerts section against memory.
6. Post the resulting brief to the handoff channel.
7. For rotations that hand off daily, start from the Engineering On-Call Coordination Agent Pack.

If you hand off once a week, a copied prompt is fine. On a daily rotation, the maintained pack pays for itself in the first month because the snooze-expiry extraction and the five-section contract don't have to be rebuilt by each engineer.

When a handoff item turns into a real incident, the same alert data flows into a writeup, so this pairs with the blameless incident postmortem prompt. For triaging the errors that show up during your shift, the Sentry error triage prompt handles the classification pass, and the Production Error Triage Agent Pack ships that contract ready to run.