On-call handover between shifts

DevOps, SRE and securityOn-call rotation across two shifts

A queue backlog started at 4 a.m. Our night on-call engineer's agent tried restarting consumers and scaling to 12 replicas; neither helped. At 9 a.m. the day engineer opened a new session and the agent suggested restarting consumers first.

How does the day on-call engineer's AI agent see what the night shift already tried?

Have the night on-call agent save each failed step with ArcRoom's remember_failure and the reason, and hold the open question with claim_task. The day engineer's agent then sees who owns it, and find_task returns the failed steps, so they aren't suggested again. Only what the night agent saved is handed over. Live alert state stays in the paging tool.

Also asked as: Our day shift's agent suggested the same fix that failed overnight. How do we stop that? · How do I hand over an incident between on-call shifts when both use AI agents? · Can an AI agent know what another engineer's agent tried during an outage?

The work

The night engineer's agent saves each failed step with remember_failure and the reason.

ArcRoom keeps
  • Last-run note
  • Reused answer
  • Claim
  • Saved answer
  • Saved decision
The next agent gets

claim_task → already claimed by ravi (claude, 5 h ago): "orders queue backlog"

Step by step

  1. The night engineer's agent saves each failed step with remember_failure and the reason.
  2. The open question is held with claim_task, so the day shift's agent sees who owns it.
  3. The day engineer's new session gets a last-run note if it opens in the same folder, and find_task returns the failed steps either way.
  4. When the cause is found, complete_task saves it for the next incident.

What the agent sees

claim_task → already claimed by ravi (claude, 5 h ago): "orders queue backlog"
Failed: restart consumers (backlog grew) · scale to 12 replicas (no change)
Open: check the poison message on partition 7

Compared with today's tools

Incident channel + handover note in the paging tool

The outgoing engineer writes a handover in Slack or the incident timeline.

Setup
Already in place
Ongoing
write a handover at 9 a.m. after a long night, ~10 min

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
2 min per person (npx arcroom init)
Ongoing
None

The paging tool stays the incident record; ArcRoom puts the night's dead ends inside the day agent's session without a tired person writing them up.

Where it stops

Only what the night agent saved is handed over. Live alert state stays in the paging tool.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in devops, sre and security