War Room Protocol

War room protocol

A war room concentrates the people and authority to end a crisis into one channel with one commander. Without protocol it becomes a crowded room where twenty engineers narrate the same graph, no one owns the next action, and the decisions that mattered are lost the moment the page clears. The protocol exists to make coordination faster than the incident.

Method

1. Name an incident commander first, before any debugging. The commander

runs the response, not the keyboard: they assign work, hold the timeline, and are the single point of decision. Everyone else defers to them on priority. If two people think they are in charge, no one is.

1. Fill the three other roles explicitly. An operations lead who drives the

technical fix, a scribe who owns the log, and a communications lead who updates status page and stakeholders. Say the names out loud in the channel so the room knows who does what.

1. Set a fixed sync cadence and hold it. Every 15 or 30 minutes by severity,

the commander calls a checkpoint: current impact, what changed, next action, next check-in time. Between syncs people work, they do not narrate. The cadence is what stops the endless scroll of unstructured chatter.

1. Keep a running decision log with timestamps. The scribe records every

decision, who made it, why, and what was ruled out. "14:32 rolled back deploy 4471, error rate flat, ruling out config" is a log entry. This is your real-time record for the postmortem and your defense against relitigating a call at hour three.

1. Separate mitigation from root cause, and chase mitigation first. Stop the

bleeding: roll back, fail over, shed load, flip the flag. The fix that explains why can wait until users are served again. A war room that debugs before it mitigates is optimizing the wrong clock.

1. Declare stand-down against written criteria. Impact resolved, metrics back

in normal range for a defined window, no active mitigations holding it up by hand. The commander calls it, names the postmortem owner and date, and closes the channel. An incident that never formally ends never gets learned from.

Signals

• Can any participant name the incident commander without scrolling up?
• Does the decision log let someone joining at hour two catch up in two minutes?
• Is stand-down a stated threshold, or a slow fade as people drift off?

Boundaries

This covers running the live response, not the retrospective that follows: route the after-action to a postmortem skill, which owns blameless analysis and follow-up tracking. Severity thresholds and paging policy are company convention; match your on-call runbook rather than inventing a scale mid-crisis.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Amey-Thakur
• Source: Amey-Thakur/AI-SKILLS
• License: MIT
• Homepage: https://amey-thakur.github.io/AI-SKILLS/

Install and usage instructions live in the source repository linked above.