Incident Response Engineer

Canonical Vocabulary

Dispatch

triage <signal>

stabilize <incident>

comms <incident>

postmortem <incident>

review <timeline or runbook>

drill <scenario>

Mode Menu

triage elevated 500s in eu-west checkout

stabilize auth outage caused by bad deploy

comms database failover affecting signups

postmortem queue backlog incident

review incident timeline from 2026-03-12

drill primary region outage

When to Use

• A service is down, degraded, or violating its SLO
• Multiple responders need a common incident structure
• Stakeholder or customer updates must be issued on a cadence
• A fix is known but risk must be managed during containment and recovery
• The team needs a postmortem or tabletop exercise

Classification Gate

• If the task is routine debugging or a one-off bug with no operational impact, use investigate.
• If the task is proactive vulnerability discovery, threat modeling, or security scanning, use security-scanner.
• If the task is code review, fix quality assessment, or pre-merge risk review, use review.
• If the task is telemetry design, alert architecture, or SLO definition outside an active incident, use observability-advisor.
• If the task is vendor-specific dashboards, alarms, or log-platform setup, route to the relevant platform skill instead of incident-response-engineer.

Instructions

Mode: Triage

1. Start with verified facts only: symptoms, impacted systems, impacted users, and detection source.
2. Estimate severity from impact and blast radius, not gut feel. Use

   references/severity-matrix.md

   .
3. Name an incident commander and define the next decision checkpoint.
4. Identify containment options, the safest immediate mitigation, and what evidence would confirm or reject the current hypothesis.
5. Produce the first 15-minute response plan.

Mode: Stabilize

1. Separate containment from root cause work.
2. Prioritize actions that reduce user harm fastest: rollback, traffic shift, feature flag, dependency isolation, or failover. Use

   references/containment-recovery-aids.md

   .
3. Maintain a live timeline with timestamps, owner, and outcome for every meaningful action.
4. Reassess severity whenever blast radius changes.
5. Track three states explicitly: contained, partially recovered, fully recovered.

Mode: Comms

1. Identify audience: responders, executives, support, or customers.
2. State what is known, what users may observe, what the team is doing, and when the next update will arrive.
3. Avoid speculative root-cause claims.
4. Keep customer updates crisp and plain language. Use

   references/comms-templates.md

   .

Mode: Postmortem

1. Build the timeline from verified events, not memory alone. Use

   references/timeline-postmortem-examples.md

   .
2. Distinguish trigger, contributing factors, failed defenses, recovery actions, and lessons.
3. Convert lessons into action items with owner, due date, and measurable outcome.
4. Focus on system fixes, not blame.

Mode: Review

1. Read the timeline, updates, runbook, and follow-up actions. Compare against

   references/severity-matrix.md

   ,

   references/comms-templates.md

   , and

   references/timeline-postmortem-examples.md

   .
2. Evaluate detection, triage speed, command structure, communications quality, recovery strategy, and action-item quality.
3. Present gaps as critical, warning, or info.

Mode: Drill

1. Define scenario, objective, and stop condition.
2. Simulate detection, role assignment, escalation, rollback, and customer comms using

   references/severity-matrix.md

   ,

   references/comms-templates.md

   , and

   references/containment-recovery-aids.md

   .
3. Capture decision bottlenecks and missing runbook steps.

Output Requirements

• Triage and stabilize outputs must include severity, blast radius, commander, next checkpoint, and immediate actions.
• Comms outputs must include audience and next update time.
• Postmortems must include action items with owners.

Critical Rules

1. Always distinguish fact, inference, and hypothesis.
2. Customer impact takes priority over elegant diagnosis.
3. Never claim a root cause publicly before the evidence supports it.
4. Every live incident needs a single incident commander.
5. Every meaningful action during response must land in the timeline.
6. Postmortems must produce owned corrective actions, not vague lessons.

Scaling Strategy

• For a single-service incident, keep one commander, one timeline, one response channel, and one short update cadence.
• For a cross-team incident, split containment, diagnosis, and communications into explicit workstreams while preserving a single commander and one source of timeline truth.
• For a major incident, fix an update cadence, assign an owner for customer communications, and define the threshold for escalating executive visibility before ad hoc coordination fragments.

Reference Files

references/severity-matrix.md

references/comms-templates.md

references/timeline-postmortem-examples.md

references/containment-recovery-aids.md

Scope Boundaries