incident-commander（事件指挥官）

Incident Commander

The agent classifies incident severity, reconstructs timelines from heterogeneous event sources, and generates structured post-incident reviews with root cause analysis and action items.

Quick Start

# Classify an incident (JSON or stdin)
echo '{"description": "Database connections timing out", "affected_users": "80%", "business_impact": "high"}' \
  | python scripts/incident_classifier.py --format text

# Multi-dimensional severity scoring
python scripts/severity_classifier.py incident.json --format markdown

# Reconstruct timeline with phase detection and gap analysis
python scripts/timeline_reconstructor.py --input events.json --detect-phases --gap-analysis --format markdown

# Build structured timeline with MTTD/MTTR metrics
python scripts/incident_timeline_builder.py incident_data.json --format markdown

# Generate Post-Incident Review
python scripts/pir_generator.py --incident incident.json --rca-method fishbone --action-items --format markdown

# Generate postmortem with benchmark comparisons
python scripts/postmortem_generator.py incident_data.json --format markdown

Tools Overview

Workflow 1: Incident Response (Detection to Resolution)

Step 1 -- Classify severity.

python scripts/severity_classifier.py incident.json --format json

The agent scores across five dimensions: revenue impact (25%), user scope (25%), data/security risk (20%), service criticality (15%), blast radius (15%).

Validation checkpoint: Severity classification includes confidence score and recommended escalation path.

Step 2 -- Establish command.

The Incident Commander:

• Assigns within 5 min (SEV-1) or 30 min (SEV-2)
• Creates war room and incident tracking ticket
• Sends initial notification using generated template
• Coordinates between technical teams and stakeholders
• Shields responders from external distractions

Step 3 -- Investigate and mitigate.

The agent generates targeted investigation commands based on the affected service:

kubectl get pods -n production -l app=<service>
kubectl logs -l app=<service> --tail=100
helm history <service> -n production

Decision framework for SEV-1/SEV-2:

• Bias toward action over analysis
• Prefer rollbacks to risky fixes under pressure
• Document every decision for later review
• Consult SMEs but do not block on them

Step 4 -- Communicate.

The agent generates three communication templates per severity:

1. Internal notification -- technical details, response team, war room link
2. Executive summary -- business impact, ETA, leadership actions required
3. Customer communication -- impact scope, what is being done, next update time

Validation checkpoint: All stakeholders notified within committed timeframes.

Workflow 2: Post-Incident Review

Step 1 -- Reconstruct the timeline.

python scripts/timeline_reconstructor.py --input events.json --detect-phases --gap-analysis --format markdown

The agent accepts events from logs, alerts, Slack messages, and deployment systems. Each event needs a timestamp and description. Optional fields: source, type, actor, severity.

Supported phases: detection, declaration, escalation, investigation, mitigation, communication, resolution.

Step 2 -- Perform root cause analysis.

python scripts/pir_generator.py --incident incident.json --timeline timeline.json --rca-method five_whys --action-items

Available RCA methods:

Step 3 -- Generate action items.

The agent categorizes action items as: immediate_fix, process_improvement, monitoring_alerting, documentation, training, architectural, tooling.

Each action item includes: title, owner, priority, deadline, success criteria, and dependencies.

Step 4 -- Validate postmortem quality.

python scripts/postmortem_generator.py incident_data.json --format json

The agent checks:

• Every contributing factor has at least one action item (coverage gap detection)
• Action items have quality scores (0-100) based on specificity
• MTTD/MTTR benchmarked against industry standards
• Missing actions suggested for uncovered themes

Validation checkpoint: Zero coverage gaps. All P0 action items have owners and deadlines within 48 hours.

Workflow 3: Escalation Management

Technical escalation path:

Business escalation:

Anti-Patterns

1. Individual blame in postmortems -- focus on system failures. "Why did the process allow this?" not "Why did Alice do this?"
2. Skipping PIR for SEV-2 -- every SEV-1 and SEV-2 gets a postmortem within 3 business days.
3. Action items without owners -- every item needs a specific person and deadline.
4. Deploying fixes under pressure without validation -- validate fixes before declaring resolution; plan for secondary failures.
5. Communication gaps -- provide updates even when there is no new information.

Troubleshooting

References

Integration Points

Last Updated: April 2026 Version: 1.1.0