---
name: incident_coordination
description: Manage in-store incidents across multiple agents — classify severity, assemble responders, coordinate response, document for follow-up.
---

# Incident Coordination

## When to use

- Equipment failure that impacts service.
- Customer incident requiring multi-role response (medical, security, dispute).
- Multi-agent operational conflict (resource contention, scheduling overlap).
- Cascading failures across workflows.

## Core procedure

1. **Receive the event** — incident events arrive from agents via the escalation mechanism, from telemetry threshold breaches, or from human operators.
2. **Classify severity** — use the severity matrix in reference data. Severity drives response composition and notification routing.
3. **Assemble the responder set** — minimum responders per severity level; for high-severity events, page the human operator via `escalation.notify_human`.
4. **Coordinate** — dispatch initial action to the lead responder, monitor in `active_workflow_state`. Hold a virtual standup every 5 minutes for severity 1-2 incidents.
5. **Document continuously** — short structured updates to `incident_history` as the incident evolves; do not batch documentation to the end.
6. **Close the incident** — once resolved, record outcome, root cause (or "RCA pending"), and any follow-up actions. Update `historical_incident_patterns` if this is a recurring class.
7. **Emit telemetry** — write `incident_frequency`, `incident_resolution_time`, and `escalation_frequency` signals.

## Reference data

<details>
<summary>Severity matrix and response playbooks</summary>

**Severity classification:**

| Severity | Examples | Responder set | Notification |
|---|---|---|---|
| 1 (critical) | Customer injury, fire, theft in progress | shift_manager_agent + cleaning_agent + human operator | Page human; call 911 if applicable |
| 2 (high) | Equipment failure during peak, dispute escalating | shift_manager_agent + relevant product agent | Notify owner via on-call channel |
| 3 (medium) | Single-drink quality issue with unhappy customer, minor spill in customer area | shift_manager_agent (informed) + originating agent (resolves) | Standard incident log |
| 4 (low) | Single recipe error, minor stockout | originating agent + inventory_agent if applicable | Standard incident log |

**Regulator-notification triggers:**
- Health code violation observed → notify shift_manager_agent who notifies compliance owner within 24h
- Food safety incident → preserve evidence, notify human operator immediately
- Injury requiring medical attention → page human operator, record per safety_policy

**Playbooks** (load on demand from `incident_history`):
- equipment_failure_playbook
- customer_dispute_playbook
- medical_emergency_playbook
- security_incident_playbook

</details>

## Guardrails to respect

- `agent.shift_manager.must_document_incidents` — documentation is required, not optional.
- `agent.orchestrator.must_log_incidents` — orchestrator's own coordination actions are logged.

## Related

- **Used by agents**: orchestrator_agent (primary), shift_manager_agent
- **Capabilities consumed**: escalate_issue, escalation_management, send_notifications
- **Memory namespaces**: incident_history, historical_incident_patterns, active_workflow_state
- **Telemetry signals emitted**: incident_frequency, incident_resolution_time, escalation_frequency
- **Workflows that invoke**: equipment_failure_workflow, refund_escalation_workflow
- **Evals**: agent.shift_manager.incident_resolution_time, system.operational_latency