name: Incident Management
description: >-
  Workflow capability for managing IT incidents using BigPanda AIOps,
  including alert ingestion, incident triage, environment management,
  and maintenance scheduling.
version: v1

imports:
  - shared/bigpanda.yaml

tools:
  - name: send-alert
    import: bigpanda.send-alert
    description: >-
      Ingest a monitoring alert from any source into BigPanda for AI-powered
      correlation and incident creation.
    inputSchema:
      type: object
      required:
        - app_key
        - status
        - host
      properties:
        app_key:
          type: string
          description: BigPanda application key for routing
        status:
          type: string
          description: Alert status (critical, warning, ok, acknowledged)
        host:
          type: string
          description: Hostname associated with the alert
        check:
          type: string
          description: Check or metric name
        description:
          type: string
          description: Human-readable alert description
  - name: list-environments
    import: bigpanda.list-environments
    description: List all BigPanda environments for incident grouping.
  - name: create-environment
    import: bigpanda.create-environment
    description: Create a new environment to group related incidents.
    inputSchema:
      type: object
      required:
        - name
        - condition
      properties:
        name:
          type: string
          description: Environment name
        condition:
          type: string
          description: Filter condition for grouping incidents
  - name: list-incidents
    import: bigpanda.list-incidents
    description: List active and resolved incidents for an environment.
    inputSchema:
      type: object
      required:
        - environment_id
      properties:
        environment_id:
          type: string
          description: Environment ID to list incidents for
        active:
          type: boolean
          description: Filter to show only active incidents
  - name: get-incident
    import: bigpanda.get-incident
    description: Get full details of a specific incident including correlated alerts.
    inputSchema:
      type: object
      required:
        - environment_id
        - incident_id
      properties:
        environment_id:
          type: string
          description: Environment ID
        incident_id:
          type: string
          description: Incident ID
  - name: list-maintenance-plans
    import: bigpanda.list-maintenance-plans
    description: List all scheduled maintenance plans.
  - name: create-maintenance-plan
    import: bigpanda.create-maintenance-plan
    description: Schedule a maintenance window to suppress alerts for specified hosts.
    inputSchema:
      type: object
      required:
        - name
        - condition
        - start
        - end
      properties:
        name:
          type: string
          description: Maintenance plan name
        condition:
          type: string
          description: Host filter condition
        start:
          type: integer
          description: Start Unix timestamp
        end:
          type: integer
          description: End Unix timestamp
  - name: send-change
    import: bigpanda.send-change
    description: Ingest a deployment or configuration change for correlation with alerts.
    inputSchema:
      type: object
      required:
        - summary
        - status
      properties:
        summary:
          type: string
          description: Change summary description
        status:
          type: string
          description: Change status (started, success, failure)
        hosts:
          type: array
          description: Affected hosts
  - name: get-audit-logs
    import: bigpanda.get-audit-logs
    description: Retrieve audit logs for compliance and troubleshooting.

expose:
  rest:
    port: 8080
  mcp:
    port: 9080

personas:
  - id: sre-engineer
    name: SRE Engineer
    description: Site reliability engineer managing incidents and alert correlation
  - id: it-ops-manager
    name: IT Ops Manager
    description: IT operations manager overseeing incident response and maintenance scheduling