Twilio Emergency Call Skill

Feature: 042-twilio-voice-mcp Priority: P1 (MVP) User Story: US1 - Emergency Alert Calls

Purpose

Automatically call John when critical network events occur (P1 incidents, core device failures). Emergency calls bypass quiet hours and are auto-approved without user confirmation.

Trigger Conditions

This skill activates when:

1. PagerDuty P1 Incident

• Source: pagerduty
• Match: severity:P1
• Auto-approved: Yes

1. Core Device Down

• Source: netclaw_monitoring
• Match: device_type:(core_router|firewall|wan_link) AND status:down
• Auto-approved: Yes

Workflow

Event Detected
    │
    ├─► Check Emergency Category
    │       └─► Not emergency? → Skip call
    │
    ├─► Pre-call Validation
    │       ├─► Rate limit check (warning only for emergencies)
    │       └─► Get whitelist target
    │
    ├─► Sanitize Message
    │       └─► Remove IPs, credentials, sensitive data
    │
    ├─► Initiate Call via Twilio
    │       └─► Uses @twilio-alpha/mcp Api20100401Call
    │
    ├─► Log to Memory MCP
    │       └─► CallRecord with trigger_id
    │
    └─► Fallback on Failure
            └─► T018: SMS or Slack notification

Tool Usage

Primary Tool

json
{
  "tool": "twilio_voice_emergency_call",
  "arguments": {
    "source": "pagerduty",
    "event_data": {
      "severity": "P1",
      "incident_id": "PD12345",
      "summary": "Core router LAX-CORE-01 is unreachable"
    }
  }
}

Manual Emergency Call

json
{
  "tool": "twilio_voice_call",
  "arguments": {
    "to": "[WHITELIST_NUMBER]",
    "message": "Emergency: Core router is down. BGP sessions lost on LAX-CORE-01.",
    "priority": "emergency"
  }
}

Message Templates

P1 Incident

NetClaw Emergency Alert.
PagerDuty Priority 1 Incident.
{incident_summary}.
Incident ID: {incident_id}.

Core Device Down

NetClaw Emergency Alert.
Core device failure detected.
{device_name} is {status}.
Last seen: {last_seen}.

Configuration

See config/twilio-voice.json:

json
{
  "emergency_categories": [
    {
      "category_name": "pagerduty_p1",
      "source": "pagerduty",
      "match_pattern": "severity:P1",
      "enabled": true
    },
    {
      "category_name": "core_device_down",
      "source": "netclaw_monitoring",
      "match_pattern": "device_type:(core_router|firewall|wan_link) AND status:down",
      "enabled": true
    }
  ]
}

Retry Logic

On call failure:

1. Wait 30 seconds
2. Retry (attempt 2)
3. Wait 60 seconds
4. Retry (attempt 3)
5. Fallback to SMS/Slack

python
retry_config = {
    "max_retries": 3,
    "backoff_base_seconds": 30,
    "backoff_multiplier": 2
}

Fallback Notifications

If all call attempts fail:

1. SMS via Twilio (if configured)
2. Slack notification to alerts channel
3. Log failure to Memory MCP

Guardrails

• Rate Limits: Logged but not blocking for emergencies
• Quiet Hours: Bypassed (p1_override: true)
• Whitelist: Must be in whitelist (no exceptions)
• Content: IPs and credentials sanitized

Example Scenarios

Scenario 1: PagerDuty P1 Alert

Input:
  - PagerDuty webhook received
  - Severity: P1
  - Summary: "Database primary node unresponsive"

Action:
  1. Event matches pagerduty_p1 category
  2. Call initiated to [WHITELIST_NUMBER]
  3. Voice message: "NetClaw Emergency Alert. PagerDuty Priority 1 Incident. Database primary node unresponsive."
  4. Call logged to Memory MCP

Scenario 2: Core Router Down

Input:
  - NetClaw monitoring detects
  - device_type: core_router
  - status: down
  - device_name: CHI-CORE-02

Action:
  1. Event matches core_device_down category
  2. Call initiated to [WHITELIST_NUMBER]
  3. Voice message: "NetClaw Emergency Alert. Core device failure detected. CHI-CORE-02 is down."

Integration Points

• PagerDuty MCP: Triggers emergency calls on P1 incidents
• Memory MCP: Stores all call records
• Slack MCP: Fallback notification
• @twilio-alpha/mcp: Call initiation

Success Criteria

• ] P1 incident triggers call within 60 seconds
• ] Call logged to Memory MCP with incident ID
• ] Quiet hours bypassed for emergencies
• ] Retry logic works on failure
• ] Fallback notification sent after max retries