The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface where untrusted data from external sources can influence agent behavior.
Ingestion points: External data enters the agent context via incident_summary and incident_id from PagerDuty webhooks, as well as device_name and status from the NetClaw monitoring system (SKILL.md).
Boundary markers: The skill lacks explicit boundary markers or instructions for the agent to ignore potentially malicious commands embedded within the alert data before interpolating them into voice messages.
Capability inventory: The skill uses @twilio-alpha/mcp to perform twilio_voice_emergency_call and twilio_voice_call, which are automated and bypass user confirmation ("Auto-approved: Yes"). It also logs to Memory MCP and sends fallback notifications via Slack (SKILL.md).
Sanitization: While the skill documentation claims to sanitize messages by removing IPs and credentials, it does not specify filters for prompt injection patterns or instructions meant to manipulate the agent's logic.

twilio-emergency-call