The Agent Skills Directory

PROMPT_INJECTION (MEDIUM): The skill presents an attack surface for Indirect Prompt Injection (Category 8) by processing and transmitting data to external parties.
Ingestion points: The send_sms method in SKILL.md accepts a body parameter, and the send_verification method accepts a to parameter, both of which ingest data from the agent's context that may originate from untrusted external sources.
Boundary markers: There are no boundary markers or explicit instructions provided to the agent to treat interpolated data as non-executable or to ignore embedded instructions within the message body.
Capability inventory: The skill possesses significant external communication capabilities via the Twilio API (self.client.messages.create and self.client.verify.v2.services), allowing it to send SMS, WhatsApp messages, and initiate voice calls.
Sanitization: While the skill validates phone number formats using regex, it performs no sanitization, filtering, or escaping of the message content itself, which could be exploited to send phishing links or unauthorized instructions if the agent is compromised.

twilio-communications