The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external mail servers through webhook events, which represents a potential injection surface.
Ingestion points: The handle_events Python snippet and Express handler in SKILL.md process JSON payloads containing fields like email, event, and reason from the SendGrid Event Webhook API.
Boundary markers: The skill body includes a specific warning note stating that the reason field originates from external mail servers and should be treated as untrusted data, advising against passing it directly into LLM system prompts without isolation.
Capability inventory: The provided code snippets perform logic-based processing and logging (print, console.log). No privileged operations (subprocess, file-writes) are included in the skill code.
Sanitization: The skill documentation advises users to isolate these fields but does not include sanitization logic in the minimal example code.

twilio-sendgrid-webhooks