The Agent Skills Directory

Indirect Prompt Injection Surface: The skill describes workflows for processing incoming emails using the email() handler. Since email content originates from external, untrusted sources, it presents a surface for indirect prompt injection where an attacker could embed instructions intended to influence an AI agent's behavior.
Ingestion points: Incoming messages processed by the email() function in references/routing.md.
Boundary markers: The documentation recommends parsing raw MIME content using postal-mime and buffering the stream, which allows for structured processing of the message parts.
Capability inventory: The skill utilizes file system storage via Durable Objects, network operations via the send_email binding, and email forwarding capabilities.
Sanitization: The implementation examples suggest extracting specific fields (subject, text, headers) and using parameter binding in SQL queries (storage.sql.exec) to prevent injection attacks.
External Dependencies: The documentation references standard ecosystem tools such as wrangler and libraries like postal-mime and mimetext for MIME processing. These are well-known packages used for Cloudflare Workers development.
Administrative Command Execution: The skill provides examples of using the wrangler CLI for service configuration and curl for interacting with the Cloudflare REST API. These commands are standard for managing Cloudflare resources and utilize placeholders for sensitive credentials (e.g., <API_TOKEN>), encouraging secure secret management practices.

cloudflare-email-service