The Agent Skills Directory

[Indirect Prompt Injection Considerations]: The skill describes how to receive and process incoming emails via Cloudflare Email Routing. Since email content comes from external sources, it represents a potential surface for indirect prompt injection if an agent processes the content without safeguards.
Ingestion points: Untrusted data enters the context through the email() handler described in references/routing.md, specifically via the message.raw stream.
Boundary markers: While the skill does not specify delimiters for the AI context, it strongly advises a 'Human-in-the-Loop' architecture where replies are reviewed before being sent.
Capability inventory: The skill uses email sending capabilities (env.EMAIL.send), forwarding (message.forward), and database operations (ctx.storage.sql.exec) within Cloudflare Workers.
Sanitization: The skill suggests using postal-mime for parsing and emphasizes storing drafts for human review as a primary safety measure.
[Command Execution Patterns]: The instructions involve the use of the wrangler CLI for service configuration and testing (e.g., npx wrangler email sending enable). These commands are standard administrative tools for the Cloudflare platform.
[External Dependencies and Downloads]: The skill references several official Cloudflare resources and standard Node.js libraries, such as postal-mime and mimetext. It also provides examples for interacting with the Cloudflare REST API at api.cloudflare.com. These integrations are consistent with the skill's purpose of managing Cloudflare services.

cloudflare-email-service