The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted incoming email data and provides the AI agent with tools to read and send emails, creating a potential surface for indirect prompt injection.\n
Ingestion points: The worker/index.ts script receives incoming emails via Cloudflare Email Routing and forwards them to a Mailbox Durable Object.\n
Capability inventory: The EmailAgentDurableObject exposes read_inbox and send_email tools to the AI agent, allowing it to act on the content of processed emails.\n
Boundary markers: No explicit delimiters or instructions to ignore embedded content are shown in the agent's system prompt or tool handlers.\n
Sanitization: There is no evidence of sanitization or filtering of email bodies before they are passed to the Workers AI model.\n- [EXTERNAL_DOWNLOADS]: Fetches the project repository and required dependencies from Cloudflare's official GitHub organization.\n
Clones the source code from https://github.com/cloudflare/agentic-inbox.git.\n- [COMMAND_EXECUTION]: Executes shell commands for deployment and secret management.\n
Uses wrangler secret put to store Cloudflare Access credentials.\n
Executes npm run deploy to provision resources and deploy the worker.

cloudflare-agentic-inbox