Skip to main content
The Agent Email Inbox skill enables AI agents to securely receive and respond to emails through a Resend Inbox. It provides security patterns that prevent untrusted email content from controlling your agent, including sender allowlists, content filtering, and sandboxed processing. The inbox groups each conversation into a thread, and your agent replies in the thread with one call. Without a custom domain, or when your plan has no inbox available, the skill falls back to Receiving with the same security levels.

Why your agent needs a secure inbox

An AI agent’s inbox receives untrusted input. Without proper security, anyone who knows your agent’s email address can send instructions that your agent will execute. The Agent Email Inbox skill solves this with a leveled security approach built in from the start. Unlike polling-based approaches, Resend uses webhooks for inbound email. Your agent is notified instantly when an email arrives. No cron jobs, no wasted API calls checking empty inboxes.

Installation

This skill is part of the resend-skills repository, which also includes the Resend skill for sending emails. When you run the install command, you’ll be prompted to choose which skills to install.

Advantages

  • 5 security levels: From strict sender allowlists to human-in-the-loop approval, choose the right level of security for your use case.
  • Real-time webhook architecture: Event-driven design means your agent reacts to emails in seconds, not minutes.
  • Threaded replies: Your agent replies inside the conversation, and Resend sets the recipients and threading headers.
  • Webhook signature verification: Built-in patterns for verifying Resend webhook signatures to prevent spoofed events.
  • Multi-language SDK support: The inbox examples use the Node.js SDK. The Receiving fallback works with Node.js, Python, Go, Ruby, PHP, Rust, Java, and .NET.
  • Common threat protection: Built-in guidance for handling prompt injection, sender spoofing, and email flooding.

Architecture

Your agent only processes emails that pass your chosen security level. Rejected emails are logged and silently acknowledged to prevent retries.

Security Levels

Choose a security level before setting up your webhook endpoint. Start with Level 1 and relax only if needed. Each level includes full implementation code in the skill. After installation, your AI agent will have access to the detailed security patterns and can help you implement the right level.

Example

A minimal webhook handler that verifies the signature, checks the sender against an allowlist, retrieves the full email, and replies in its thread:
The inbox methods need resend 6.34.1 or later.

Learn More

OpenClaw Setup Guide

Step-by-step guide for giving an OpenClaw agent a fully functional email inbox.

View on GitHub

See the full source code, security patterns, and documentation.