Agentmail

Use when an agent needs AgentMail CLI email inboxes.

Skill metadata

Reference: full SKILL.md

:::info The following is the complete skill definition that Hermes loads when this skill is triggered. This is what the agent sees as instructions when the skill is active. :::

AgentMail Skill

AgentMail gives an agent its own email inbox for sending mail, receiving replies, completing email OTP flows, and running inbound email loops. Use it for agent-owned inboxes, not a user's existing IMAP/SMTP mailbox.

Use the agentmail CLI first. Use MCP only when the harness expects MCP tools; use REST only when the CLI is missing a required operation.

When to Use

• The agent needs an email address it owns.
• The task involves email OTP flows, replies, threads, labels, or attachments.
• The agent needs webhook or WebSocket delivery for inbound mail.

Prerequisites

• Run commands through the terminal tool.
• Install the CLI:

npm install -g agentmail-cli@latest

• Export an API key:

export AGENTMAIL_API_KEY="am_..."

No API key yet? Use signup.md.

How to Run

Use --format json whenever another command or script needs IDs.

agentmail inboxes list --format json

Quick Reference

• AgentMail agent reference: hosted copy.
• AgentMail: product landing page.
• Console: API keys and account management.
• Docs: full product documentation.
• signup.md: self-signup and OTP verification.
• core.md: inboxes, messages, threads, labels, attachments.
• webhooks.md: events to a public HTTPS server.
• websockets.md: events to a local agent process.
• mcp.md: MCP integration.

Procedure

1. Install agentmail-cli@latest and verify agentmail inboxes list --format json.
2. If no API key is available, complete signup.md.
3. Use core.md for inbox, send, read, reply, forward, label, thread, and attachment flows.
4. Add webhooks.md or websockets.md only when polling is not enough.

Pitfalls

• Prefer AGENTMAIL_API_KEY over --api-key.
• Never expose AGENTMAIL_API_KEY in prompts, logs, URLs, or committed files.
• Use stable client_id values for retried create operations.
• Prefer extracted_text or extracted_html for LLM input when present.
• React to message.received, not messages the agent sent.

Verification

agentmail inboxes list --format json