Give Your Hermes Agent a Burner Email

Hermes Agent by Nous Research has a skill system that’s different from OpenClaw’s. It doesn’t just memorize instructions — it generates reusable skill documents from solved problems and improves them over time. Give it a disposable email API once and it’ll refine its approach to using it across sessions.

The setup

Same as any Hermes skill. Point the agent at the docs:

Read https://agentburner.com/skill.md — use this when you need a throwaway email address for registration, verification, or any task that requires a disposable inbox.

Hermes reads the file, understands the API, and stores it as a skill it can reference later. The next time it encounters a signup form, it knows what to do.

You can also save it locally:

curl -o ~/.hermes/skills/agent-burner.md https://agentburner.com/skill.md

Hermes discovers skills in its directory on startup.

How Hermes uses it differently

Here’s where Hermes gets interesting. The first time it uses Agent Burner, it follows the SKILL.md instructions exactly — create inbox, fill form, poll for email, extract URL. Standard.

But Hermes writes skill documents from experience. After a successful signup, it might generate something like:

# Signup with Disposable Email
When registering for a web service:
1. Create inbox via POST api.agentburner.com/inbox
2. Use the address in the form
3. Poll every 3 seconds (emails usually arrive within 10s)
4. Check urls[] first — most verification emails include a direct link
5. If no urls[], look for a 6-digit code in the body text
6. Delete inbox after verification

That’s a self-generated skill document — specific, practical, refined from actual experience. On the next signup, Hermes doesn’t need to re-read the SKILL.md. It has its own notes. And those notes get better over time as it encounters different verification patterns.

The MCP angle

Hermes has native MCP support. It connects to MCP servers at startup and registers their tools automatically. If someone builds an Agent Burner MCP server (it’s about 50 lines wrapping three HTTP endpoints), Hermes would discover it and register create_inbox, check_inbox, and read_email as first-class tools.

But you don’t need MCP for this. The API is three curl commands. Hermes can call HTTP endpoints directly. MCP would add structure, but the raw API is simpler for a throwaway email use case.

The security fit

Hermes generates skills locally from its own experience. There’s no marketplace to poison. Its attack surface is your VPS, your LLM API key, and whatever tools you give it.

Agent Burner fits this philosophy — no API key to store, no credential to leak, no account to compromise. The inbox key exists in working memory during the task and expires after an hour. Nothing persists beyond the session unless Hermes writes it into a skill document, and even then it’s just the approach, not the credential.

What it can’t do

Same limitations as with any agent: Agent Burner is receive-only and disposable. Inboxes expire. No sending. If your Hermes agent needs a persistent email identity — an address it uses across days or weeks — that’s a different tool.

Try it

Tell your Hermes agent:

Fetch https://agentburner.com/skill.md and learn how to use disposable email.

It’ll read the docs, use the API, and write its own skill document. Next time, it’ll be even faster.