Installable skillMIT

Agent Email CLI

An MCP server that provides on-demand skill loading for AI coding assistants.

BY modbenderUPDATED Oct 27, 2026
INTENDED FOR◈OpenClawInstall in a listed agent.

instructions.md

---
name: agent-email-cli
description: Operate the agent-email CLI to create disposable inboxes, poll for new mail, retrieve full message details, and manage local mailbox profiles. Use when the user needs terminal-based email inbox access for LLM or agent automation workflows.
---

# Agent Email CLI

## Overview

Use this skill to operate the `agent-email` command safely and predictably for agent workflows that need inbox access.

Prefer JSON-native command output and return key fields (`email`, `messageId`, `subject`, `createdAt`, `from.address`) in your summaries.

## Workflow

1. Verify CLI availability.

```bash
command -v agent-email
agent-email --help
```

If missing, install:

```bash
npm install -g @zaddy6/agentemail
# or
bun install -g @zaddy6/agentemail
```

2. Create a mailbox account.

```bash
agent-email create
```

Record these fields from JSON output:

- `data.email`
- `data.accountId`
- `data.activeEmail`

Do not record, repeat, or print secret values such as mailbox passwords or tokens.

3. Read latest messages.

```bash
agent-email read <email|default>
```

For inbox waiting/polling:

```bash
agent-email read <email|default> --wait 30 --interval 2
```

For full message payloads:

```bash
agent-email read <email|default> --full
```

4. Retrieve one message in detail.

```bash
agent-email show <email|default> <messageId>
```

Use `show` when you need body/source details for verification links, codes, or full content extraction.

5. Manage mailbox profiles.

```bash
agent-email accounts list
agent-email use <email|default>
agent-email accounts remove <email>
```

Avoid commands that require entering secrets on the command line in agent logs.

6. Delete processed/irrelevant message when requested.

```bash
agent-email delete <email|default> <messageId>
```

## Operational Guidance

- Keep command output machine-readable; avoid forcing human output unless requested.
- Prefer `default` alias when user does not specify an email.
- Never echo, store, or summarize secret values (`password`, `token`) from command output.
- If command fails, surface the JSON error `code` and `hint` fields directly.
- For auth failures (`AUTH_REQUIRED`/401), rerun command once and request user intervention if credentials must be re-established.
- For rate limits (`RATE_LIMITED`/429), retry after short delay.

## Troubleshooting

- `command not found`: ensure `~/.bun/bin` or npm global bin path is on `PATH`.
- `NO_ACTIVE_ACCOUNT`: run `agent-email create` or `agent-email use <email>`.
- `ACCOUNT_NOT_FOUND`: run `agent-email accounts list` and pick a valid address.
- `EOTP` during npm publish: use npm trusted publishing for CI or publish locally with OTP.

## References

- For command cheat sheet and JSON field map, read [references/commands.md](references/commands.md).
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.