AG
agentmail
Send emails from Simone's own inbox using the native `mcp__internal__send_agentmail` MCP tool. Use when you need to send an email, reply, deliver a report, or communicate with anyone via email. Do NOT use bash, curl, SDK scripts, or CLI commands — just call the MCP tool directly.
Install
mkdir -p .claude/skills/agentmail-kjdragan && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/15630" && unzip -o skill.zip -d .claude/skills/agentmail-kjdragan && rm skill.zipInstalls to .claude/skills/agentmail-kjdragan
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Send emails from Simone's own inbox using the native `mcp__internal__send_agentmail` MCP tool. Use when you need to send an email, reply, deliver a report, or communicate with anyone via email. Do NOT use bash, curl, SDK scripts, or CLI commands — just call the MCP tool directly.280 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
About this skill
AgentMail — Simone's Email
Simone sends email via mcp__internal__send_agentmail. One tool call — no scripts, no CLI, no SDK.
Sending Email
mcp__internal__send_agentmail({
"to": "[email protected]",
"subject": "Subject line",
"body": "Email body content — plain text or HTML"
})
Required: to, subject, body. Optional: cc, bcc, dry_run.
Before Sending, Ask Yourself
- Who is the sender? If Simone → this tool. If Kevin's Gmail →
gmailskill instead. - Is the body formatted? For reports/structured content, wrap in basic HTML (
<h2>,<p>,<ul>) for readable email. Raw markdown renders poorly in email clients. - Is this a duplicate? The tool has built-in dedup guards. If it returns a "duplicate delivery blocked" error, your email was already sent — do NOT retry.
- Is the recipient correct? Kevin's email:
[email protected].
Anti-Patterns — NEVER Do These
- NEVER use bash, curl, or Python scripts to send email. The MCP tool handles auth, formatting, and Task Hub lifecycle tracking automatically. Scripts bypass all of this.
- NEVER use
mcp__AgentMail__send_message— that's the raw AgentMail MCP endpoint. It bypasses delivery tracking and dedup guards, causing phantom sends and orphaned tasks. - NEVER send receipt acknowledgements like "Got it, working on it" — the system blocks these. Only send the final, substantive response.
- NEVER construct AgentMail SDK code (
from agentmail import AgentMail,client.inboxes.messages.send(...)) — that's for external app development, not for Simone. - NEVER install
agentmail-clivia npm — the CLI is not needed; the native MCP tool is the correct path.
Error Handling
| Error Message | Meaning | Action |
|---|---|---|
'to' is required | Missing recipient | Add the to field |
AgentMail service is not available | Service not configured | Report error to user — cannot send |
duplicate final delivery blocked | Email already sent for this task | Stop — delivery succeeded earlier. Do NOT retry. |
Receipt acknowledgement blocked | Tried to send a "got it" message | Skip the ack — only send final substantive content |
| Connection/timeout error | Transient failure | Retry once, then report failure |
Quick Routing
| Request | Action |
|---|---|
| "Email this to Kevin" | mcp__internal__send_agentmail → [email protected] |
| "Send this report" | mcp__internal__send_agentmail → specified recipient |
| "Reply to this email" | mcp__internal__send_agentmail → original sender |
| "Send from Kevin's Gmail" | Use gmail skill (gws CLI) — completely different channel |