License
Apache License 2.0 — See LICENSE file for full text.
Email via Resend
Send and receive emails using the Resend API.
Configuration
No config file needed. The skill auto-discovers settings from:
- OpenClaw context — channel, chat_id, user email/name (from context)
- Environment variables — for API keys
- Interactive prompts — asks user if required info missing
Required Environment Variables
export RESEND_API_KEY="re_123456789" # Resend API key (required)
export DEFAULT_FROM_EMAIL="you@example.com" # Sender email (required)
export DEFAULT_FROM_NAME="Your Name" # Sender display name (required)
# Channel/target set by agent from context/memory - not hardcoded
First-Time Setup
When the skill is first invoked, the sub-agent should:
Check context — OpenClaw context already has:
context.user.email(from USER.md)context.channel(from current session)context.chat_idcontext.thread_id(for topics)
Check memory — Use
memory_searchtool:- Query: "email notification channel preference"
- Query: "email from address"
If missing, ask user — Via chat message:
- "Which email should I send from?"
- "Which channel for email notifications?"
Commit to memory — Write preferences to persist across sessions:
write path="memory/email-preferences.md" content="--- from_email: $EMAIL from_name: $NAME notification_channel: $CHANNEL chat_id: \"$CHAT_ID\" thread_id: \"$THREAD_ID\" --- # Email Notification Preferences Saved $(date) "This ensures memory_search finds it in future sessions.
Format (YAML frontmatter):
---
from_email: you@company.com
from_name: Your Name
notification_channel: telegram # or discord, signal, whatsapp, etc - from memory or ask user
chat_id: "123456789"
thread_id: "334" # optional, for topics
---
Context Fields (Available in Sub-Agent)
| Field | Source | Example |
|---|---|---|
user.email |
USER.md | you@company.com |
user.name |
USER.md | Your Name |
channel |
OpenClaw | from context |
chat_id |
OpenClaw | 123456789 |
thread_id |
OpenClaw | 334 |
The skill uses these directly from OpenClaw context — no parsing needed.
Usage
Inbound (Receive)
Cron Setup
openclaw cron add \
--name "email-resend-inbound" \
--every 15m \
--session-target isolated \
--message "Follow instructions in skills/email-resend/cron-prompts/email-inbound.md" \
--to "YOUR_CHAT_ID" \
--announce
Manual Check
python3 skills/email-resend/scripts/inbound.py
Notification Format
Each new email triggers a notification with:
- From, Subject, Date
- Body preview (~2000 chars)
- Attachment list (if any)
- Importance: 🔥 HIGH / 📅 MEETING / 📬 NORMAL
Acknowledge Flow
Reply to the notification message (using notification reply) to acknowledge, or type:
done/ack- mark as read
Use draft-reply.py to compose replies with proper quoting.
Important: Always use inline replies ([[reply_to_current]]) to keep messages linked in the thread. This enables:
- Proper custody chain tracking
- Reply-to-email tracing
- Better conversation flow
CRITICAL: When responding via OpenClaw message tool, use replyTo parameter (not [[reply_to_current]] tag):
message(action="send", channel="<from-context>", replyTo="<msg_id>", ...)
Scripts
| Script | Purpose |
|---|---|
inbound.py |
Check emails, send notifications |
draft-reply.py |
Draft reply workflow with quoting & threading |
outbound.py |
Send emails directly |
download_attachment.py |
Download attachments from inbound emails |
Downloading Attachments
To download attachments from an inbound email:
# List attachments (shows IDs)
python3 scripts/download_attachment.py <email_id> --list
# Download all to directory
python3 scripts/download_attachment.py <email_id> --output-dir ./attachments
# Download specific attachment
python3 scripts/download_attachment.py <email_id> --attachment-id <attachment_id>
Note: The API path is /emails/receiving/{email_id}/attachments (not the standard /emails/ path).
State Files
memory/email-resend-inbound-notified.json— pending/acknowledged emailsmemory/email-message-map.json— notification message_id → email_id (legacy)memory/email-custody-chain.json— Full DAG of email → notification → actionsmemory/email-msg-to-chain.json— notification message_id → chain lookupmemory/email-draft-state.json— Active draft state (email_id, status, reply_content)
See docs/custody-chain.md for DAG design.
Outbound (Send)
python3 skills/email-resend/scripts/outbound.py \
--to "recipient@example.com" \
--subject "Hello" \
--body "Message text"
# With attachments
python3 skills/email-resend/scripts/outbound.py \
--to "recipient@example.com" \
--subject "Here's the file" \
--body "See attachment" \
--attachment ./file.pdf \
--attachment ./image.png
⚠️ CRITICAL: Email Threading Rule (2026-02-22)
MANDATORY: Always use draft-reply.py for replying to emails.
This is non-negotiable. Failure to follow this rule will result in broken Gmail threading.
Why This Matters
- Gmail threads emails based on
In-Reply-ToANDReferencesheaders - Using wrong headers = reply appears as NEW thread = context lost
- There's no way to fix this after sending
✅ Correct Workflow (ALWAYS USE THIS)
# Step 1: Start draft (fetches Message-ID automatically)
python3 skills/email-resend/scripts/draft-reply.py start <email_id>
# Step 2: Set reply content
python3 skills/email-resend/scripts/draft-reply.py content "Your reply"
# Step 3: Send
python3 skills/email-resend/scripts/draft-reply.py send
⚠️ CRITICAL: Approval Execution Rule (2026-02-22)
When user approves a draft, you MUST execute the send command immediately.
The mistake to avoid:
- ❌ Show draft for approval → User says "send" → Only acknowledge, don't execute
- ✅ Show draft for approval → User says "send" → RUN
draft-reply.py send→ Then confirm
Correct workflow:
1. Show draft for approval
2. User replies "approve", "send", "yes", or "ok"
3. IMMEDIATELY run: draft-reply.py send
4. Only THEN confirm to user
Never:
- Only acknowledge the approval without executing
- Ask for confirmation after user already approved
- Wait to send - do it immediately
❌ NEVER Do These Things
NEVER use outbound.py for replies:
# WRONG - will break threading
python3 skills/email-resend/scripts/outbound.py \
--to "x@y.com" --subject "Re: Original" --body "Reply"
NEVER manually construct --reply-to flags:
# WRONG - guessing Message-ID format never works
python3 skills/email-resend/scripts/outbound.py \
--to "x@y.com" --subject "Re: Original" --body "Reply" \
--reply-to "<some-guess>@resend"
NEVER skip the workflow when subject starts with "Re:":
# WRONG - replying without threading headers breaks thread
python3 skills/email-resend/scripts/outbound.py \
--to "x@y.com" --subject "Re: Previous Thread" --body "Quick reply"
outbound.py Only For New Emails
outbound.py is for new emails only (not replies):
- First contact
- Announcements
- Emails where you intentionally want a NEW thread
For anything that could be a reply, use draft-reply.py.
Requirements
RESEND_API_KEYenvironment variable set- Python
requestslibrary
Draft Reply Best Practices
When composing a reply via draft-reply.py:
Always quote the original — Include the original message with
>prefix so recipient knows what you're responding toUse proper threading — Set
In-Reply-ToandReferencesheaders using the original email's Message-IDKeep subject line — Start with
Re:prefix to maintain thread (but avoid "Re: Re:")Structure:
Your reply here --- On [date] [original sender] wrote: > quoted original message > continues hereMultiple replies supported — After sending, draft is marked as "sent" so you can reply again to the same thread. Use
resumecommand to continue.No double Re: — If original subject already starts with "Re:", don't add another
Custody Chain — Track full lineage:
- Email → notification → All replies/actions
- DAG structure with parent links
- Any message traces back to original email
Draft Reply Commands
| Command | Purpose |
|---|---|
start <email_id> |
Start a draft reply to an email |
resume |
Continue a sent thread to reply again |
content "text" |
Set reply content |
send |
Send the reply |
cancel |
Cancel the draft |
status |
Show current draft status |
After sending, use resume to reply again to the same thread — threading headers are preserved.
Run tests:
python3 skills/email-resend/tests/test_inbound.py
Expected: 40 tests, all passing.
If tests fail:
- Check which test failed and why
- Fix the feature/code to match expected behavior
- Or update tests if feature intentionally changed