Claude Code sends a reviewed 16-recipient campaign
Vo reports using Claude Code to build and send 16 personalised B2B invitations through Resend, fixing clipping and Vietnamese encoding before an explicit send instruction.
Original prompt
Shared by the source author. See source notes for context.
send the full batch to all 16. track open and click. do report after 3 hours.
Reach people exploring AI agents
See placements →What happened
Vo reports using Claude Code to build and send 16 personalised B2B invitations through Resend, fixing clipping and Vietnamese encoding before an explicit send instruction.
Reported by Tri VoHow the agent did it
- Build recipient and account-manager data with HTML and plain-text templates.
- Preview every row and send a test to the author’s own inbox.
- Repair image delivery and UTF-8 rendering problems from the actual test.
Where a person stepped in
Vo inspects previews and real test emails, verifies personalised contact isolation and explicitly authorises the final batch. Open/click status is not treated as verified human engagement.
Tri Vo uses Claude Code to build a real sixteen-recipient invitation campaign, with two account managers and recipient-specific merge fields. The project grows from a test email into a recipient CSV, manager registry, HTML and plain-text templates, previews, dry-run mode and send script.
Testing catches problems the code alone did not reveal. Embedded base64 images cause Gmail clipping and broken graphics; attachment references replace them. Vietnamese accented names display incorrectly until the template declares UTF-8. Vo also checks that each rendered recipient version contains only the correct account manager's contact details, and flags uncertain honorifics instead of assuming every name can be classified.
Only after previews, a test to his own inbox and a full dry run does Vo give the explicit instruction to send all sixteen. A later API query corrects the agent's earlier mistaken claim that opened and clicked status is unavailable. The resulting perfect open figure is not proof of human interest, because corporate security scanning can affect it. The useful result is a reviewed merge-send pipeline and completed small campaign, not an independently measured sales gain. The public starter kit uses fictional sandbox data rather than exposing the real recipients.
Prompt
Original prompt
Source: Blog · Tri Vo
Frequently asked questions
What stopped the Gmail clipping?
The workflow replaced embedded base64 images with real attachments referenced by content identifiers.
How were recipient contacts checked?
Vo checked rendered versions for both managers’ addresses to ensure each contained only its assigned manager’s details.
Did a perfect open rate prove interest?
No. The source warns that corporate email-security scanning can register opens or clicks before a person reads the message.
More details
Remaining steps
- Verify manager-contact isolation and run the full list in dry-run mode.
- Send only after explicit approval, then inspect provider API status.
About this account
Claude Code builds and repairs the send pipeline, renders previews, executes the authorised Resend batch and queries the API for its subsequent report.
Firsthand account by Tri Vo; original publication date verified against Public page metadata datePublished 2026-07-31T17:37:25.000Z. Outcomes are the author’s reports, not independently reproduced. Original final-send instruction, issued after previews, a test email and dry run; not a suggested safe default.
Credit: Tri Vo
Source published 31 Jul 2026
Source checked 14 Oct 2026
Tools and topics
Claude Code · Bun · Resend API · Recipient CSV · HTML previews · Gmail
Reviewed campaign send · Merge-field isolation · Vietnamese encoding
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.