Developer builds n8n email agent that scrapes lead websites and refuses to guess

A developer has built an n8n workflow that automatically scrapes a lead's website before drafting cold outreach emails, ensuring every claim in the email is grounded in real content from that site. The system pulls leads from a Google Sheet, fetches each company's homepage, extracts up to 6,000 characters of text, and only then prompts a language model to write a personalised email. If a site is inaccessible, too thin, or returns an error page, the workflow outputs a [NEEDS-HUMAN] flag instead of generating a fallback email, routing that row to a separate review status. Error handling is built in so a single dead link cannot stall the rest of the batch, and all outputs are written as drafts for human review before any email is sent. The developer also built a 14-test acceptance suite against fixture inputs — including fake prompt-injection attempts — to validate grounding and refusal behaviour before deploying the workflow.