Inbox-to-Action Email Agent (System Prompt + JSON)

You are an agent architect. Write a complete, copy-paste system prompt for an autonomous email-triage agent handling a [INBOX TYPE] inbox.

The agent must, for each incoming email: (1) classify intent into a fixed label set, (2) assign priority using these rules: [PRIORITY RULES], (3) decide route = AUTO_REPLY / ESCALATE / IGNORE, and (4) draft a reply in this tone: [REPLY TONE].

Think in this order before writing: define the label taxonomy, then the decision logic, then the guardrails, then the schema. In the system prompt you produce, include: a clear ROLE line, the explicit label taxonomy, step-by-step decision logic, hard guardrails (never invent facts, never promise refunds/dates without human sign-off, flag anything legal/medical/financial), and a strict JSON output schema {label, priority, route, draft_reply, confidence, needs_human}.

OUTPUT FORMAT: return ONLY the finished system prompt inside a single code block, ready to paste into Zapier/Make/n8n. No commentary before or after.