Healthcare Diagnostic AI Agent

Write a production-ready system prompt for a healthcare symptom triage AI agent for [CLINIC/HEALTH SYSTEM NAME]. The agent helps patients understand their symptoms and determine appropriate care level. Requirements: 1) Role: a clinical triage assistant — NOT a doctor, cannot diagnose, cannot prescribe, 2) Emergency detection: ALWAYS check for emergency symptoms first (chest pain, difficulty breathing, stroke symptoms, severe bleeding, suicidal ideation) — immediately direct to 911 or ER, 3) Symptom collection: ask about symptom onset, severity (1-10), duration, associated symptoms, relevant medical history, 4) Care level recommendation: self-care at home / urgent care / ER / call 911 — with specific reasoning, 5) Escalation: offer to connect with a nurse or schedule an appointment, 6) Prohibited: never diagnose, never recommend specific medications, never minimize serious symptoms, 7) Tone: calm, empathetic, clear. Include 3 few-shot examples: mild symptom, moderate symptom, emergency symptom.

Build a Node.js TypeScript HIPAA-compliant conversation logging system for a healthcare AI chatbot. Requirements: 1) PHI detection: before logging any message, scan for PHI using regex patterns for: SSN, date of birth, phone numbers, email addresses, medical record numbers, full names (first + last), 2) PHI redaction: replace detected PHI with [REDACTED_TYPE] placeholders before storing, 3) Audit log schema: { sessionId, timestamp, role (user/assistant), messageHash (SHA-256 of original), redactedContent, phiTypesDetected[], agentAction, agentReasoning }, 4) Storage: encrypted at rest — use AES-256 encryption before writing to database, 5) Retention: auto-delete logs after 7 years (HIPAA minimum), 6) Access log: log every time a conversation record is accessed (who, when, why), 7) TypeScript with full type safety. Include the PHI regex patterns and encryption utilities.

Build a conversational appointment booking flow for a healthcare AI agent. The agent should: 1) Collect: patient name, date of birth (for identity verification), reason for visit (free text), preferred provider (if any), preferred date/time range, 2) Check availability: call GET /api/appointments/availability?providerId=X&date=Y to get open slots, 3) Present options: show 3 available slots in a natural conversational format ("I have Tuesday at 2pm with Dr. Smith, or Wednesday at 10am with Dr. Jones — which works better?"), 4) Confirm: summarize the appointment details and ask for confirmation, 5) Book: call POST /api/appointments/book with collected data, 6) Send confirmation: trigger email confirmation with calendar invite (.ics file), 7) Handle edge cases: no availability (offer waitlist), patient cancels mid-flow (save partial data), provider not available (suggest alternatives). Write as a TypeScript state machine with explicit states and transitions.