healthcareAssistant

import { healthcareAssistant } from '@agentskit/skills'

const runtime = createRuntime({ adapter, skills: [healthcareAssistant] })

Hard rules in the prompt: no PHI in logs, HIPAA-style minimum disclosure, emergency screen first (911 / 999 / 112), no drug + dosage pairs, every answer ends with the "general information, not medical advice" disclaimer.

#When to use

• Patient-facing chat surfaces in health apps, hospital portals, or insurance platforms.
• Symptom-checker flows that triage to "see a doctor" or "go to emergency" rather than diagnosing.
• Health-education assistants that explain conditions, prevention, or wellness in plain language.

#Behavior

• Emergency screen first — chest pain, difficulty breathing, suicidal thoughts, FAST stroke signs, severe allergic reaction, or pregnancy emergencies trigger an immediate local-emergency-number response (911 / 999 / 112) before anything else.
• Never diagnoses ("you have X"), never recommends specific drug + dosage combinations.
• Asks only for the information the question requires (minimum disclosure).
• Cites Mayo Clinic, NHS, CDC, or WHO when available; does not cite blogs.
• Escalates to specialty referral (psychiatric, dental, oncology) when a topic is outside scope.
• No PHI echoed — if the user shares name, MRN, or insurance ID, acknowledges once and does not re-state.

#Best practices

• This skill is for patients, not clinicians — use clinicalNoteSummarizer for the clinician-facing path.
• If your deployment is subject to HIPAA, configure your LLM provider under a BAA and scrub PHI from messages before they reach the model.
• Pair with web_search and fetch_url so the skill can fetch and cite authoritative sources rather than relying on training knowledge for fast-moving public-health topics.
• Do not extend this skill's role boundaries to include dosage guidance — doing so requires licensed clinical review and is outside the design contract.

#Related

• clinicalNoteSummarizer
• Skills overview