clinicalNoteSummarizer

import { clinicalNoteSummarizer } from '@agentskit/skills'

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

Output is fixed (Subjective / Objective / Assessment / Plan). Empty sections show (not documented) rather than inferred. Vitals and lab values are copied verbatim.

#When to use

• Clinician-facing documentation tools that summarize free-text SOAP notes from an EHR.
• Discharge-summary pipelines where structured output is required before handoff.
• QA tools that diff a clinician's note against the structured summary to catch gaps.

#Behavior

• Always emits all four SOAP sections in order; missing sections get (not documented), not invented content.
• Copies BP, HR, lab values, and dosages verbatim — no rounding unless the source rounds.
• Strips identifiers: patient name → [patient], removes MRN, DOB, address. Preserves age range, sex, and clinical context.
• Flags inconsistencies (e.g. conflicting BP readings) with a note for clinician review rather than silently resolving them.
• Never adds clinical interpretation, differential, or diagnosis if the source note does not state one.

#Best practices

• Audience is clinicians, not patients. Do not expose this skill to patient-facing surfaces.
• Strip or redact PHI before passing notes to the model if your deployment is subject to HIPAA BAA requirements with the LLM provider.
• Gate output with a human review step before ingesting into any EHR or billing system.
• Pair with healthcareAssistant only if you need a patient-facing summary from a separate plain-language pass — keep the clinician and patient paths separate.

#Related

• healthcareAssistant
• Skills overview