Skip to main content

Clinical AI Engineering

  • · review due

Clinical Prompt Library

A governed, open-source library of reproducible prompt templates for Nurse Citizen Developers.

When executing high-risk clinical administration or generating complex synthetic data, ad-hoc prompting is insufficient. The Clinical Prompt Library provides heavily tested, peer-reviewed templates that ensure AI outputs are formatted correctly, clinically safe, and compliant with NHS guidelines.

The Problem with Ad-Hoc Clinical Prompting​

If five different nurses prompt an AI to "summarize this patient handover," they will get five vastly different summaries. In a clinical environment, this lack of standardisation introduces unacceptable risk.

How to Use This Library​

Each template in this library follows a strict structure (based on the GOV.UK AI lab framework):

  1. Clear Constraints: Defining what the AI must not do (e.g., "Do not infer diagnoses").
  2. Explicit Formatting: Demanding specific outputs (e.g., SBAR format, JSON structure).
  3. Safety Rails: Injecting mandatory disclaimers or escalations.

These principles mirror current (2026) best practice for agentic prompting: a good clinical prompt reads like a job description — role, goal, boundaries, definition of done — not a one-line question. Current reasoning models (such as Claude Opus 5.5, GPT-6 Astra and Gemini 3.1 Pro) handle the reasoning themselves, so the value you add is in constraints, examples, and verification.

Harden Templates in a Loop​

No template is safe on the first draft. Every template here is refined through the Prompt Refinement Loop — Brief → Run → Critique → Refine → Verify — and re-tested against edge cases before it earns a place in the library.

Contributing​

To ensure rigorous quality control, all new prompt templates must use our standard prompt-template.md.

Next: See the Clinical Prompt Template to structure your own.