the-paramedic-foundation/paramedic-narrative-skill

AI-assisted PCR narrative documentation for paramedics and EMTs

Published by The Paramedic Foundation Licensed under CC BY 4.0 · Version 3.0.0

---

Overview

Paramedic-Narrative is an AI skill for patient care report (PCR) narrative documentation. It is an editorial tool designed for paramedics and EMTs who use AI language model tools and want a structured, clinically rigorous, non-hallucinating documentation assistant.

It produces PCR narratives in your agency’s declared format – SOAP with Clinical Summary by default, with native support for SOAPE, SOAPIER, CHART, CHARTE, DCHART-E, CHRONOLOGICAL, HEAD-TO-TOE, DRAATT, AT CHART, FACT, dedicated REFUSAL/NON-TRANSPORT and IFT templates, and CUSTOM agency-defined section orders – all grounded in one principle: the narrative explains the WHY. Structured PCR fields capture the WHAT and WHEN.

The skill does not restate vital signs, medication doses, or exam findings already captured in your ePCR platform. It captures what structured fields cannot: clinical reasoning, differential rationale, history source and reliability, medication indication and response, controlled substance audit trails, forensic evidentiary detail, and transfer-of-care narrative.

It never invents, assumes, or infers clinical detail. If you did not provide it, it does not appear in the draft. Missing items are flagged [VERIFY].

---

Critical Limitations – Read Before Use

This is an editorial tool only. It must never be used to make clinical decisions.

Using AI-generated content to inform clinical decisions in paramedicine is unethical, unsupported by evidence, and may be illegal under applicable professional licensing and scope-of-practice law. This tool has not been tested or validated for clinical decision support and is not intended for that purpose.

The Paramedic Foundation makes no warranty regarding the accuracy or completeness of any output. The provider bears full professional and legal responsibility for every submitted document.

PHI warning: Never photograph or upload patient-identifying information, including patient faces, license plates, or documents containing names, dates of birth, or other protected health information (PHI). Camera metadata may also transmit location data that constitutes PHI. The Paramedic Foundation is not responsible for privacy breaches resulting from provider conduct.

Read ETHICS.md before use.

---

What This Tool Does

• Produces PCR narrative drafts from provider-supplied information in your agency’s declared narrative format (14 formats supported, per-call override)
• Accepts photo plus dictation intake: monitor screens, med vials, facility paperwork, and handwritten notes, transcribed exactly and verified – never inferred from blur
• Supports fragmented documentation across sessions and targeted memory-jogging recall questions when documenting hours after a call
• Assembles an on-request IMIST-AMBO handoff prep during transport from facts already collected, and appends a retrospective IMIST-AMBO training example to each completed draft
• Produces an on-request prearrival notification note: a copy-paste block matched to notification platform fields (patient type as you stated it, chief complaint, brief note with vitals, destination), with no patient identifiers
• Captures clinical reasoning, differential rationale, and history source attribution that structured fields cannot hold
• Applies an elevated evidentiary standard for forensic and high-acuity calls
• Documents controlled substance audit trail elements the provider supplies
• Flags missing information as [VERIFY] rather than inventing it
• Keeps care performed by another agency, care performed before arrival, and interventions prepared but not performed out of structured fields and written out in full in the narrative, attributed to whoever performed them
• Keeps each patient’s facts separate from every other patient’s in a multi-patient incident
• Applies universal paramedicine documentation standards when no agency policy is provided

What This Tool Does Not Do

• Does not make or support clinical decisions of any kind
• Does not access ePCR platforms or submit documents
• Does not invent, assume, or infer clinical detail
• Does not reproduce information already in structured fields
• Does not reference a structured field entry that does not exist
• Does not assume the documenting crew performed an act whose performer is unstated
• Does not fabricate controlled substance audit trail elements
• Does not provide clinical advice

---

What is in this repository

If you have never used GitHub before, this page may look confusing. Here is what everything is. As of version 3.0.0, the standard has two hand-edited source files (ALWAYS-BLOCK.md and knowledge/) and a build script (build.py) that generates everything a provider actually installs, into dist/.

paramedic-narrative-skill/          ← the repository (this whole page)
├── README.md                       ← this file -- start here
├── ETHICS.md                       ← ethical framework -- read before use
├── LICENSE.md                      ← CC BY 4.0 license terms
├── ALWAYS-BLOCK.md                 ← SOURCE (hand-edited). The always-loaded
│                                     rules, byte-identical on every platform:
│                                     editorial-tool boundary, the ten NEVER
│                                     rules, attribution boundary (short form),
│                                     PHI, patient/incident separation,
│                                     during-a-call guardrails, pre-draft check,
│                                     style, standing disclaimer, router table
├── build.py                        ← builds dist/ from ALWAYS-BLOCK.md and
│                                     knowledge/, and verifies the result --
│                                     run `python3 build.py && python3
│                                     build.py --verify`
├── knowledge/                      ← SOURCE (hand-edited). 17 topic files,
│   │                                 each named for the situation that
│   │                                 requires it and opening with
│   │                                 "CONSULT THIS FILE WHEN: " /
│   │                                 "DO NOT CONSULT OTHERWISE."
│   ├── WHEN-ABC-LOC-VITALS.md
│   ├── WHEN-BARRIERS.md
│   ├── WHEN-BEHAVIORAL-HEALTH.md
│   ├── WHEN-CARDIAC-OR-STROKE.md
│   ├── WHEN-CHILDBEARING-AGE.md
│   ├── WHEN-DRAFTING.md
│   ├── WHEN-FORENSIC.md
│   ├── WHEN-FORMAT.md
│   ├── WHEN-HANDOFF.md
│   ├── WHEN-INTAKE.md
│   ├── WHEN-MEDICATION.md
│   ├── WHEN-MULTIPLE-PATIENTS.md
│   ├── WHEN-NO-TRANSPORT.md
│   ├── WHEN-OTHER-AGENCY.md
│   ├── WHEN-SCORING-TOOL.md
│   ├── WHEN-SESSION-STARTS.md
│   └── WHEN-SUBSTANCE-USE.md
├── agency-config-template.md       ← standardized configuration template for
│                                     agencies and Chief Paramedics to complete
├── provider-profile-template.md    ← personal provider profile template --
│                                     one per provider, never shared
├── assets/                         ← brand and design assets
│   ├── pn-header.png               ← horizontal header lockup (PNG, used in README)
│   ├── pn-header.svg               ← horizontal header lockup (SVG source)
│   ├── pn-badge.png                ← square badge mark (PNG)
│   ├── pn-badge.svg                ← square badge mark (SVG source)
│   ├── pn-ecg.png                  ← ECG stylistic element (PNG)
│   ├── pn-ecg.svg                  ← ECG stylistic element (SVG source)
│   ├── pn-social-preview.png       ← GitHub social preview card (1280x640)
│   └── pn-social-preview.svg       ← social preview card (SVG source)
├── docs/                           ← provider-facing documents
│   ├── TPF_ParamedicNarrative_DictationPocketCard_2026_v1.txt
│   │                               ← 12-point dictation skeleton (text)
│   ├── TPF_ParamedicNarrative_DictationPocketCard_2026_v1.pdf
│   │                               ← printable pocket-size card (5in x 6.55in)
│   ├── TPF_ParamedicNarrative_DictationBusinessCard_2026_v1.pdf
│   │                               ← wallet-size card, category names only
│   ├── TPF_ParamedicNarrative_MobileReferenceCard_2026_v1.png
│   │                               ← tall reference image for a phone screen
│   └── spec-addendum-sections-4-4A-4B-4C.md  ← v1.3.0 design addendum
└── dist/                           ← GENERATED by build.py. Never hand-edit
    │                                 anything in this folder -- it is
    │                                 overwritten on every build.
    ├── system-prompt.md            ← full system prompt for Gemini, direct
    │                                 API, Copilot's per-session paste, and
    │                                 any platform without file retrieval --
    │                                 the always-loaded block plus all 17
    │                                 topic files, inlined into one document
    ├── chatgpt-instructions.md     ← condensed Instructions field for
    │                                 ChatGPT and for Copilot's one-time
    │                                 Custom Instructions
    ├── claude-project-instructions.md  ← Instructions field for a Claude
    │                                     Project
    ├── paramedic-narrative.skill   ← the install file for Claude Skills --
    │                                 a single upload containing everything
    │                                 below plus the root docs
    └── paramedic-narrative/
        ├── SKILL.md                ← full standard, packaged with the
        │                             frontmatter Claude's Skill feature reads
        └── references/             ← 17 files, generated -- identical
                                       content to knowledge/, copied here so
                                       the .skill package and manual Knowledge
                                       File uploads can use them directly

You only need one or two of these depending on who you are:

• Line provider using Claude: download dist/paramedic-narrative.skill
• Line provider using ChatGPT: see the ChatGPT setup below
• Line provider using Gemini or another platform: open dist/system-prompt.md and copy its contents
• Line provider using Microsoft Copilot: see the Copilot section below
• Setting up on a phone: see the Mobile section below
• New provider setting up their profile: open provider-profile-template.md (or say “build my provider profile” to the skill and it will guide you)
• Agency administrator or Chief Paramedic: open agency-config-template.md (or say “set up agency configuration” to the skill and it will guide you)
• Read the full standard: open dist/paramedic-narrative/SKILL.md, or browse ALWAYS-BLOCK.md and knowledge/ directly – those two are the source
• Read the ethical framework: open ETHICS.md

---

Start here if you are new to all of this

You do not need to understand GitHub to use this tool. GitHub is just where the files are stored so anyone can find and download them for free. You only need to visit this page once to get set up. After that, you use the tool entirely inside Claude, ChatGPT, Gemini, or whatever AI platform you already use.

Here is the simplest possible path to get started:

If you use Claude (claude.ai):

1. Go to claude.ai and sign in.

2. Click Projects in the left sidebar, then New Project. Give it a short, recognizable name – something like PCR Narratives or My Documentation Assistant. You will come back to this Project every time you document a call, so name it something you will recognize instantly.

3. In the Project, open Settings and paste the full contents of dist/claude-project-instructions.md (everything below the horizontal rule) into the Instructions field.

4. In the Project’s Files section, upload all 17 files in dist/paramedic-narrative/references/ (the WHEN-*.md files; choose UTF-8 if you’re asked about encoding).

   That is 17 uploads where earlier versions of this tool needed 2. It is a one-time setup, and it exists on purpose: a Project’s file retrieval is chunked and not guaranteed to pull every relevant file into a given turn, so a single large standard risks a safety rule simply not reaching the model on the turn it was needed. Splitting the standard into 17 files, each named for the exact situation that requires it, gives retrieval a much better chance of surfacing the right rule when it matters. If you’d rather do this once and be done with it, see the Skill package option below – it wraps all 17 files into a single upload.

   Do not paste dist/system-prompt.md into a Claude Project. It inlines everything the 17 files contain, and the Instructions field is re-sent with every message, so you would pay for the whole standard on every turn. dist/claude-project-instructions.md carries only the rules that must be in context at all times and relies on the Project’s files for the rest. dist/system-prompt.md is for platforms with no file retrieval.

5. Start a new conversation in the Project and say “I want to set up my provider profile” – the skill asks a few questions and remembers your name, credential, and agency for every future session.

6. That is it. Come back to this same Project whenever you need to document a call.

This works identically whether you’re on the Claude iOS app, the Android app, or a browser. Some accounts also have a separate Skills feature under Project Settings – if yours does, you can instead download dist/paramedic-narrative.skill from the Releases page and upload it there as a single file. If there is no Skills option, or the uploaded file shows up as unreadable text instead of an installed skill, your account doesn’t have that feature – use the steps above.

If you use ChatGPT:

ChatGPT Custom GPTs have a character limit on the Instructions field. Paramedic-Narrative keeps only the rules that must never depend on retrieval in the Instructions field, and puts the rest in 17 Knowledge Files, each covering one topic, retrieved when that topic comes up.

1. On this page, go to dist/chatgpt-instructions.md. Click the copy icon.
2. In ChatGPT, go to Explore GPTs > Create > Configure and paste the copied text into the Instructions field.
3. Still in the Configure panel, scroll down to Knowledge and click Upload files.
4. Download all 17 files in dist/paramedic-narrative/references/ from this repository and upload them as Knowledge Files. This gives the GPT access to the full clinical documentation standard, retrieved by topic instead of all at once.
5. Give it a recognizable name – Paramedic Narratives or PCR Assistant.
6. Click Save. Use that Custom GPT for every documentation session.
7. In your first conversation, say “I want to set up my provider profile” – paste the resulting file below the Instructions field or upload it as an additional Knowledge File.

If you use Google Gemini:

1. Copy the contents of dist/system-prompt.md as described above.
2. In Gemini, go to Gems > New Gem.
3. Paste the copied text into the instructions field.
4. Name it something recognizable – PCR Narratives or Paramedic Documentation.
5. Save. Use that Gem for every documentation session.
6. In your first conversation, say “I want to set up my provider profile” and paste the resulting file below your instructions.

Gemini Gems have no file-retrieval mechanism, so this is still a single-file setup – dist/system-prompt.md inlines the always-loaded block and all 17 topic files into one document. Only the platforms with retrieval (Claude Projects, ChatGPT) take the 17-file route above.

If none of that makes sense yet: go to paramedicfoundation.org/resources/ai-documentation for plain-language instructions.

---

Using this tool on your phone or without a work computer

Paramedic-Narrative works fully on a phone. Claude, ChatGPT, and Gemini all have free mobile apps for iOS and Android and all work in a phone browser without an app. Once the tool is set up on your phone, you can use it anywhere.

Important policy notice: The Paramedic Foundation does not condone, encourage, or authorize the violation of any agency policy, jurisdictional regulation, employer requirement, or regulatory prohibition on the use of AI tools in clinical settings. Providers are solely responsible for knowing and complying with all applicable policies before using this or any AI tool. If your agency, Chief Paramedic, or jurisdiction prohibits AI-assisted documentation tools, do not use this tool in that context. If you are unsure whether your use is permitted, ask your supervisor or Chief Paramedic before proceeding.

Describing your call – voice input:

You do not have to type anything. Every AI platform has a microphone button you can tap to speak your call description out loud – exactly the way physicians have used transcription services for decades. Tap the microphone, describe the call the way you would tell it to a colleague, and the AI will work from what you said. This is often faster and more natural than typing, especially at the end of a shift.

Adding clinical data – camera input:

You can take a photo and add it to the conversation using the camera button in the AI app. This works well for:

• Vital signs printouts from your monitor
• 12-lead ECG strips
• Medication lists or medication administration records
• Discharge summaries from a transferring facility
• Printed protocol or procedure reference cards
• Handwritten run sheet notes

PHI warning – read this before using the camera: Before photographing anything, check that the image does not contain the patient’s name, date of birth, address, medical record number, Social Security number, or any other information that identifies a specific person. Crop or cover those fields if present. Camera images also contain hidden metadata including GPS coordinates that may itself constitute PHI – be aware of where you are when you take clinical photos. The Paramedic Foundation is not responsible for privacy breaches resulting from provider conduct. See Section 6 of ETHICS.md for the full standard.

Copying a draft to your ePCR:

Once you have reviewed and approved the narrative draft:

1. Select all the narrative text and copy it.
2. Paste it directly into your ePCR platform – most platforms (ESO, ImageTrend, Zoll, and others) have mobile apps where you can paste narrative text.

Email workaround for constrained work environments:

If you complete documentation on a personal device and need to transfer it to a work computer:

1. Review and fully approve the narrative on your phone.
2. Copy the narrative text and paste it into an email to your work address.
3. Open the email on your work computer and paste the narrative into your ePCR.

If you use this workflow, do not include any patient identifiers in the AI session. Use a case number, initials, or generic descriptors (“the patient,” “a 68-year-old female”) rather than names, dates of birth, or other direct identifiers. You are responsible for ensuring this workflow complies with your agency’s privacy and data handling policies.

---

Installation

Why 17 files, not one

Starting with version 3.0.0, every platform with file retrieval (Claude Projects, ChatGPT Custom GPTs) is set up with 17 separate topic files instead of one or two combined reference files. That is more uploading – say so plainly. It is also a one-time cost, done once per Project or GPT, never repeated.

It exists because retrieval from uploaded files is chunked and not guaranteed: a platform’s retrieval system decides on each turn which parts of which files are worth pulling into context, and a single monolithic file competes with itself for that decision. A safety-critical rule buried in the middle of one large document can simply not be the part retrieval selects on the turn it was needed. Seventeen files, each opening with "CONSULT THIS FILE WHEN: " and named for the situation that requires it (WHEN-MEDICATION.md, WHEN-FORENSIC.md, and so on), give retrieval a much narrower, much more legible target to match against.

Two things soften the cost. First, the rules that must never depend on retrieval at all – the editorial-tool boundary, the ten NEVER rules, the attribution boundary, PHI, and the rest of ALWAYS-BLOCK.md – are pasted directly into the Instructions field on every platform, so they are never subject to retrieval in the first place. Second, Claude users can skip the 17-file upload entirely: the .skill package (dist/paramedic-narrative.skill) bundles the always-block and all 17 files into a single upload, and Claude’s Skill mechanism handles retrieval from the bundle internally. Gemini, direct API, and Copilot’s per-session paste have no retrieval at all, so they were never affected by this change – they get everything inlined into one file (dist/system-prompt.md) either way.

Claude

Claude.ai Projects support two setup methods. Which one is available depends on your account and plan, not on your device.

Skill package (single upload, only if your account has this feature):

1. Download dist/paramedic-narrative.skill from the Releases page
2. In the Project, look for a Skills option under Project Settings
3. If you find it, upload the .skill file there – this one file already contains the always-loaded block, all 17 topic files, and the templates
4. If there is no Skills option, or the uploaded file shows up as unreadable or garbled text instead of an installed skill, your account does not have this feature – use Files + Instructions below instead

Files + Instructions (works on every account):

1. In Claude.ai, open or create a Project, and go to Project Settings
2. Paste the full contents of dist/claude-project-instructions.md from this repository into the Instructions field
3. Download all 17 files in dist/paramedic-narrative/references/ from this repository
4. Upload them to the Project’s Files section (choose UTF-8 if asked about encoding)
5. Optionally upload your agency documentation standard as a Project file (see Configuration)

Do not paste dist/system-prompt.md into a Claude Project’s Instructions field – see Why 17 files, not one above for why the two files serve different platforms.

ChatGPT (Custom GPT)

ChatGPT Custom GPTs have a character limit on the Instructions field. Paramedic-Narrative keeps only the rules that must never depend on retrieval in Instructions, and puts the full standard in 17 Knowledge Files, retrieved by topic.

Instructions field:

1. Open dist/chatgpt-instructions.md in this repository and copy its full contents
2. In ChatGPT, go to Explore GPTs > Create > Configure
3. Paste the contents into the Instructions field

Knowledge Files: 4. Download all 17 files in dist/paramedic-narrative/references/ from this repository 5. In the Configure panel, scroll to Knowledge and click Upload files 6. Upload all 17 files – this gives the GPT the full clinical documentation standard, retrieved one topic at a time instead of as one large document

Finish setup: 7. Optionally paste your agency configuration below the instructions, or upload it as an additional Knowledge File 8. Give the GPT a recognizable name and save

Why Instructions plus Knowledge Files, and why 17 of them? The Instructions field holds the behavioral rules that must always be in effect – how the tool thinks and what it refuses to do, identical to every other platform. The 17 Knowledge Files hold the clinical reference content – scoring tools, the medication documentation standard, the forensic standard, vital thresholds, and handoff frameworks – split by topic instead of combined, so retrieval has a narrower, better-named target on each turn. See Why 17 files, not one above for the full rationale. The Knowledge Files are consulted during conversation without counting against the Instructions character limit.

Google Gemini (Gem)

1. Copy the full contents of dist/system-prompt.md from this repository
2. In Gemini, go to Gems > New Gem
3. Paste the contents into the instructions field
4. Optionally paste your agency documentation standard below
5. Save the Gem

Gemini Gems have no file-retrieval mechanism, so this remains a single-file setup: dist/system-prompt.md inlines the always-loaded block and all 17 topic files.

Microsoft Copilot (M365)

Microsoft 365 Copilot supports partial setup through Custom Instructions. The full reference standard must be pasted at the start of each session.

One-time setup (Custom Instructions):

1. Open dist/chatgpt-instructions.md in this repository and copy its full contents
2. In Copilot, click the three-dot menu (top right) > Settings > Personalization > Edit instructions
3. Paste the contents and click Save instructions

Per-session (full reference): 4. Go to dist/system-prompt.md in this repository, click Raw, select all (Ctrl+A / Cmd+A), and copy 5. At the start of each Copilot conversation, paste the text with the note: “This is your documentation standard for this session.” 6. Then describe your call

The behavioral rules persist via Custom Instructions. The full reference is pasted each session because the Personalization > Custom Instructions path used above has no persistent knowledge-upload feature equivalent to ChatGPT Custom GPTs. dist/system-prompt.md is used for the per-session paste, in place of SKILL.md in earlier versions, because it is the single file with the full standard already inlined – matching a platform that has no retrieval to split work across. Microsoft 365 Copilot separately supports declarative agents (built via Copilot Studio / Agent Builder) that do support persistent knowledge sources alongside instructions – if your organization has that capability available and wants a one-time setup instead of a per-session paste, a declarative agent may be a better fit than the workflow above; that setup is not yet documented here.

Note: Some organizations restrict Copilot customization through IT policy. If the Custom Instructions field is unavailable, paste both dist/chatgpt-instructions.md and dist/system-prompt.md at the start of each session.

Mobile (iOS and Android)

Paramedic-Narrative works fully on a smartphone. Claude, ChatGPT, and Gemini all have free mobile apps and also work in any phone browser without an app.

Getting files from GitHub on your phone:

The GitHub mobile app does not support raw file downloads. Use your phone’s browser (Safari on iOS, Chrome on Android) instead:

1. Go to github.com/The-Paramedic-Foundation/paramedic-narrative-skill in your phone browser
2. Navigate to the file you need
3. Tap the file name to open it
4. Tap Raw (top right of the file view) – this opens the plain text
5. Tap and hold anywhere in the text > Select All > Copy (or use your browser’s share sheet to save to Files/Downloads)

Claude on mobile: Use the Files + Instructions method (see Installation above) – it works identically on the Claude iOS app, the Android app, and any browser, because Project Files and Instructions are both available from the phone app itself. No desktop or browser detour is required for this method.

1. In the Claude app, open or create your Project and go to Project Settings
2. Paste dist/claude-project-instructions.md into Instructions
3. Download the 17 files in dist/paramedic-narrative/references/ using the phone browser steps above, then upload them to the Project’s Files section from the app. This is the most tedious step on a phone – 17 individual downloads instead of 1 – but it is one-time
4. Your agency config uploads the same way, directly from the Claude app

The Skill-package method (.skill upload) is a single file instead of 17 and is a separate, account-dependent feature. If you’ve confirmed your account has it, the upload step itself still needs a browser (see the note in Installation); the Claude app can then use the Project normally afterward.

ChatGPT on mobile: Creating and editing a Custom GPT’s Instructions and Knowledge Files is a web-only feature – the ChatGPT mobile app can only use a GPT that already exists, not create or edit one. Do the setup itself in your phone’s browser, then use the ChatGPT app afterward.

1. In your phone browser, go to chatgpt.com, sign in, and go to Explore GPTs > Create > Configure
2. In a second browser tab, open dist/chatgpt-instructions.md in this repository, tap Raw, select all, copy, and paste it into the Instructions field
3. Scroll to Knowledge, tap Upload files, and upload each of the 17 files in dist/paramedic-narrative/references/ in turn (download each from this repository first if your browser requires a local file) – this is the most tedious step on a phone, but it is one-time
4. Save the GPT. From then on, open the ChatGPT app and use this GPT normally – the app is only unable to edit it, not use it

Gemini on mobile: Creating, editing, or deleting a Gem is a web-only feature – the Gemini mobile app can only use a Gem that was already created at gemini.google.com. Do the setup itself in your phone’s browser, then use the Gemini app afterward.

1. In your phone browser, go to gemini.google.com, sign in, and go to Gems > New Gem
2. In a second browser tab, open dist/system-prompt.md in this repository, tap Raw, select all, copy, and paste it into the instructions field
3. Save the Gem. From then on, open the Gemini app and use this Gem normally – the app is only unable to create or edit it, not use it

Copilot on mobile:

1. In the Copilot app, tap your profile > Settings > Personalization > Edit instructions > paste contents of dist/chatgpt-instructions.md > Save
2. At the start of each session: open dist/system-prompt.md in your phone browser, tap Raw, select all, copy, paste as your first message in Copilot

Using the tool on shift:

• Tap the microphone and describe the call out loud – no typing required
• Use the camera to photograph clinical data (vital sign printouts, 12-lead strips, medication lists) – confirm no PHI visible before photographing
• Copy the approved narrative and paste into your ePCR mobile app

Any Other LLM Platform

dist/system-prompt.md is formatted as a plain system prompt compatible with any AI platform that accepts system-level or custom instructions. Paste it into the appropriate field for your platform.

---

Configuration

For individual providers

This skill works out of the box using universal paramedicine documentation standards. If your agency has completed an agency configuration file (see below), download it from wherever your agency hosts it and upload it to your Claude Project alongside the skill. For ChatGPT or Gemini, paste its contents below the system prompt.

If your agency has not yet created a configuration file, you can provide basic context at the start of any session:

“I work for [Agency] in [State]. We use [ePCR platform]. Our protocols are [protocol system name]. My Chief Paramedic is [name].”

The skill will apply that context for the session.

For ChatGPT specifically: paste your agency configuration below the instructions in the Configure panel, or upload it as an additional Knowledge File alongside the 17 WHEN-*.md files. Both approaches work.

For agencies and Chief Paramedics

The repository includes a standardized agency configuration template that Chief Paramedics and agency administrators can complete once and distribute to all providers. When every provider uses the same configuration file, every narrative produced in your system reflects your protocols, your ePCR platform, your documentation standard, and your Chief Paramedic’s clinical expectations automatically.

The template file is: agency-config-template.md

It covers:

• Agency identity and service area context
• Chief Paramedic endorsement and scope of authorization
• ePCR platform and narrative field specifications
• Documentation standard and minimum narrative requirements
• Protocol system name, version, and CPG sources adopted by Chief Paramedic
• Specific protocol titles for common call types
• Controlled substance policy including witness, waste, and reconciliation requirements
• Optional prompt settings – turn specific skill prompts on, off, or required
• Transfer of care standards and receiving facility list
• Alternate destination authorization
• Privacy and data handling policy
• Version history for configuration updates

How to set it up:

1. Download agency-config-template.md from this repository
2. Open it in any text editor (Notepad, TextEdit, Word set to plain text)
3. Fill in each section – incomplete sections are fine, the skill applies universal standards for anything left blank
4. Have your Chief Paramedic review and complete Section 2 (endorsement)
5. Save the completed file as agency-config.md
6. Post it somewhere all your providers can download it – your agency intranet, shared drive, ePCR document library, or a protected page on your website
7. Tell providers to download it and upload it to their Claude Project, or paste its contents below the system prompt in ChatGPT or Gemini
8. When your protocols or policies change, update the file, increment the version number in Section 11, and notify providers to download the new version

Contact The Paramedic Foundation if you need assistance configuring the skill for your agency or would like consultation on AI documentation governance: [email protected]

---

Setting Up Your Provider Profile

Your provider profile is a personal file that tells the skill who you are, which agencies you work for, which roles you operate in, and any standing documentation preferences. You create it once and it is active for every session without having to re-enter your information.

To build it: say “I want to set up my provider profile” and the skill will guide you through a short conversation and produce the file automatically.

Alternatively, download provider-profile-template.md from this repository, fill it in, and save it as provider-profile.md.

To use it: upload provider-profile.md to your Claude Project alongside the skill. For ChatGPT or Gemini, paste its contents below the system prompt.

---

Working for Multiple Agencies

If you work for more than one agency, upload all your agency configuration files to your Claude Project at once. The skill detects all loaded configurations and knows which agencies are available to you.

At the start of a session, say which agency you are working for:

“I’m working an emergency shift for [Agency Name] today.”

To switch agencies mid-session:

“Switch to [agency short name]” or “I’m working for [Agency B] today.”

The skill confirms the switch, preserves your provider profile and preferences completely, and applies the new agency’s protocols and documentation standards.

---

Role Contexts

If you work in more than one clinical role, tell the skill which role is active at the start of each session. This changes which documentation framework, prompts, and disposition options apply.

Available roles:

• Emergency paramedic – 911 response. Full SOAP with all scoring tools, forensic standard, IMIST-AMBO handoff, ATLS trauma standard. Care pathway documentation for low-acuity calls and refusals.

• Rescue paramedic – Technical or special operations. Adds scene safety and hazard documentation, rescue mechanism, extrication detail, specialized equipment, and multi-agency role attribution.

• Community paramedic – CP/MIH visits. Shifts to longitudinal care framework: visit reason, functional status, medication adherence, resource connections, care plan. Alternative disposition is the primary outcome. Barriers to care always active.

• Hospital paramedic – CCT, interfacility transport, in-hospital response, or procedure support. Emphasizes transport indication and medical necessity, pre-transport stability, monitoring and interventions en route, and structured handoff to receiving team.

To activate a role:

“I’m doing community paramedicine today.” “This is a CCT run.” “I’m on rescue.” “Emergency shift.”

To switch roles mid-session:

“Switch to emergency mode for this call.”

The skill applies the emergency framework for that narrative and returns to your prior role context when complete.

---

How to Use It

Begin a session and describe your call. You can provide information in any format:

• A brief verbal summary: “55M, chest pain, 10/10, diaphoretic, pressure-like…”
• A structured list of findings and interventions
• A detailed walkthrough of the encounter

Never photograph or upload an image in which any direct patient identifier is visible. A document may be photographed only once every identifier on it has been cropped or covered before the shot; if it can’t be fully redacted first, dictate the content instead. See Section 6 of ETHICS.md.

The skill will identify what additional information is needed for a complete narrative, ask only for what is missing and narrative-relevant (not for information already captured in your structured PCR fields), and produce a draft in your agency’s declared narrative format (Clinical Summary / S / O / A / P by default).

Dictating a call: talk through the 12-point verbal report skeleton in any order – call frame, arrival picture, patient, story, pertinent negatives, exam highlights, numbers, thinking, doing, moving, handoff, exceptions. A printable pocket card, a wallet-size business card, and a mobile-reference image sized for a phone screen are all in docs/. Fragments are fine: you can document across multiple messages over hours, and the skill tracks what is still missing rather than re-interviewing you. Documenting long after a call? Say so – the skill switches to targeted recall questions built from what it already knows.

Every draft ends with a mandatory provider review disclaimer. Review the draft carefully, correct any errors, verify all [VERIFY] items, and approve before submitting.

Example session

“Help me write a narrative. 68F, found unresponsive by family, last known well approximately 2 hours ago per family at bedside. GCS 8 on arrival. Cincinnati positive. BGL 112. Transported priority to Regional Stroke Center.”

The skill will ask targeted follow-up questions about ABC/LOC quality, clinical reasoning, transport rationale, and transfer of care, then produce the draft.

---

Controlled Substance Documentation

When a call involves controlled substance administration, the skill applies a strict audit trail standard. It will ask for:

• Medication source and container identifier
• Witness identity and credential
• Amount drawn vs. amount administered
• Waste amount, method, and witness to waste
• Chain of custody for any remainder
• Reconciliation if performed

It will not fabricate any of these elements. Missing audit trail items are always flagged [VERIFY]. See Section 7.4 of ETHICS.md for the governing ethical standard.

---

Forensic Documentation

When a call involves suspected criminal activity, victimization, or harm, the skill automatically applies an elevated evidentiary standard:

• Source attribution for every factual statement about who did what to whom
• Verbatim patient and witness quotes where possible
• Observation distinguished from inference throughout
• Law enforcement presence and identifiers documented
• Chain of custody for any items transferred
• Neutral, non-legal-conclusion language throughout

Call types that trigger this standard include: assault, domestic violence, sexual assault, child or vulnerable adult abuse, suspicious death, gunshot or stab wound, suspected non-accidental trauma, intoxication-related harm, arson, motor vehicle collision with potential impairment or fatality, threats, and any scene where law enforcement is investigating.

---

Multi-Platform Files

All renderings share the identical always-loaded block – build.py --verify fails the build if they do not match byte-for-byte. Only the 17 topic files are consulted through retrieval rather than read in full on every turn, on any platform with file upload (Claude Project Files, ChatGPT Knowledge Files), so behavior there is not guaranteed identical to Gemini or a direct-API session, where everything in system-prompt.md is inlined. This is why the always-loaded block itself – privacy, the ten NEVER rules, the attribution boundary, controlled substance and forensic non-fabrication – is pasted directly into every platform’s Instructions field, rather than left to retrieval alone.

---

Ethical Framework

Read ETHICS.md before clinical use. It covers:

• The governing principle (augmentation, not delegation)
• Disclaimer of warranty and limitation of liability
• Absolute prohibition on clinical decision use
• Privacy and PHI requirements in detail
• Provider ethical commitments
• Equity obligations
• Organizational governance guidance
• Version control and contribution process

This framework is grounded in Nudell, N.G. (2026). Clinical governance in the age of artificial intelligence: A profession-wide imperative for paramedicine. Governing Care. The Paramedic Foundation / American College of Paramedics.

---

Version Control

Current version: 3.0.0

3.0.0 – File-layout restructure (breaking change). Replaces the single paramedic-narrative/SKILL.md and its two reference files with two hand-edited sources: ALWAYS-BLOCK.md, containing only the rules that must never depend on retrieval (the editorial-tool boundary, the ten NEVER rules, the attribution boundary in short form, PHI, patient/incident separation, during-a-call guardrails, the pre-draft check, style, the standing disclaimer, and a router table), and knowledge/, 17 topic files each named for the situation that requires it (WHEN-MEDICATION.md, WHEN-FORENSIC.md, and so on) and opening with "CONSULT THIS FILE WHEN: " / “DO NOT CONSULT OTHERWISE.” Every file a provider actually installs – system-prompt.md, chatgpt-instructions.md, claude-project-instructions.md, and paramedic-narrative/SKILL.md – is now generated from those two sources by build.py rather than hand-maintained separately. A build verifier (build.py --verify) fails the build if the always-loaded block is not byte-identical across every rendering, if a knowledge/ file is missing from the router or the router points at a file that does not exist, if the ChatGPT rendering exceeds 6,500 characters, or if house style is violated. This is the fix for the whole class of defect found in the 3.0.0 audit, where three hand-maintained paraphrases of one standard drifted apart and lost safety-critical content – with generation from a single source, that class of defect can no longer occur, because there is only one place to make each change. documentation-standards-primer.md and narrative-formats.md no longer exist as separate files; their content, which had duplicated across the two on seven topics, has been deduplicated into the 17 topic files. This is a breaking change for every 2.x installation: the old files must be removed, not left alongside the new ones. See Updating an existing installation below before you upload anything.

3.0.0 – Cross-file parity corrections. No change to clinical content, thresholds, or standards, so agencies that have reviewed 2.1.0 carry no re-verification burden. An audit of all instruction files found that safety-critical text had been paraphrased into the platform-specific renderings rather than reproduced, and the paraphrases had lost content. Corrected: chatgpt-instructions.md regains the full eleven-item forensic trigger list (it had carried five, omitting domestic violence and sexual assault among others), the complete standing disclaimer including the AI-tool and no-clinical-advice sentences, the camera-metadata PHI rule, the container-identifier and reconciliation elements of the controlled substance audit trail, the pre-draft verification step as an explicit workflow step, and the rule against assuming your crew performed an act whose performer is unstated. system-prompt.md regains “threats” in the forensic trigger list and the neonate temperature threshold, and now uses “Chief Paramedic” rather than “medical director.” claude-project-instructions.md now carries incident and patient workspace isolation inline rather than only by reference to a project file, since retrieval is not guaranteed. The dictation pocket card and its printable variants gain the attribution prompts that 2.1.0 described but did not ship. Updating from 2.1.0 replaces the instruction and reference files only.

2.1.0 – Attribution and data-integrity boundary. Adds a core section covering care performed by another agency’s provider, care performed before the documenting crew arrived and assumed responsibility, and interventions prepared or considered but not performed. Structured ePCR entries are attributed entries and, where structured data feeds external reporting, become the record of what an agency provided; the three categories above therefore stay out of structured fields and are carried in full by the narrative. Core Operating Principle 1 now extends the non-fabrication rule to other clinicians’ clinical reasoning. Core Operating Principle 5 now requires confirming that a structured entry exists before referencing it, which closes a real failure mode: a narrative that says “as charted” for an intervention the agency’s own standard forbids charting leaves that intervention documented nowhere. Related changes in the medication administration standard, workflow prompts, dictation pocket card, narrative formats, primer, and a new agency configuration Section 4A. Updating from 2.0.x replaces the instruction and reference files only; provider profiles and existing agency configurations remain valid, though agencies should complete the new Section 4A.

Version history and release notes are maintained in this repository. Check Releases for updates. Providers and agencies using this tool in ongoing documentation workflows should review updates periodically.

Significant updates affecting clinical content, privacy standards, or the ethical framework will be clearly noted in release notes.

Updating an existing installation to a new version

Updates are not automatic on any platform. When a new version is released, each provider updates their own installation:

If you are updating from any 2.x version to 3.0.0, read this first. The file layout changed: paramedic-narrative/SKILL.md, documentation-standards-primer.md, and narrative-formats.md no longer exist. They are replaced by the Instructions content generated from ALWAYS-BLOCK.md and the 17 WHEN-*.md files in dist/paramedic-narrative/references/. Before uploading anything new, remove every file from the old layout – the old SKILL.md, the old documentation-standards-primer.md, and the old narrative-formats.md – from your Project Files or Knowledge Files. Do not leave the old files in place alongside the new ones: if both are present, retrieval may surface the retired copy instead of the current one, and the two will disagree.

Claude:

1. In your Project’s Instructions field, paste the new dist/claude-project-instructions.md over whatever is there now. If your Project currently holds the full dist/system-prompt.md in Instructions, replacing it with this file is part of the upgrade: same standards, a fraction of the per-message cost.
2. In Files, remove the old SKILL.md, documentation-standards-primer.md, and narrative-formats.md if present, then upload all 17 files from dist/paramedic-narrative/references/ – don’t leave the old files in place, or the assistant may consult the outdated copy.
3. Your provider profile, agency configuration files, and past conversations are unaffected – only the instructions and reference files are replaced.
4. This works the same on the Claude iOS app, the Android app, or a browser.

If your account uses the Skill package instead: download the new dist/paramedic-narrative.skill from the Releases page, remove the old skill under Settings > Skills, and upload the new file. This step requires a browser (web or desktop).

ChatGPT (Custom GPT):

1. Copy the new dist/chatgpt-instructions.md and paste it over the old contents of the Instructions field in your GPT’s Configure panel.
2. In Knowledge, delete the old SKILL.md (and documentation-standards-primer.md / narrative-formats.md if you had uploaded them separately), then upload all 17 files from dist/paramedic-narrative/references/. Do not leave the old files in place – the GPT may consult the outdated copy.
3. Save the GPT.

Gemini (Gem):

1. Copy the new dist/system-prompt.md and paste it over the old contents of your Gem’s instructions field. Re-paste your agency configuration below it if you had one. Save.

Microsoft Copilot:

1. Update Custom Instructions with the new dist/chatgpt-instructions.md contents.
2. Use the new dist/system-prompt.md for your per-session paste going forward, in place of SKILL.md.

Agency administrators: when you distribute a new agency configuration file version, the same rule applies – providers must replace the old file in their Claude Project or re-paste it on other platforms. Increment the version in Section 11 of the configuration and notify providers.

How to tell which version you are running: ask the assistant “what version are you?” – the version is stated in the skill and system prompt – or compare against the version shown at the top of this README.

---

Contributing

Contributions are welcome from providers, agencies, educators, Chief Paramedics, and researchers. See Section 11 of ETHICS.md for full guidance.

Ways to contribute:

• Clinical corrections: Open a GitHub Issue with your proposed correction and reasoning
• Agency configurations: Share de-identified agency-config.md examples for specific ePCR platforms
• Platform support: Configuration guidance for additional AI platforms
• International adaptations: Non-US paramedicine frameworks and documentation standards
• Research and evaluation: Systematic evaluations of tool outputs

All contributions are reviewed by The Paramedic Foundation before incorporation.

If you are opening a pull request

dist/ is entirely generated. Every file in it – system-prompt.md, chatgpt-instructions.md, claude-project-instructions.md, paramedic-narrative/SKILL.md, and paramedic-narrative/references/ – is produced by build.py and overwritten the next time it runs. Pull requests should never edit anything under dist/ directly.

Instead:

• Edit ALWAYS-BLOCK.md for a rule that must apply on every platform.

• Edit the relevant file in knowledge/ for a topic-specific rule.

• Then run, from the repository root:

  python3 build.py && python3 build.py --verify
  

python3 build.py regenerates everything in dist/. python3 build.py --verify must pass before a pull request is submitted – it fails the build if the always-loaded block is not byte-identical across every rendering, if a knowledge/ file is missing from the router table in ALWAYS-BLOCK.md or the router points at a file that does not exist, if the ChatGPT rendering exceeds its character budget, or if house style is violated (no em dashes, “Chief Paramedic” never “medical director”, “ePCR platform” never “PCR platform”).

---

Citation

If you use or adapt this skill in research, policy work, or publications:

Nudell, N. G. (2026). paramedic-narrative-skill: AI-assisted PCR narrative documentation for paramedics and EMTs (Version 3.0.0) [Software]. The Paramedic Foundation. https://github.com/The-Paramedic-Foundation/paramedic-narrative-skill

---

License

Creative Commons Attribution 4.0 International (CC BY 4.0)

You may share and adapt this material for any purpose, including commercial use, provided you give appropriate credit to The Paramedic Foundation and link to paramedicfoundation.org.

---

The Paramedic Foundation · 23 W Central Entrance PMB 321 · Duluth, MN 55811 paramedicfoundation.org · [email protected] · EIN 46-3271401 · 501©(3)