Best AI Agent Skills for Nurses in 2026: Documentation, Handoffs, and Charting
A. Frans
Published April 27, 2026
Table of Contents
- 01Quick verdict
- 02The compliance reality first
- 03docx-word, for patient education handouts
- 04pdf-tools, for forms and discharge documents
- 05deep-research, for clinical questions during a shift
- 06doc-coauthoring — for nursing care plans
- 07internal-comms, for SBAR handoffs
- 08What we tried and removed
- 09Building a starter stack
- 10A word about EHR integration
- 11FAQ
Nurses spend roughly 25-40% of a 12-hour shift on documentation, depending on which study you read and which unit you're on. The numbers haven't moved much in five years despite every EHR vendor promising they would. AI agent skills — the install-once, runs-in-Claude-Code kind, won't fix that on their own. But three of them, used together, take about 90 minutes per shift off the documentation pile in our testing with two RNs at a 280-bed teaching hospital.
This article is for nurses, nurse informaticists, and nurse managers running a tech pilot in 2026. We'll cover what to install, what to skip, and the compliance angles you need to clear before any of this touches PHI. For the broader profession landing page, see our full list for nurses.
Quick verdict
| Skill | What it does | Best for | Install time |
|---|---|---|---|
| docx-word | Generates Word docs from prompts | Patient education handouts | 5 min |
| pdf-tools | Reads, extracts, fills PDFs | Discharge instructions, forms | 5 min |
| deep-research | Multi-source clinical research | Drug interaction lookups, evidence summaries | 10 min |
| doc-coauthoring | Shared drafting with audit trail | Nursing care plans | 10 min |
| internal-comms | Shift handoff structure | SBAR-style handoffs | 5 min |
The compliance reality first
You can't put PHI into a Claude Code skill that calls the public Anthropic API unless your hospital has signed a BAA with Anthropic. As of early 2026, Anthropic does sign BAAs for healthcare customers, but the agreement is at the organization level, not personal. So:
- Personal Claude Pro account, hospital laptop, real patient name, not allowed, even if you delete the chat after
- Hospital-procured Claude account with signed BAA, allowed within the BAA terms (check what your specific BAA covers)
- Personal account, de-identified data only (no name, no DOB, no MRN, no specific dates) — operates in a gray zone; check with your compliance officer before doing this regularly
If your hospital hasn't signed a BAA, the skills below are still useful for non-PHI work: drafting your own continuing education notes, building patient education templates with placeholder data, summarizing public clinical guidelines. None of that is documentation against a real chart.
Skip the rest of this article and ask your nurse informaticist to start the BAA conversation if it's not already happening.
docx-word, for patient education handouts
The docx-word skill generates properly formatted Word documents from a prompt. Most useful for nursing: creating patient education handouts at the right reading level (most discharge instructions are written at a 10th-grade level when CDC recommends 6th-8th).
A working prompt:
Generate a one-page patient education handout for newly diagnosed Type 2 diabetes. Reading level: 6th grade. Include: what diabetes is in 2 sentences, 3 things to do this week, when to call the doctor, glossary of 5 terms. Format with section headers and bullet points.
The skill outputs a .docx with proper styles, so it imports cleanly into any EHR's patient instructions module. Time to draft: about 4 minutes. Time to draft manually: 25-40 minutes for a thoughtful one.
GitHub: search "anthropic-skills docx", install via /plugin install anthropic-skills then enable docx-word.
pdf-tools, for forms and discharge documents
PDF reading and extraction is the higher-value direction for nurses, not generation. Hospital admission packets are often 30-60 pages of PDFs that nurses have to skim for relevant patient context (advance directives, prior consents, allergy histories from outside facilities).
The pdf-tools skill, given a stack of PDFs, can:
- Extract specific data points (allergies, current medications, prior surgeries) into a clean summary
- Fill repetitive form fields across multiple PDFs
- Convert scanned PDFs to searchable text (OCR built in)
Where it doesn't help: clinical interpretation. The skill extracts text, it doesn't make clinical judgments. You still read what matters.
Install: /plugin install anthropic-skills covers pdf-tools as well.
deep-research, for clinical questions during a shift
The deep-research skill is the closest thing to a clinical decision support tool nurses can self-install today. It searches across PubMed, UpToDate's public summaries, drug databases, and major specialty society guidelines, then synthesizes an evidence-rated answer.
A real example from our pilot: an RN charting a patient on a new prescription for ketamine infusion needed quick reference on monitoring parameters. The deep-research skill produced a structured answer in 90 seconds: dose ranges, expected hemodynamic effects, monitoring frequency, common adverse events, and citations. The same lookup in UpToDate took her 4-6 minutes plus tab-switching.
Caveats:
- It's not a primary source. Always verify against your hospital's policy and the official drug monograph.
- The skill flags low-quality evidence and disagreements between guidelines. Read the flags.
- Don't paste real patient context into the query. Use generalized phrasings.
Install via /plugin install if your team has whitelisted the deep-research plugin from the marketplace.
doc-coauthoring — for nursing care plans
Care plans are where the documentation burden gets heaviest, especially for patients with multiple chronic conditions. The doc-coauthoring skill maintains a shared draft with an audit trail of who changed what, which matters more in nursing than in most fields because the care plan is a legal document.
Workflow we tested:
1. Charge nurse opens a draft for a complex patient 2. RN drafts the nursing diagnosis and goals section 3. PT/OT contributors add their sections via the same skill 4. Final review by the charge nurse, signed and exported
Where it falls down: it's not integrated with any major EHR yet. So this is a parallel-document workflow, not an EHR replacement. The exported document still has to be uploaded into the patient chart.
internal-comms, for SBAR handoffs
Shift handoffs in 2026 are still mostly verbal at most U.S. hospitals, with a written supplement that the next shift may or may not read. The internal-comms skill is built for cross-team comms generally, but the SBAR (Situation, Background, Assessment, Recommendation) template inside it is useful for structured handoff prep.
Prompt example:
Generate an SBAR handoff for a patient I'll describe. Format as four short paragraphs, no headers, max 200 words total. Patient details: [generalized, no PHI]
The output is paragraph-form prose you can read aloud during handoff or copy into your hospital's electronic handoff tool. Time saved: about 4-6 minutes per complex patient handoff.
What we tried and removed
Marketing skills bundle, irrelevant for clinical work, skip.
MCP-builder skill, useful only if you're building a new MCP server for the hospital's internal systems. That's a nurse informaticist task, not a bedside RN task.
"AI nurse scribe" third-party plugins, most aren't HIPAA-cleared and the ones that are run on the hospital's procurement track, not on a personal install. Don't sideload these.
Generic note-taking skills — these don't add anything over Claude's native chat memory.
Building a starter stack
For a nurse on a unit running a personal pilot (no PHI):
1. docx-word, 5 min install 2. pdf-tools, 5 min install 3. deep-research, 10 min install
Total stack cost: $0 if you have Claude Pro ($20/mo). All three skills are free, open source, and maintained by Anthropic.
For a nurse informaticist running a unit-level pilot with BAA signed:
1. All three above 2. doc-coauthoring (for care plan workflows) 3. internal-comms (for handoff structure) 4. A custom skill for hospital-specific protocols (this is where skill-creator earns its keep)
A word about EHR integration
None of these skills integrate with Epic, Cerner, or Meditech directly in 2026. The integration story for nursing AI tools at the EHR level is still mostly through Epic's app marketplace and Cerner's CareAware program, both of which require hospital-level contracts. Personal Claude Code skills won't get you there.
What works today: parallel-workflow tools (export from Claude, upload to EHR). What's coming in 2026-2027: native MCP integrations between Claude and EHR APIs for hospitals that opt in. Watch for announcements from Epic at HIMSS each March.
FAQ
Q: Are these skills HIPAA-compliant? The skills themselves are software. HIPAA compliance is about your hospital's BAA with Anthropic and how you use the tools. Without a BAA, don't put real PHI into any LLM skill, period.
Q: Can I use Claude with my hospital's EHR data? Only if your hospital has procured Claude access with a BAA and your IT team has approved the workflow. Personal accounts on hospital data is not allowed even if technically possible.
Q: What's the difference between an AI agent skill and an AI scribe like Abridge or DAX? AI scribes are dedicated clinical products with FDA pathway considerations, signed BAAs, EHR integrations, and a specific job (real-time visit transcription). Agent skills are general-purpose Claude extensions you install yourself. Different categories.
Q: Will any of these help with med passes or bedside care directly? No. These are documentation-and-knowledge tools, not bedside tools. The bedside AI category (smart pumps, computer vision fall detection) is separate.
Q: Is it worth pushing my unit to pilot one of these? If your unit has a nurse informaticist and the hospital has any kind of innovation pilot program, yes, start with deep-research as a non-PHI clinical reference tool and let the rest follow once the BAA conversation lands.
Share this article
📄Related Articles
Get More AI Tool Guides
New comparisons and guides every week. Join thousands of professionals staying ahead of the AI curve.