Skip to main content
Guide9 min read·Updated June 9, 2026
🧩

Best AI Agent Skills for Doctors in 2026

B

A. Frans

Published June 9, 2026

AI Agent SkillsDoctorsHealthcareClaude SkillsMedical AI

Most of a doctor's day that isn't seeing patients is reading and writing: scanning literature, drafting referral letters, reconciling spreadsheets of lab values, summarizing a guideline that just changed. AI agent skills target exactly that load, and unlike a medical scribe product, you can install several for free in an afternoon.

A quick boundary before the list. None of these is a regulated clinical decision support system, and none should touch identifiable patient data unless your organization runs it under a signed Business Associate Agreement in a HIPAA-compliant setup. Everything here helps with the work around clinical decisions, not the decisions themselves. For patient-facing and clinical products, see our full list of AI tools for doctors.

The skills at a glance

SkillWhat it doesBest forInstall
deep-researchMulti-source literature synthesis with citationsStaying current on evidence~10 min
pdf-toolsRead, merge, extract from PDFsJournal articles, scanned forms~5 min
docx-wordGenerate and edit Word documentsReferral letters, reports~5 min
xlsx-spreadsheetRead and analyze spreadsheetsLab data, audit tracking~5 min
scientific-skillsResearch, stats, and analysis toolkitAcademic clinicians, research~10 min

1. deep-research, for keeping up with the evidence

The hardest part of practising medicine is that the evidence moves faster than anyone can read. The deep-research skill runs a multi-step pipeline: it fans out across sources, pulls the actual papers, checks claims against each other, and returns a synthesis with citations you can verify. It's the closest a doctor can get to a self-installed evidence assistant.

Use it to summarize what changed in a guideline, to pull the recent trials on a drug before a tumor board, or to brief yourself on a condition outside your specialty. Always click through to the cited sources, because a synthesis is a starting point, not a primary reference.

Install: /plugin install the deep-research skill from its marketplace once your team has whitelisted it. Source and documentation live on GitHub under the deep-research project.

2. pdf-tools, the journal-and-paperwork workhorse

Medicine still runs on PDFs: journal articles, society guidelines, scanned consent forms, faxed records. The pdf-tools skill reads them, extracts text and tables, merges several into one file, and splits long documents into the pages you need.

The practical win is feeding a 40-page guideline PDF in and asking for the dosing table, or merging a stack of articles into a single reading file. It's part of Anthropic's own skill bundle, so it's one of the safer installs.

Install: /plugin install anthropic-skills, then enable pdf-tools.

3. docx-word, referral letters and reports without the formatting fight

Every doctor writes the same documents repeatedly: referral letters, discharge summaries, medico-legal reports, patient information sheets. The docx-word skill generates and edits Word files directly, with proper headings, tables, and letterhead, so you draft the content and skip the formatting.

A useful pattern: keep a de-identified template, then have the skill fill the clinical reasoning while you add the patient specifics by hand in your own secure system. Never paste identifiable details into the skill itself.

Install: ships in the Anthropic bundle, so /plugin install anthropic-skills and enable docx-word.

Clinicians end up in spreadsheets more than they expected: tracking a cohort's lab values, running a quality audit, reconciling a clinic's numbers. The xlsx-spreadsheet skill reads Excel files, computes formulas, builds charts, and answers questions about a sheet without you writing the formula.

For audit and research work on de-identified data, it turns an afternoon of spreadsheet wrangling into a few prompts. Keep the data anonymized before it goes anywhere near the skill.

Install: also in the Anthropic bundle alongside pdf-tools and docx-word.

5. scientific-skills, for the academic clinician

If you publish, sit on a research committee, or run trials, the scientific-skills bundle is built for you. It packages research, statistics, data analysis, and scientific writing helpers into one set, aimed at the analysis-heavy work that surrounds a paper or a grant.

This is overkill for a pure clinician but valuable for anyone whose week includes study design, stats, or manuscript drafting. Install it from its GitHub marketplace via /plugin marketplace add then /plugin install.

Don't install these

Be skeptical of third-party "AI medical scribe" or "AI diagnosis" skills floating around skill directories. Most aren't HIPAA-cleared, many run on infrastructure you can't audit, and a few request far more access than the job needs. Real clinical scribes are dedicated products with FDA pathway considerations, signed BAAs, and EHR integration, not a skill you sideload in five minutes. If a skill claims to diagnose or to safely handle PHI out of the box, that's a reason to walk away, not install.

A safe install order

1. pdf-tools and docx-word, 5 minutes each, from the Anthropic bundle 2. xlsx-spreadsheet, also in the bundle 3. deep-research, ~10 minutes, once whitelisted 4. scientific-skills, only if you do academic work

Before any install, open the skill's SKILL.md and GitHub repo, check who maintains it, and prefer Anthropic-published or well-starred sources. We walk through the full process in our guide on auditing a skill before installing it.

One workflow, several skills

The skills get more useful when they hand off to each other. A realistic morning: a colleague forwards three PDFs of recent trials before a tumor board. You use pdf-tools to merge and extract the dosing tables, deep-research to cross-check those trials against the current guideline and surface anything newer, and docx-word to draft a one-page summary for the team, with the citations carried through so anyone can verify them. What used to be an hour of reading and reformatting becomes fifteen minutes of prompting and a careful human review.

The review step is not optional. Every output here is a draft to check, not a fact to trust. The skills save you the mechanical work; the clinical judgment, and the responsibility, stay entirely with you.

FAQ

Can a doctor put patient data into a Claude skill? Only under a signed Business Associate Agreement in a HIPAA-compliant deployment. Consumer Claude and self-installed skills are not HIPAA-cleared, so keep protected health information out and de-identify test data.

Do these skills replace clinical decision support? No. None is a regulated clinical decision support tool. They handle literature, paperwork, and data around your clinical work; the decisions stay with you.

How do I install a Claude skill? Add the marketplace with /plugin marketplace add and the GitHub repo, then /plugin install and enable it. Anthropic's bundle installs via /plugin install anthropic-skills.

Are AI agent skills safe to install? Skills run code, so treat each like any software install: read the SKILL.md, check the GitHub source, and prefer trusted maintainers.

What's the difference between an AI tool and an AI skill for doctors? A tool is a standalone product like a scribe app. A skill extends an assistant such as Claude to do one job, like a cited literature summary. Skills are cheaper and more flexible but need setup.

Share this article

📬

Get More AI Tool Guides

New comparisons and guides every week. Join thousands of professionals staying ahead of the AI curve.