Clinical Agent Skills  ·  v2.4

Clinical Decision
Protocol Skills

A standardised, version-controlled format for packaging specialty clinical knowledge, decision protocols, and institutional guidelines into portable AI skills — deployable across specialties, hospitals, and clinicians in DoctorAssist.AI.

Open Standard HIPAA-Aligned Guideline-Linked

What are Clinical Agent Skills?

Clinical Agent Skills are structured, portable packages of medical expertise. Each skill encodes a specialty's decision logic — guidelines, risk scores, dosing rules, alert thresholds, and institutional protocols — into a folder that DoctorAssist.AI agents load on demand at the point of care.

Modelled on the open Agent Skills standard, clinical skills carry additional medical metadata: evidence grades, guideline sources (AHA, ESC, WHO, NCCN), specialty context, and patient safety constraints built into the architecture rather than applied as an afterthought.

Progressive Disclosure

Agents load only a skill's name and description at startup. The full clinical protocol loads into context only when a matching patient scenario is detected — keeping the AI's reasoning efficient and precise without overwhelming the context window.

The SKILL.md Format

Every clinical skill is a folder with a SKILL.md file as its entrypoint, using frontmatter for metadata and Markdown for clinical instructions. Additional directories bundle calculators, drug protocols, and reference guidelines.

oncology-skill/ ├── SKILL.md # Required: metadata + protocol ├── guidelines/ # NCCN, ESMO, WHO PDFs / links ├── calculators/ # Risk scores — ECOG, TNM, etc. ├── drug-protocols/ # Regimen templates, dose tables ├── templates/ # SOAP notes, MDT summaries └── institution.json # Hospital-specific overrides
oncology-skill / SKILL.md
Specialty
Medical Oncology
Guideline Source
NCCN v2.2025 · ESMO
Evidence Grade
Level 1A
--- name: oncology-clinical-decision version: "2.4.1" specialty: "Medical Oncology" subSpecialties: ["Breast", "GI", "Lung", "Haematology"] guidelineSources: ["NCCN v2.2025", "ESMO 2024", "WHO AMR"] evidenceGrade: "1A" safetyClass: "HIGH — physician approval required for all actions" institution: "Tata Memorial Centre, Mumbai" description: "Activate for oncology patients — staging, regimen selection, toxicity management, MDT summaries, and palliative transitions." --- ## Activation Criteria # Activates when patient has active malignancy, chemotherapy # orders, staging workup, or oncology consult request. ## Clinical Decision Workflow 1. Confirm pathological diagnosis and molecular markers (ER/PR/HER2, EGFR, PD-L1) 2. Apply TNM staging using bundled calculator 3. Cross-reference NCCN guidelines for recommended regimen 4. Check renal/hepatic dose adjustments via drug-protocols/ 5. Flag drug-drug interactions with current medication list 6. Generate MDT summary template with evidence links 7. Set monitoring schedule and toxicity alert thresholds ## Safety Constraints - NEVER initiate chemotherapy without consultant co-sign - ALWAYS display ECOG performance status before regimen - Flag fertility preservation for patients under 40 years
cardiology-skill / SKILL.md
Specialty
Interventional Cardiology
Guideline Source
AHA / ACC 2024 · ESC
Evidence Grade
Level 1A
--- name: cardiology-acs-protocol version: "3.1.0" specialty: "Cardiology" subSpecialties: ["ACS", "Heart Failure", "Arrhythmia", "Structural"] guidelineSources: ["AHA/ACC 2024", "ESC 2023", "STEMI Guidelines"] evidenceGrade: "1A" safetyClass: "CRITICAL — real-time troponin and ECG monitoring" description: "Activate on chest pain presentations, troponin elevation, ECG changes, or known cardiac history. Runs ACS risk stratification, activates cath lab protocol, manages antiplatelet therapy." --- ## ACS Risk Stratification 1. Compute HEART Score (History · ECG · Age · Risk Factors · Troponin) 2. Delta troponin at 0h / 3h / 6h — flag above 0.04 3. STEMI → immediate cath lab activation 4. NSTEMI/UA → risk-stratified timeline 5. Load dual antiplatelet therapy per renal function 6. Hold metformin if eGFR below 45 and contrast planned ## Alert Thresholds - Troponin delta > 0.04 → CRITICAL → auto-page cardiology - MAP below 65 mmHg → haemodynamic instability protocol - New LBBB on ECG → treat as STEMI equivalent

How Skills Work in Practice

DoctorAssist.AI agents use a three-stage progressive disclosure model to activate clinical skills at the right moment — without loading unnecessary context or adding friction to the clinical workflow.

01
Discovery — Lightweight Scan

At session start, the agent loads only each skill's name and description fields. This gives it a complete map of available clinical expertise without loading any protocol detail into active memory.

02
Activation — Pattern Match

As the agent reads incoming data — labs, vitals, notes, orders — it matches clinical signals to skill descriptions. A rising troponin activates the Cardiology ACS skill. A new lung mass activates the Oncology skill. Activation is automatic, no prompting required.

03
Execution — Protocol in Context

The full SKILL.md loads into the agent's reasoning context. It follows the decision workflow, cross-references bundled guidelines, runs calculators, and surfaces evidence-linked recommendations before you ask.

Available Specialty Skills

DoctorAssist.AI ships with a core library of specialty skills, each maintained against the latest international guidelines. Institutions can extend, fork, or layer these with local protocols.

Cardiology

ACS protocols, HEART score, heart failure staging, arrhythmia management, and antiplatelet therapy — aligned to AHA/ACC 2024 and ESC 2023.

AHA · ACC · ESC 2024
Oncology

Staging, molecular profiling, regimen selection, toxicity grading, MDT summaries, and palliative transitions across solid tumours and haematological malignancies.

NCCN · ESMO · WHO
Neurology

Acute stroke pathways (NIHSS, tPA eligibility, thrombectomy window), epilepsy management, Parkinson's titration, and neuro ICU monitoring protocols.

AHA Stroke · EAN
Critical Care / ICU

Sepsis bundles (Surviving Sepsis 2024), ventilator management, vasopressor titration, ARDS protocols, and haemodynamic monitoring decision trees.

SCCM · Surviving Sepsis 2024
Endocrinology

Diabetes management (ADA 2025), insulin titration, DKA protocol, thyroid workup, adrenal crisis management, and HbA1c stratification by comorbidity.

ADA 2025 · ETA
Nephrology

AKI staging (KDIGO), CKD dose adjustment, contrast nephropathy prevention, renal replacement therapy thresholds, and electrolyte correction protocols.

KDIGO · ERA-EDTA

Adding a Specialty Skill

Any licensed clinician or hospital administrator with author access can add a specialty skill to their DoctorAssist.AI instance — sourced from the Skill Registry, forked from another institution, or authored from scratch.

Step-by-step: Deploy Oncology to your instance

1
Open the Skill Registry

In your DoctorAssist.AI dashboard, go to Settings → Clinical Skills → Browse Registry. Search by specialty or filter by guideline source, evidence grade, or institution adoption count.

2
Review the Skill

Preview the full SKILL.md, evidence grades, guideline versions, and last-updated date. Each skill shows how many institutions have validated it in production use.

3
Configure Institution Overrides

Optionally add an institution.json overlay for local formulary restrictions, department-specific thresholds, or preferred drug brands — without modifying the base skill or breaking future updates.

4
Assign to Clinicians or Departments

Deploy hospital-wide, to a single department, or to individual physician accounts. Skills are immediately active for new patient encounters on assigned accounts with no restart required.

Instant Activation

Once deployed, the skill begins pattern-matching on all incoming patient data for assigned clinicians — surfacing staging alerts, regimen suggestions, and MDT prompts automatically before the next encounter opens.

Copying Skills: Hospital to Hospital

Skills are portable, version-controlled folders. A validated protocol from one institution can be adopted — with or without customisation — by any other hospital on DoctorAssist.AI in minutes, not months of committee review.

Tata Memorial Centre
oncology-skill v2.4.1
Validated · 847 patients
Transfer
AIIMS New Delhi
Fork + local overrides
institution.json applied
Method Use Case Customisation Time to Deploy
Registry Install Adopt a published skill as-is Institution overlay only < 5 minutes
Fork & Customise Adapt a peer institution's protocol Full edit access 1–2 hours
Export / Import Air-gapped or on-prem transfer Full edit access < 30 minutes
API Sync Automated multi-site updates Version-locked or floating Automated

Copying Skills: Doctor to Doctor

A senior consultant's curated skill set — refined over years of clinical practice — can be shared directly with a registrar, fellow, or colleague joining the team. This transfers institutional knowledge that no textbook contains.

Controlled Sharing

Skill sharing between clinicians requires administrator approval and full audit logging. All transferred skills inherit the receiving clinician's institution's safety constraints and approval workflows. No action can be taken without the receiving clinician's explicit consent.

Consultant shares a skill package

From their profile, a senior clinician selects one or more skills and shares via a secure link or direct push to a colleague's account. The package includes the SKILL.md, calculators, and any personalised decision thresholds.

Recipient reviews and accepts

The receiving clinician previews the full protocol before accepting. They can merge it with their existing skills or apply it selectively — for example, only for paediatric oncology patients within their panel.

Audit trail preserved

Every transfer is logged with timestamp, sender, recipient, version, and any post-transfer modifications. This creates a traceable lineage of protocol adoption for compliance, accreditation, and quality governance.

Advantages in Clinical Practice

Standardising clinical decision logic through portable skills has measurable benefits across patient safety, operational efficiency, and institutional knowledge — grounded in how medicine is actually practised.

Patient Safety

Eliminate protocol drift between shifts

When a night registrar takes over, the same evidence-based protocol governs every decision — not memory or habit. A STEMI patient at 3am receives the same cath lab activation logic as at 3pm with the attending consultant present.

Dose safety across renal and hepatic impairment

Drug protocols embedded in skills automatically adjust recommendations based on the patient's current eGFR, liver enzymes, and weight — catching prescribing errors before orders are placed, not after.

Institutional Knowledge Transfer

Preserve senior expertise when consultants leave

A retiring oncologist's curated treatment approach — built from decades of MDT discussions — persists in a skill file rather than departing with them. The institution retains that clinical intelligence in a queryable, auditable form.

Accelerate junior clinician training

Registrars and fellows operate with a consultant-level clinical framework from day one. Skills scaffold decision-making with evidence links and confidence scores, reducing the learning curve without removing clinical autonomy.

Operational Efficiency

Reduce time-to-treatment in time-critical conditions

For ACS, stroke, and sepsis — where minutes alter outcomes — automated skill activation means the right protocol is in front of the clinician before they finish reviewing the presenting complaint. No search required.

One guideline update propagates everywhere

When NCCN releases new oncology guidelines, updating the base skill in the registry propagates the change to every hospital and physician using it — with full changelog visibility and rollback support.

Research and Quality Improvement

Cross-institutional outcome comparison

When multiple hospitals run the same skill version against similar patient cohorts, DoctorAssist.AI can generate de-identified comparative outcome data — turning everyday clinical practice into real-world evidence.

Guideline adherence becomes measurable

Because every recommendation links to a specific skill version and guideline clause, adherence can be tracked per-physician, per-department, and per-institution — making audit a continuous feedback loop rather than a periodic event.

The Clinical Skill Registry

DoctorAssist.AI maintains a curated registry of peer-reviewed, guideline-anchored clinical skills. Each entry is versioned, attributed, and carries a transparent evidence grade.

Skill Version Guidelines Institutions Updated
cardiology-acs-protocol 3.1.0 AHA/ACC 2024 · ESC 42 Jan 2025
oncology-clinical-decision 2.4.1 NCCN v2.2025 · ESMO 31 Feb 2025
sepsis-surviving-sepsis 4.0.0 Surviving Sepsis 2024 58 Dec 2024
stroke-acute-pathway 2.2.0 AHA Stroke 2024 · EAN 27 Nov 2024
diabetes-ada-management 5.0.0 ADA Standards 2025 64 Mar 2025
aki-kdigo-protocol 1.8.3 KDIGO 2024 · ERA-EDTA 19 Oct 2024

Deploy your first skill

Start with a core specialty from the registry, or bring your own institutional protocols into DoctorAssist.AI in under an hour.

Browse Skill Registry