Clinical Skills Documentation
Everything you need to use and create clinical AI skills. Practical guidance for doctors, paramedics, and clinic staff.
Quick Start
Clinical Skills – Quick Start
Get up and running with TherapyPod's clinical skills in under 5 minutes.
What Are Clinical Skills?
Clinical skills are AI capability modules that help TherapyPod's assistant perform specialized healthcare tasks. Think of them as "specialist training" for the AI—each skill teaches it to handle specific clinical workflows like treatment planning, triage, or medication review.
You don't need to be technical to use skills. The AI automatically activates the right skill when you ask it to perform a related task.
For Doctors: 3 Ways to Use Skills Today
1. Treatment Plan Generation
Ask the AI to create a treatment plan, and it will use the treatment-plans skill automatically.
Try saying:
"Create a treatment plan for a 58-year-old diabetic patient with uncontrolled HbA1c of 9.2%"
The AI will generate a structured plan with:
- SMART goals (Specific, Measurable, Achievable, Relevant, Time-bound)
- Evidence-based interventions
- Monitoring parameters
- Follow-up schedule
2. Mental Health Screening
Request mental health assessments using validated instruments.
Try saying:
"Conduct a PHQ-9 depression screening for this patient" "What GAD-7 questions should I ask for anxiety assessment?"
3. Clinical Documentation
Generate structured notes quickly.
Try saying:
"Create a SOAP note for today's diabetes follow-up visit" "Draft a discharge summary for this patient"
For Paramedics & Staff: Daily Use Cases
Patient Triage
When patients contact the clinic, the AI uses the patient-triage skill to classify urgency.
What it does:
- Identifies emergency symptoms (chest pain, breathing difficulty, stroke signs)
- Classifies as Emergency / Urgent / Routine
- Routes to appropriate care pathway
- Works in English and Hindi
Your role: Review triage recommendations and escalate when the AI flags urgent/emergency cases.
Patient Education Materials
Generate easy-to-understand materials for patients.
Try saying:
"Create diabetes education material in Hindi for a patient with low health literacy" "Generate medication instructions for this patient's new prescriptions"
Care Coordination
When coordinating with other providers:
Try saying:
"Create a referral letter to cardiology for this patient" "Generate a handoff summary for the evening shift"
Quick Reference: Available Skills
| Skill | What It Does | Example Request | |-------|--------------|-----------------| | Treatment Plans | Creates structured care plans | "Make a treatment plan for hypertension" | | Patient Triage | Classifies urgency of symptoms | "Is this patient's chest pain urgent?" | | Mental Health | PHQ-9, GAD-7, crisis protocols | "Screen for depression" | | Medication Review | Checks interactions, adherence | "Review this medication list for interactions" | | Clinical Notes | SOAP, H&P, discharge summaries | "Write a progress note" | | Patient Education | Literacy-adapted materials | "Explain diabetes in simple Hindi" | | Care Coordination | Handoffs, referrals | "Create a referral to nephrology" | | Chronic Disease | Diabetes, HTN, HF protocols | "What's the target HbA1c for this patient?" |
Safety First
All skills are designed with patient safety as the top priority:
- Emergency Detection – The AI will immediately flag life-threatening symptoms
- Human Review – Complex cases are escalated to clinical staff
- No Definitive Diagnoses – The AI assists but doesn't replace clinical judgment
- HIPAA Compliant – All interactions follow privacy requirements
When in doubt, the AI will ask for human review.
Getting Help
- Questions about a skill? Ask the AI: "How does the treatment-plans skill work?"
- Technical issues? Contact your clinic administrator
- Feedback? We use it to improve the skills continuously
For Doctors
Understanding Clinical Skills
Clinical skills extend TherapyPod's AI with specialized medical knowledge and workflows. Each skill is designed by healthcare professionals and validated against clinical benchmarks before deployment.
Key Principle: Skills assist your clinical decision-making—they never replace it. The AI provides structured support, but you remain the clinical authority.
How Skills Work in Your Workflow
During Patient Encounters
-
Pre-Visit Preparation
- Review AI-generated summaries of chronic disease status
- Check medication adherence trends
- See pending screenings or follow-ups
-
During the Visit
- Request real-time clinical notes drafting
- Ask for evidence-based treatment recommendations
- Generate patient education materials on the spot
-
Post-Visit Documentation
- Complete SOAP notes with AI assistance
- Create referral letters
- Update treatment plans
Treatment Planning Best Practices
When using the treatment-plans skill:
Be Specific About Context
Good: "Create a treatment plan for a 62-year-old male with Type 2 diabetes (HbA1c 8.5%), hypertension, and early CKD (eGFR 55)" Less helpful: "Make a diabetes plan"
Review and Customize SMART Goals
The AI generates default targets based on guidelines. Adjust for individual patients:
- Elderly or frail patients may need relaxed targets
- Patients with limited life expectancy—focus on comfort
- Highly motivated patients—consider more aggressive goals
Verify Medication Recommendations
Always verify:
- Contraindications for this specific patient
- Current medication interactions
- Patient's ability to afford/access medications
- Renal/hepatic dosing adjustments
Specialty-Specific Tips
Cardiology
- Use
chronic-disease-managementfor heart failure protocols - Request GDMT (Guideline-Directed Medical Therapy) checklists
- Ask for patient-friendly heart failure zone education
Mental Health
mental-health-assessmentincludes safety screening- Crisis protocols are built-in—AI will escalate appropriately
- Session notes follow standard formats
Primary Care
- Combine multiple skills for complex patients
- Use
medication-reviewbefore adding new medications - Request annual review checklists for chronic conditions
Clinical Validation
All skills are tested against our ClinEval benchmark:
| Dimension | What's Tested | |-----------|---------------| | Safety Detection | Does it catch emergencies? (30% weight) | | Triage Accuracy | Is classification correct? (25% weight) | | Escalation Quality | Does it know when to involve humans? (20% weight) | | Response Appropriateness | Is advice clinically sound? (15% weight) |
Skills only ship when they pass safety thresholds.
For Paramedics & Clinic Staff
Your Role in the Skills Workflow
You are the frontline for patient interactions. Skills help you:
- Triage incoming patient concerns accurately
- Prepare documentation for doctor review
- Coordinate care across providers
- Educate patients in their preferred language
Patient Triage Workflow
When a patient contacts the clinic:
Step 1: Gather Information Ask about:
- Main concern (in patient's own words)
- Duration of symptoms
- Severity (1-10 scale)
- Any alarming symptoms (chest pain, breathing problems, etc.)
Step 2: Use AI Triage Tell the AI what you learned. Example:
"Patient calling with headache for 3 days, rates pain 6/10, no fever, no vision changes"
Step 3: Review Classification The AI will respond with:
- Emergency → Immediate escalation to doctor/emergency services
- Urgent → Same-day appointment or callback within hours
- Routine → Standard scheduling
Step 4: Take Action
- For emergencies: Follow your emergency protocols immediately
- For urgent: Notify clinical staff, schedule priority appointment
- For routine: Schedule appointment, provide interim guidance
Red Flags to Always Escalate
Even if the AI classifies as routine, escalate immediately if patient mentions:
| Category | Red Flags | |----------|-----------| | Cardiac | Chest pain, arm/jaw pain, severe sweating | | Neuro | Sudden weakness, slurred speech, worst headache ever | | Respiratory | Severe breathing difficulty, blue lips/fingers | | Mental Health | Suicidal thoughts, intent to harm self/others | | Pediatric | High fever in infant, not responding normally |
When in doubt, escalate. False alarms are better than missed emergencies.
Patient Education Support
You can help patients understand their conditions:
For Limited Hindi/English Speakers
"Create simple instructions for taking metformin, in Hindi"
For Low Health Literacy
"Explain blood pressure in very simple terms, 6th grade reading level"
For Visual Learners
"Create a medication schedule chart for this patient"
Care Coordination Tasks
Referral Preparation
"Draft a referral letter to nephrology for Dr. [Name] to review"
Shift Handoffs
"Summarize today's urgent cases for the evening team"
Follow-up Tracking
"Which patients need HbA1c rechecks this month?"
Creating New Skills
Who Can Create Skills?
Skills can be proposed by anyone, but clinical validation requires:
- Medical review by qualified healthcare professional
- ClinEval benchmark testing
- Safety assessment
Skill Structure
Every skill needs a SKILL.md file with:
--- name: your-skill-name # lowercase, hyphens only description: | What the skill does and when to use it. Keep under 1024 characters. license: Proprietary - TherapyPod compatibility: Designed for TherapyPod clinical AI agents metadata: author: Your Name / Team version: "1.0" category: Category Name clineval_validated: false # Set true after validation --- # Your Skill Title Detailed instructions for the AI agent...
Step-by-Step: Creating a New Skill
Step 1: Identify the Need
Ask yourself:
- What clinical task am I doing repeatedly?
- Could structured AI assistance help?
- Is there a standard workflow or guideline to follow?
Good candidates:
- Protocol-based workflows (preoperative checklists)
- Standardized assessments (fall risk, nutritional screening)
- Guideline-driven care (anticoagulation management)
- Documentation templates (specific note types)
Step 2: Document the Workflow
Write out exactly how the task should be performed:
- What information is needed as input?
- What are the steps in order?
- What are the decision points?
- What is the expected output?
Example for a Falls Risk Skill:
INPUT NEEDED: - Patient age - Medication list (especially sedatives, antihypertensives) - Gait/balance observations - History of falls STEPS: 1. Calculate Morse Fall Scale score 2. Identify high-risk medications 3. Assess environmental risk factors 4. Generate risk level: Low/Moderate/High OUTPUT: - Risk score and level - Contributing factors - Recommended interventions - Patient education points
Step 3: Write the SKILL.md
Structure your skill file:
-
Frontmatter (required)
- name, description, metadata
-
Overview Section
- What the skill does
- When to use it
- Scope and limitations
-
Detailed Instructions
- Step-by-step workflow
- Decision criteria
- Examples of inputs and outputs
-
Clinical Guidelines Reference
- Link to evidence base
- Cite relevant guidelines
-
Safety Considerations
- When to escalate
- What NOT to do
- Limitations
Step 4: Test Informally
Before formal validation:
- Try the skill with sample cases
- Does it produce expected outputs?
- Are edge cases handled?
- Is the language clear?
Step 5: Submit for Validation
Contact the TherapyPod Labs team to:
- Review clinical accuracy
- Run ClinEval benchmark tests
- Assess safety considerations
- Approve for production use
Skill Writing Tips
Be Explicit About Scope
✓ "This skill handles Type 2 diabetes medication adjustment in adults" ✗ "This skill handles diabetes"
Include Examples
✓ "Example: For a patient with HbA1c 8.5% on metformin alone, recommend adding SGLT2 inhibitor per ADA guidelines" ✗ "Add medications as appropriate"
Define Escalation Triggers
✓ "Escalate to physician if: - Glucose > 400 mg/dL - Signs of DKA - Hypoglycemia with altered consciousness" ✗ "Escalate when needed"
Reference Guidelines
✓ "Follow ADA 2024 Standards of Care, Section 9: Pharmacologic Approaches to Glycemic Treatment" ✗ "Follow standard guidelines"
Template: Minimal Skill
--- name: my-new-skill description: Brief description of what this skill does and when the AI should use it. metadata: author: Your Name version: "1.0" category: Clinical Documentation --- # My New Skill
Data & Privacy
Patient Data Handling
- Skills process information but don't store patient data
- All interactions are HIPAA-compliant
- De-identification is applied to any persisted content
- No real patient identifiers in skill training or testing
Audit Trail
All skill activations are logged:
- Which skill was used
- Who requested it (staff ID, not patient)
- Timestamp
- Whether escalation occurred
This supports quality improvement and compliance.
Feedback & Improvement
Reporting Issues
If a skill produces incorrect or unsafe output:
- Do not follow the recommendation
- Document what happened
- Report to your clinic administrator
- We'll investigate and improve the skill
Suggesting Improvements
We welcome feedback:
- "This skill should also consider..."
- "The output format would be better if..."
- "I wish there was a skill for..."
Your clinical expertise makes our skills better.
