Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generate clinical trial protocols for medical devices or drugs. This skill should be used when users say "Create a clinical trial protocol", "Generate protocol for [device/drug]", "Help me design a clinical study", "Research similar trials for [intervention]", or when developing FDA submission documentation for investigational products.
.claude/skills/clinical-trial-protocol-skill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | — | — |
| case-11 | ✗→✓ | ▲ Improved | — | — |
| case-08 | ✗→✓ | ▲ Improved | — | — |
| case-15 | ✗→✓ | ▲ Improved | — | — |
| case-18 | ✗→✓ | ▲ Improved | — | — |
CRITICAL: This orchestrator follows a SIMPLE START approach:
Why this matters:
This skill generates clinical trial protocols for medical devices or drugs using a modular, waypoint-based architecture
Starting with an intervention idea (device or drug), this orchestrated workflow offers two modes:
🔬 Research Only Mode (Steps 0-1):
📄 Full Protocol Mode (Steps 0-5):
All analysis data is stored in waypoints/ directory as JSON/markdown files:
waypoints/
├── intervention_metadata.json # Intervention info, status, initial context
├── 01_clinical_research_summary.json # Similar trials, FDA guidance, recommendations
├── 02_protocol_foundation.md # Protocol sections 1-6 (Step 2)
├── 03_protocol_intervention.md # Protocol sections 7-8 (Step 3)
├── 04_protocol_operations.md # Protocol sections 9-12 (Step 4)
├── 02_protocol_draft.md # Complete protocol (concatenated in Step 4)
├── 02_protocol_metadata.json # Protocol metadata
└── 02_sample_size_calculation.json # Statistical sample size calculationRich Initial Context Support: Users can provide substantial documentation, technical specifications, or research data when initializing the intervention (Step 0). This is preserved in intervention_metadata.json under the initial_context field. Later steps reference this context for more informed protocol development.
Each step is an independent skill in references/ directory:
references/
├── 00-initialize-intervention.md # Collect device or drug information
├── 01-research-protocols.md # Clinical trials research and FDA guidance
├── 02-protocol-foundation.md # Protocol sections 1-6 (foundation, design, population)
├── 03-protocol-intervention.md # Protocol sections 7-8 (intervention details)
├── 04-protocol-operations.md # Protocol sections 9-12 (assessments, statistics, operations)
└── 05-generate-document.md # NIH Protocol generationscripts/
└── sample_size_calculator.py # Statistical power analysis (validated)Installation:
.mcpb file into Claude DesktopAvailable Tools: search_clinical_trials - Search by:
condition - Disease or condition (e.g., "pancreatic cancer") intervention - Drug, device, or treatment (e.g., "pembrolizumab", "CAR-T") sponsor - Sponsor or collaborator name (e.g., "Pfizer", "NIH") location - City, state, or country (e.g., "California", "Boston") status - "recruiting" (default), "active", "completed", "all" phase - Trial phase: "1", "2", "3", "4", "early_phase1" max_results - Default 25, max 100
get_trial_details - Get comprehensive details for a specific trial using its nct_id (e.g., "NCT04267848"). Returns eligibility criteria, outcomes, study design, and contact information.
Verification: Step 1 will automatically test MCP connectivity at startup.
Purpose: FDA regulatory pathway research via explicit database URLs
Sources:
Template Files: Any .md files in the assets/ directory
Purpose: Reference template for protocol structure and content guidance. The system automatically detects available templates and uses them dynamically.
Installation:
bashpip install -r requirements.txt
Dependencies:
Purpose: Accurate statistical sample size calculations for clinical protocols
Simply invoke the skill and select your desired mode:
🔬 Research Only Mode:
📄 Full Protocol Mode:
Resume Capability: If interrupted, simply restart the skill and it will automatically resume from your last completed step.
When skill is invoked, display the following message:
🧬 CLINICAL TRIAL PROTOCOL
Welcome! This skill generates clinical trial protocols for medical devices or drugs.
[If waypoints/intervention_metadata.json exists:]
✓ Found existing protocol in progress: [Intervention Name]
Type: [Device/Drug]
Completed: [List of completed steps]
Next: [Next step to execute]
📋 SELECT MODE:
1. 🔬 Research Only - Run clinical research analysis (Steps 0-1)
• Collect intervention information
• Research similar clinical trials
• Find FDA guidance and regulatory pathways
• Generate comprehensive research summary as .md artifact
2. 📄 Full Protocol - Generate complete clinical trial protocol (Steps 0-5)
• Everything in Research Only, plus:
• Generate all protocol sections
• Create professional protocol document
3. ❌ Exit
Please select an option (1, 2, or 3):🛑 STOP and WAIT for user selection (1, 2, or 3)
execution_mode = "research_only" and proceed to Research Only Workflow Logicexecution_mode = "full_protocol" and proceed to Full Workflow LogicThis workflow executes only Steps 0 and 1, then generates a formatted research summary artifact.
Step 1: Check for Existing Waypoints
waypoints/intervention_metadata.json exists: Load metadata, check if steps 0 and 1 are already completeStep 2: Execute Research Steps (0 and 1)
For each step (0, 1):
references/00-initialize-intervention.md (collect intervention info)references/01-research-protocols.md (clinical research and FDA guidance)Step 3: Generate Research Summary Artifact
After Step 1 completes successfully:
waypoints/intervention_metadata.json (intervention details)waypoints/01_clinical_research_summary.json (research findings)markdown# Clinical Research Summary: [Intervention Name] ## Intervention Overview - **Type:** [Device/Drug] - **Indication:** [Target condition/disease] - **Description:** [Brief intervention description] - **Mechanism of Action:** [How it works] ## Similar Clinical Trials [List top 5-10 similar trials with NCT ID, title, phase, status, key findings] ## FDA Regulatory Pathway - **Recommended Pathway:** [510(k), PMA, De Novo, IND, NDA, BLA, etc.] - **Regulatory Basis:** [Rationale for pathway selection] - **Key Requirements:** [Major regulatory considerations] ## FDA Guidance Documents [List relevant FDA guidance documents with links and key excerpts] ## Study Design Recommendations - **Suggested Study Type:** [RCT, single-arm, etc.] - **Phase Recommendation:** [Phase 1, 2, 3, etc.] - **Primary Endpoint Suggestions:** [Based on similar trials] - **Sample Size Considerations:** [Preliminary thoughts] ## Key Insights and Recommendations [Synthesized recommendations for protocol development] ## Next Steps [If user wants to proceed with full protocol development] --- *Generated by Clinical Trial Protocol Skill* *Date: [Current date]*
waypoints/research_summary.md✅ RESEARCH COMPLETE
Research Summary Generated: waypoints/research_summary.md
📊 Key Findings:
• Similar Trials Found: [X trials]
• Recommended Pathway: [Pathway name]
• FDA Guidance Documents: [X documents identified]
• Study Design: [Recommended design]
📄 The research summary has been saved as a formatted markdown artifact.
Would you like to:
1. Continue with full protocol generation (steps 2-5)
2. Exit and review research summary
Option 1 Logic (Continue to Full Protocol):
execution_mode = "full_protocol"Option 2 Logic (Exit):
Step 1: Check for Existing Waypoints
waypoints/intervention_metadata.json exists: Load metadata, check completed_steps array, resume from next incomplete stepStep 2: Execute Steps in Order
For each step (0, 1, 2, 3, 4, 5):
references/00-initialize-intervention.md (read when Step 0 executes)references/01-research-protocols.md (read when Step 1 executes)references/02-protocol-foundation.md (read when Step 2 executes - sections 1-6)references/03-protocol-intervention.md (read when Step 3 executes - sections 7-8)references/04-protocol-operations.md (read when Step 4 executes - sections 9-12)references/05-concatenate-protocol.md (read when Step 5 executes - final concatenation)Step 2.5: Protocol Completion Menu
After Step 4 completes successfully, display the EXACT menu below (do not improvise or create alternative options):
✅ PROTOCOL COMPLETE: Protocol Draft Generated
Protocol Details:
• Study Design: [Design from metadata]
• Sample Size: [N subjects from metadata]
• Primary Endpoint: [Endpoint from metadata]
• Study Duration: [Duration from metadata]
Protocol file: waypoints/02_protocol_draft.md
File size: [Size in KB]
📋 WHAT WOULD YOU LIKE TO DO NEXT?
1. 📄 Review Protocol in Artifact - click on the .md file above
2. 📄 Concatenate Final Protocol (Step 5)
3. ⏸️ Exit and Review Later
Option 1 Logic (Review in Artifact): Pause, let user open the section files, wait for further instruction
Option 2 Logic (Concatenate Protocol):
references/05-concatenate-protocol.mdOption 3 Logic (Exit):
Step 3: Final Summary
Display completion message with:
JSON Waypoints (Steps 0, 1):
Markdown Waypoints (Steps 2, 3, 4):
02_protocol_foundation.md (Sections 1-6)03_protocol_intervention.md (Sections 7-8)04_protocol_operations.md (Sections 9-12)02_protocol_draft.md (concatenated complete protocol)Each step implements aggressive summarization:
Each subskill is designed to:
⚠️ IMPORTANT: This protocol generation tool provides preliminary clinical study protocol based on NIH/FDA guidelines and similar trials. It does NOT constitute:
REQUIRED before proceeding with clinical study:
PROFESSIONAL CONSULTATION STRONGLY RECOMMENDED
Clinical trial protocols are complex, high-stakes documents requiring expertise across multiple disciplines. Professional consultation with clinical trial experts, biostatisticians, and regulatory affairs specialists is essential before proceeding with clinical study planning.
When this skill is invoked:
references/01-research-protocols.md and follow its instructionswaypoints/research_summary.mdwaypoints/intervention_metadata.json after each step| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-23 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +70 percentage points is the difference between those two pass rates over the 22 comparable cases.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.