clinical-data-review— agent skill

You are in AUTONOMOUS MODE. Do NOT ask questions. Review the entire codebase's clinical data layer systematically.

INPUT

$ARGUMENTS (optional). If no arguments provided, review all data models, schemas, and APIs in the current working directory for clinical data standards compliance. If a specific standard is named (e.g., "FHIR only", "terminology"), focus on that area.

PHASE 0: CLINICAL SYSTEM DETECTION

Auto-detect the project stack and clinical context:

1. Detect tech stack (package.json, requirements.txt, pom.xml, go.mod, *.csproj, etc.).
2. Identify clinical libraries and dependencies:

• FHIR: hapi-fhir, fhir.js, fhirclient, pyFHIR, Firely SDK, fhir-net-api
• HL7v2: node-hl7-complete, python-hl7, HAPI, nHAPI
• DICOM: dcmjs, pydicom, fo-dicom, cornerstone.js
• Terminology: SNOMED packages, LOINC libraries, ICD-10 validators
• CDA: CDA generators/parsers, CCDA templates

1. Identify database and ORM (Prisma, TypeORM, SQLAlchemy, Hibernate, EF Core).
2. Locate data model definitions:

• Schema files (.prisma, .graphql, *.proto)
• Model/entity classes
• Migration files
• OpenAPI/Swagger specs
• TypeScript/Python types/interfaces

PHASE 1: FHIR CONFORMANCE REVIEW

Evaluate data models against FHIR R4 resource definitions.

1.1 Resource Modeling

Map project data models to FHIR resources:

• Patient: demographics, identifiers, contact, communication preferences
• Practitioner / PractitionerRole: provider info, specialties, qualifications
• Organization: facilities, departments, healthcare organizations
• Encounter: visits, admissions, appointments
• Condition: diagnoses, problems, health concerns
• Observation: vitals, lab results, social history, assessments
• MedicationRequest / MedicationStatement: prescriptions, current meds
• AllergyIntolerance: allergies, adverse reactions
• Procedure: surgical, diagnostic, therapeutic procedures
• DiagnosticReport: lab reports, imaging reports, pathology
• DocumentReference: clinical documents, notes, external records
• CarePlan: treatment plans, goals, activities
• Immunization: vaccination records

For each mapped resource, check:

• Required FHIR elements present (status, subject, code, etc.).
• Correct cardinality (0..1, 0.., 1..1, 1..).
• Proper data types (CodeableConcept vs string, Reference vs ID, Period vs DateTime).
• Resource references use proper Reference type with resource type + ID.
• Extensions are properly defined (not ad-hoc fields breaking FHIR structure).

1.2 Search Parameters

• Verify FHIR search parameter support on API endpoints.
• Check standard search params: _id, _lastUpdated, _tag, _profile.
• Check resource-specific params: Patient?name, Observation?code, etc.
• Verify search modifiers: :exact, :contains, :missing.
• Check chained search support: Observation?subject:Patient.name.
• Verify _include and _revinclude support.

1.3 Capability Statement

• Check for /metadata endpoint returning CapabilityStatement resource.
• Verify it accurately reflects implemented resources and operations.
• Check for declared profiles and supported search parameters.

1.4 Bundle Support

• Verify transaction Bundle support (POST to root with type: transaction).
• Check for batch Bundle support.
• Verify searchset Bundle response format for search endpoints.
• Check Bundle entry fullUrl and resource consistency.

PHASE 2: TERMINOLOGY AND CODING STANDARDS

2.1 ICD-10 (Diagnoses)

• Search for ICD-10-CM code handling in data models and business logic.
• Verify code format validation (letter + 2 digits + optional decimal + up to 4 digits).
• Check for code versioning (ICD-10 updates annually in October).
• Verify code descriptions are stored or lookable.
• Check for ICD-10-PCS (procedure codes) if surgical/procedural data exists.
• Flag hardcoded ICD-10 codes without version tracking.

2.2 CPT / HCPCS (Procedures and Services)

• Search for CPT code handling in billing, orders, or procedure models.
• Verify CPT code validation (5 digits or 4 digits + letter).
• Check for HCPCS Level II codes (letter + 4 digits) for supplies/equipment.
• Verify modifier support (CPT modifiers: -25, -59, -76, etc.).
• Check for annual code updates handling.

2.3 SNOMED CT (Clinical Terms)

• Search for SNOMED concept IDs in data models.
• Verify SNOMED codes use proper SCTID format (6-18 digit numeric).
• Check for concept hierarchy navigation capability.
• Verify SNOMED-to-ICD-10 mapping if cross-coding is needed.
• Check for SNOMED version/edition tracking.

2.4 LOINC (Lab and Observations)

• Search for LOINC codes in observation/lab models.
• Verify LOINC code format validation (numeric with optional dash and check digit).
• Check for proper units of measure (UCUM standard) paired with LOINC codes.
• Verify lab result value sets align with LOINC answer lists.

2.5 RxNorm (Medications)

• Search for medication coding in prescription/medication models.
• Check for RxNorm concept unique identifiers (RxCUI).
• Verify NDC (National Drug Code) handling if pharmacy integration exists.
• Check for drug-drug interaction checking capability.

2.6 Terminology Service

• Check for terminology server integration ($lookup, $validate-code, $expand).
• Verify ValueSet binding on coded fields.
• Check for CodeSystem resources or external terminology service configuration.
• Flag coded fields that accept free text without code validation.

PHASE 3: INTEROPERABILITY ASSESSMENT

3.1 CDA / CCDA Documents

• Search for CDA document generation or parsing.
• Check for CCDA template conformance (CCD, Discharge Summary, Progress Note, etc.).
• Verify required sections: allergies, medications, problems, procedures, results.
• Check for structured vs narrative-only sections.
• Verify XML schema validation on generated documents.

3.2 Bulk Data Export

• Check for FHIR Bulk Data Access ($export) implementation.
• Verify NDJSON output format.
• Check for group-level and system-level export support.
• Verify async export pattern (kick-off, status polling, file download).

3.3 ADT Messaging

• Search for HL7v2 ADT (Admit/Discharge/Transfer) message handling.
• Check for A01 (admit), A02 (transfer), A03 (discharge), A04 (register), A08 (update) message type support.
• Verify PID, PV1, NK1 segment parsing/generation.

3.4 ORU / ORM Messaging

• Search for HL7v2 ORU (results) and ORM (orders) message handling.
• Check for OBR, OBX segment handling in results.
• Verify order/result linking via placer/filler order numbers.

3.5 Direct Messaging

• Check for Direct protocol support (secure email for clinical data).
• Verify S/MIME encryption compliance.

PHASE 4: CLINICAL WORKFLOW VALIDATION

4.1 Order Management

• Check order lifecycle: draft -> active -> completed/cancelled.
• Verify order validation (appropriate order for patient context).
• Check for duplicate order detection.
• Verify order modification and cancellation workflows.
• Check for clinical decision support at order entry.

4.2 Results Management

• Verify result status workflow: preliminary -> final -> corrected -> amended.
• Check for abnormal result flagging (reference ranges, critical values).
• Verify result acknowledgment tracking.
• Check for result routing based on ordering provider.

4.3 Medication Management

• Check prescription lifecycle: draft -> active -> stopped/completed.
• Verify drug allergy checking against patient allergies.
• Check for formulary validation.
• Verify medication reconciliation support.
• Check for e-prescribing (NCPDP SCRIPT) readiness.

4.4 Documentation

• Check for clinical note types (progress notes, H&P, discharge summary).
• Verify note signing/cosigning workflow.
• Check for addendum support (append, not edit).
• Verify template support for structured documentation.

4.5 Data Quality

• Check for required field enforcement on critical clinical data.
• Verify referential integrity between related clinical records.
• Check for data validation rules (date ranges, numeric ranges, code validation).
• Verify duplicate detection mechanisms (patient matching, record deduplication).

============================================================ SELF-HEALING VALIDATION (max 2 iterations) ============================================================

After producing output, validate data quality and completeness:

1. Verify all output sections have substantive content (not just headers).
2. Verify every finding references a specific file, code location, or data point.
3. Verify recommendations are actionable and evidence-based.
4. If the analysis consumed insufficient data (empty directories, missing configs),

note data gaps and attempt alternative discovery methods.

IF VALIDATION FAILS:

• Identify which sections are incomplete or lack evidence
• Re-analyze the deficient areas with expanded search patterns
• Repeat up to 2 iterations

IF STILL INCOMPLETE after 2 iterations:

• Flag specific gaps in the output
• Note what data would be needed to complete the analysis

OUTPUT FORMAT

## Clinical Data Model Review

**Project:** [name]
**Stack:** [detected technologies]
**Clinical Domain:** [EHR/lab/pharmacy/imaging/billing/etc.]
**Date:** [date]

### Standards Conformance Summary

| Standard | Coverage | Conformance | Issues |
|---|---|---|---|
| FHIR R4 | [N resources mapped] | [FULL/PARTIAL/NONE] | N |
| ICD-10-CM/PCS | [usage found?] | [VALID/PARTIAL/NONE] | N |
| CPT/HCPCS | [usage found?] | [VALID/PARTIAL/NONE] | N |
| SNOMED CT | [usage found?] | [VALID/PARTIAL/NONE] | N |
| LOINC | [usage found?] | [VALID/PARTIAL/NONE] | N |
| RxNorm | [usage found?] | [VALID/PARTIAL/NONE] | N |
| CDA/CCDA | [support found?] | [CONFORMANT/PARTIAL/NONE] | N |
| HL7v2 | [usage found?] | [VALID/PARTIAL/NONE] | N |

### FHIR Resource Mapping

| Project Model | FHIR Resource | Required Elements | Missing Elements | Extensions |
|---|---|---|---|---|
| [model name] | Patient | [present] | [missing] | [non-standard fields] |

### Data Model Findings

| # | Severity | File | Model/Field | Standard | Issue | Fix |
|---|----------|------|-------------|----------|-------|-----|
| 1 | High | path/to/model.ts | Patient.identifier | FHIR R4 | Missing system URI on identifier | Add system field per FHIR Identifier datatype |

### Terminology Gaps

| Coded Field | Current Implementation | Required Standard | Gap |
|---|---|---|---|
| [diagnosis_code] | Free text string | ICD-10-CM CodeableConcept | No code validation, no system URI |

### Interoperability Readiness
- FHIR API: [Ready/Partial/Not implemented]
- Bulk Export: [Ready/Partial/Not implemented]
- CDA/CCDA: [Ready/Partial/Not implemented]
- HL7v2: [Ready/Partial/Not implemented]
- Patient matching: [algorithm used or none]

### Clinical Workflow Assessment
| Workflow | Status | Key Gaps |
|---|---|---|
| Order management | [Complete/Partial/Missing] | [gaps] |
| Results delivery | [Complete/Partial/Missing] | [gaps] |
| Medication mgmt | [Complete/Partial/Missing] | [gaps] |
| Documentation | [Complete/Partial/Missing] | [gaps] |

RULES

• Do NOT modify any code -- this is a review skill, not a build skill.
• Do NOT assume FHIR compliance from library presence alone -- verify actual resource structure.
• Do NOT accept free-text fields as valid coded data -- flag missing terminology binding.
• Do NOT skip checking cardinality and data types against FHIR spec.
• Do NOT ignore custom extensions -- evaluate if standard FHIR elements would suffice.
• Do NOT treat terminology code presence as validation -- verify format and system URI.
• Do NOT overlook data quality rules -- missing validation is a finding.
• Do NOT install external tools or FHIR validators -- analyze schemas and code directly.

NEXT STEPS

• "Run /healthcare-compliance to audit regulatory compliance of the clinical data layer."
• "Run /compliance-ops to evaluate broader organizational compliance operations."
• "Run /api-surface to evaluate API design patterns for clinical endpoints."

============================================================ SELF-EVOLUTION TELEMETRY ============================================================

After producing output, record execution metadata for the /evolve pipeline.

Check if a project memory directory exists:

• Look for the project path in ~/.claude/projects/
• If found, append to skill-telemetry.md in that memory directory

Entry format:

### /clinical-data-review — {{YYYY-MM-DD}}
- Outcome: {{SUCCESS | PARTIAL | FAILED}}
- Self-healed: {{yes — what was healed | no}}
- Iterations used: {{N}} / {{N max}}
- Bottleneck: {{phase that struggled or "none"}}
- Suggestion: {{one-line improvement idea for /evolve, or "none"}}

Only log if the memory directory exists. Skip silently if not found. Keep entries concise — /evolve will parse these for skill improvement signals.