Source: https://github.com/aipoch/medical-research-skills

HIPAA Compliance Auditor

A clinical-grade PII/PHI detection and de-identification tool for healthcare text data.

Quick Check

Use this command to verify that the packaged script entry point can be parsed before deeper execution.

python -m py_compile scripts/main.py

Audit-Ready Commands

Use these concrete commands for validation. They are intentionally self-contained and avoid placeholder paths.

python -m py_compile scripts/main.py
python scripts/main.py --help
python scripts/main.py --text "Audit validation sample with explicit methods, findings, and conclusion."

When to Use

• Use this skill when the task needs A clinical-grade PII/PHI detection and de-identification tool for healthcare text data.
• Use this skill for academic writing tasks that require explicit assumptions, bounded scope, and a reproducible output format.
• Use this skill when you need a documented fallback path for missing inputs, execution errors, or partial evidence.

When NOT to Use

• Do NOT use for DICOM image de-identification (use dicom-anonymizer skill)
• Do NOT use for general data anonymization of structured databases (use dedicated tools)
• Do NOT use as a legal compliance certification — this tool assists but does NOT replace human HIPAA review

Workflow

Step 1: Provide Input Text

Provide clinical text in one of two ways:

• File input: python scripts/main.py --input patient_text.txt --output deidentified.txt
• Direct text: python scripts/main.py --text "Patient John Doe, SSN 123-45-6789..." --audit-log audit.json

If neither input provided: Request the text or file path from the user. Do not proceed without input.

Step 2: Configure Detection Parameters

• --confidence 0.7 (default): Minimum confidence threshold (0.0-1.0). Lower = more detections but more false positives.
• --preserve-structure true (default): Maintain document formatting after redaction
• --custom-patterns <path>: Optional custom regex patterns JSON for institution-specific identifiers

🔍 Checkpoint 1: If confidence threshold is changed from default, inform the user about the trade-off (lower threshold catches more PII but may produce more false positives).

Step 3: Run De-identification

from scripts.main import HIPAAAuditor

auditor = HIPAAAuditor()
result = auditor.deidentify("Patient John Doe was admitted on 2024-01-15...")
# result.cleaned_text → De-identified output
# result.detected_pii → List of found PII entities with types and confidence scores

If de-identification fails (missing spaCy model): Install with python -m spacy download en_core_web_trf, then retry.

Step 4: Review Detection Results

Check the audit log for:

• High-confidence detections (≥0.9): Verify replacements are correct
• Medium-confidence detections (0.7-0.9): Manual review recommended
• Categories detected: Confirm all 18 HIPAA identifier categories were scanned

If unexpected PII remains: Increase custom patterns or lower confidence threshold.

Step 5: Output and QA Reminder

• Output de-identified text to file or stdout
• Generate audit log JSON with all detection details
• ⚠️ CRITICAL REMINDER: This tool is a helper, NOT a replacement for human review. Always perform manual QA before HIPAA-compliant release.

Overview

This skill analyzes text for HIPAA-protected identifiers and automatically redacts or anonymizes them. It uses a combination of regex patterns, NLP entity recognition, and contextual analysis to identify 18 HIPAA identifier categories.

Features

• 18 HIPAA Identifiers Detection: Names, dates, SSN, MRN, phone/fax, email, geographic data, etc.
• Automatic De-identification: Replace PII with semantic tokens (e.g., [PATIENT_NAME], [DATE_1])
• Context-Aware Detection: Distinguishes between similar patterns (dates vs. lab values)
• Audit Logging: Track all redaction actions for compliance documentation
• Confidence Scoring: Flag uncertain detections for manual review

Usage

Command Line

python scripts/main.py --input "patient_text.txt" --output "deidentified.txt"
python scripts/main.py --text "Patient John Doe, SSN 123-45-6789..." --audit-log audit.json

Python API

from scripts.main import HIPAAAuditor

auditor = HIPAAAuditor()
result = auditor.deidentify("Patient John Doe was admitted on 2024-01-15...")
print(result.cleaned_text)  # De-identified output
print(result.detected_pii)  # List of found PII entities

Parameters

Parameter	Type	Default	Required	Description
--input, -i	string	-	No	Path to input text file
--text	string	-	No	Direct text input (alternative to file)
--output, -o	string	-	No	Path for de-identified output file
--audit-log	string	-	No	Path for JSON audit log
--confidence	float	0.7	No	Minimum confidence threshold (0.0-1.0)
--preserve-structure	bool	true	No	Maintain document structure
--custom-patterns	string	-	No	Path to custom regex patterns JSON

HIPAA Identifier Categories Detected

1. Names (patient, relatives, employers)
2. Geographic subdivisions smaller than state
3. Dates (except year) related to individual
4. Phone numbers
5. Fax numbers
6. Email addresses
7. SSN
8. Medical record numbers
9. Health plan beneficiary numbers
10. Account numbers
11. Certificate/license numbers
12. Vehicle identifiers
13. Device identifiers
14. URLs
15. IP addresses
16. Biometric identifiers
17. Full-face photos
18. Any other unique identifying numbers

Output Format

De-identified Text

Original identifiers replaced with semantic tags:

• [PATIENT_NAME_1], [PATIENT_NAME_2] ...
• [DATE_1], [DATE_2] ...
• [SSN_1]
• [PHONE_1], [PHONE_2] ...
• [EMAIL_1]
• [MRN_1] (Medical Record Number)
• [ADDRESS_1]

Audit Log JSON

{
  "timestamp": "2024-01-15T10:30:00Z",
  "input_hash": "sha256:abc123...",
  "detections": [
    {
      "type": "PATIENT_NAME",
      "position": [10, 18],
      "confidence": 0.95,
      "replacement": "[PATIENT_NAME_1]",
      "original_length": 8
    }
  ],
  "statistics": {
    "total_pii_found": 5,
    "categories_detected": ["NAME", "DATE", "PHONE", "SSN"]
  }
}

Technical Architecture

1. Preprocessing: Normalize text encoding, handle line breaks
2. Regex Engine: Pattern matching for structured identifiers (SSN, phone, email, MRN)
3. NLP Pipeline: spaCy NER for names, organizations, locations
4. Context Filter: Remove false positives (e.g., "Dr. Smith" vs. "smith fracture")
5. Replacement Engine: Sequential replacement with semantic tokens
6. Validation: Ensure no original PII remains in output

Dependencies

• Python 3.9+
• spaCy (en_core_web_trf or en_core_web_lg)
• regex (for advanced pattern matching)
• Presidio (optional, for enhanced PII detection)

See references/requirements.txt for full dependency list.

Limitations & Warnings

⚠️ CRITICAL: This tool is designed as a helper, not a replacement for human review.

• Context-dependent PII (e.g., rare disease names + location) may not be fully detected
• Unstructured narrative text may contain identifying information not caught by patterns
• Always perform manual QA on output before HIPAA-compliant release
• AI Autonomous Acceptance Status: Requires Manual Review (Requires Manual Review)

References

• references/hipaa_safe_harbor_guide.pdf - HIPAA Safe Harbor de-identification standards
• references/pii_patterns.json - Complete regex pattern definitions
• references/test_cases/ - Sample clinical texts with expected outputs
• references/requirements.txt - Python dependencies

Technical Difficulty: High

Complex NLP pipelines, contextual disambiguation, regulatory compliance requirements.

Risk Assessment

Risk Indicator	Assessment	Level
Code Execution	Python/R scripts executed locally	Medium
Network Access	No external API calls	Low
File System Access	Read input files, write output files	Medium
Instruction Tampering	Standard prompt guidelines	Low
Data Exposure	Output files saved to workspace	Low

Security Checklist

• No hardcoded credentials or API keys
• No unauthorized file system access (../)
• Output does not expose sensitive information
• Prompt injection protections in place
• Input file paths validated (no ../ traversal)
• Output directory restricted to workspace
• Script execution in sandboxed environment
• Error messages sanitized (no stack traces exposed)
• Dependencies audited

Prerequisites

# Python dependencies
pip install -r requirements.txt

Evaluation Criteria

Success Metrics

• Successfully executes main functionality
• Output meets quality standards
• Handles edge cases gracefully
• Performance is acceptable

Test Cases

1. Basic Functionality: Standard input → Expected output
2. Edge Case: Invalid input → Graceful error handling
3. Performance: Large dataset → Acceptable processing time

Lifecycle Status

• Current Stage: Draft
• Next Review Date: 2026-03-06
• Known Issues: None
• Planned Improvements:
  • Performance optimization
  • Additional feature support

Output Requirements

Every final response should make these items explicit when they are relevant:

• Objective or requested deliverable
• Inputs used and assumptions introduced
• Workflow or decision path
• Core result, recommendation, or artifact
• Constraints, risks, caveats, or validation needs
• Unresolved items and next-step checks

Error Handling

• If required inputs are missing, state exactly which fields are missing and request only the minimum additional information.
• If the task goes outside the documented scope, stop instead of guessing or silently widening the assignment.
• If scripts/main.py fails, report the failure point, summarize what still can be completed safely, and provide a manual fallback.
• Do not fabricate files, citations, data, search results, or execution outcomes.

Input Validation

This skill accepts requests that match the documented purpose of hipaa-compliance-auditor and include enough context to complete the workflow safely.

Do not continue the workflow when the request is out of scope, missing a critical input, or would require unsupported assumptions. Instead respond:

hipaa-compliance-auditor only handles its documented workflow. Please provide the missing required inputs or switch to a more suitable skill.

Response Template

Use the following fixed structure for non-trivial requests:

1. Objective
2. Inputs Received
3. Assumptions
4. Workflow
5. Deliverable
6. Risks and Limits
7. Next Checks

If the request is simple, you may compress the structure, but still keep assumptions and limits explicit when they affect correctness.