Medical Report to JSON — OpenClaw Skill

Extract structured health data from medical report images or PDFs and save as JSON files for long-term tracking and querying by your AI health assistant.

Features

• 📷 Image & PDF input — supports PNG, JPG, WebP images and PDF files
• 🌍 Multi-language — handles Chinese, German, English, and other languages; all output in English
• 📑 Multi-page reports — send multiple images across messages for a single report
• 🔀 Auto-categorization — splits multi-category reports into separate JSON files
• 💉 Vaccination booklet — accumulative vaccination record in a single file
• 🔍 Query & compare — agent can list, read, and compare saved records across dates
• 🔒 Privacy-first — no PII extraction; user redacts before sending

Installation

Copy the skill directory to your agent's workspace:

cp -r medical-report-to-json/ ~/.openclaw/workspace-<AGENT-ID>/skills/

Or for all agents:

cp -r medical-report-to-json/ ~/.openclaw/skills/

PDF Support

The skill includes a Python helper for PDF conversion. Install the dependency:

pip3 install pymupdf

The agent will also prompt to install this automatically on first PDF encounter.

Usage

Sending a Report

1. Redact personal information from your report image/PDF.
2. Send the image to your OpenClaw agent.
3. The agent extracts the data and saves a JSON file.

Single image:

"Here's my blood test from March 15th."

Multi-page report:

"Blood test page 1 of 3" → send image "Page 2 of 3" → send image "Last page" → send image

PDF file:

"Process this lab report PDF" → attach PDF

Querying Records

"List all my medical records"

"What were my last blood test results?"

"Compare my cholesterol levels over the past year"

"Show my vaccination record"

Output

Storage Location

All records are saved to:

<workspace>/memory/medical_records/

File Naming

Type	Pattern	Example
Standard report	{YYYY-MM-DD}_{category}_{seq:02d}.json	2024-03-15_blood_routine_01.json
Vaccination record	vaccination_record.json	vaccination_record.json

Supported Categories

Category	Description
blood_routine	CBC, complete blood count
blood_chemistry	Metabolic panel, CMP, BMP
liver_function	ALT, AST, bilirubin, albumin
kidney_function	Creatinine, BUN, GFR
lipid_panel	Cholesterol, triglycerides, LDL, HDL
blood_glucose	Fasting glucose, HbA1c
thyroid_function	TSH, FT3, FT4
urine_routine	Urinalysis
coagulation	PT, INR, APTT
tumor_markers	AFP, CEA, PSA
hormone_panel	Testosterone, estrogen, cortisol
immune_panel	IgG, IgM, complement
infection_markers	CRP, PCT, ESR
imaging_report	X-ray, CT, MRI, ultrasound
ecg	Electrocardiogram
pathology	Biopsy, cytology
vaccination_record	Vaccination booklet (accumulative)
other	Anything not fitting above

JSON Schema

Standard Report

{
  "id": "2024-03-15_blood_routine_01",
  "category": "blood_routine",
  "date": "2024-03-15",
  "items": [
    {
      "name": "WBC",
      "value": 7.2,
      "unit": "10^9/L",
      "ref_range": "3.5-9.5",
      "flag": "normal"
    }
  ],
  "findings": "All values within normal range."
}

Fields:

Field	Type	Required	Description
id	string	✅	{date}_{category}_{seq}
category	string	✅	Category slug
date	string	✅	Report date YYYY-MM-DD
items	array	✅	Test result items
items[].name	string	✅	Standard English abbreviation
items[].value	number/string	✅	Numeric or qualitative
items[].unit	string	✅	Standard unit
items[].ref_range	string	❌	Reference range
items[].flag	string	✅	normal, high, low, critical_high, critical_low, abnormal, positive, negative
findings	string	❌	Brief summary of abnormalities

Vaccination Record

{
  "id": "vaccination_record",
  "category": "vaccination_record",
  "last_updated": "2024-11-20",
  "items": [
    {
      "vaccine": "COVID-19 mRNA (Pfizer-BioNTech) Dose 1",
      "date": "2021-06-15"
    }
  ]
}

Unique behavior:

• Single accumulative file: vaccination_record.json
• New entries are appended, never overwritten
• Only vaccine (name/type) and date per entry

Privacy

This skill is designed with privacy in mind:

• User responsibility: Redact PII before sending images to the agent
• No PII extraction: The skill instructions explicitly tell the agent to ignore personal information
• Local storage only: JSON files are stored locally in the agent's workspace
• No external API calls: Vision analysis uses the agent's built-in LLM capability

Skill Structure

medical-report-to-json/
├── SKILL.md                    # Skill definition and instructions
├── scripts/
│   └── pdf_to_images.py        # PDF → PNG converter (PyMuPDF)
├── references/
│   └── schema_example.json     # JSON schema examples
└── README.md                   # This file

License

MIT