How to use the AI Health Data Analyzer agent skill (ai-analyzer)

What is an agent skill, and what does ai-analyzer make the agent do?

An agent skill is a folder containing a SKILL.md file that tells an AI agent when and how to perform a specific task. When the skill is installed and your request matches its description, the agent follows its instructions.

With ai-analyzer installed, the agent reads the health files you supply, such as a profile, lab results, and fitness, sleep, nutrition and mental health trackers, plus medications and allergies if you provide them. It then summarises trends, flags statistical outliers, describes general risk factors, answers questions about your data, and can write a text or HTML report.

The skill states that it is not a clinically validated prediction system. It must not give diagnoses, dosing advice or prognoses.

• Integrates multiple data sources into one analysis
• Detects anomalies and change points using methods such as CUSUM and Z-score, as described in the skill
• Calculates correlations between metrics, such as sleep and mood
• Writes a text report or an HTML report

How do you install the ai-analyzer skill?

Download the ZIP for ai-analyzer from the skill's page in the directory. Unzip it so that the skill folder sits in one of these locations: ~/.claude/skills/ for all your projects, or .claude/skills/ inside a specific project folder.

Alternatively, in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

The catalog copy includes only SKILL.md. The skill's HTML report script is not bundled, so check the file list before relying on it.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• App install: Settings → Capabilities → Skills

What files and tools does the skill need?

You supply the data. The skill expects local JSON files, for example data/profile.json and data-example/*-tracker.json such as fitness-tracker.json, sleep-tracker.json, nutrition-tracker.json and mental-health-tracker.json. Medication and allergy files (data/medications.json and data/allergies.json) are optional. Missing files are treated as missing information, not as normal or zero values.

The skill uses Read, Grep, Glob and Write tool access. It is written to read JSON data and write HTML reports and history records.

The optional HTML report depends on scripts/generate_ai_report.py. That script is not included in this catalog copy and has no verified download source. Without it, the text report still works, and the HTML step is unavailable.

• Local JSON health files you provide
• Read, Grep, Glob and Write tool access
• Optional: scripts/generate_ai_report.py for HTML reports (not bundled)

How do you trigger the skill with natural requests?

Ask in plain language about your health data. The skill is written to respond to general requests about your status, risks and reports, and to specific ones about sleep, exercise or blood pressure risk.

Be clear about what you want: a summary, a risk discussion, a question about one tracker, or a report. The more specific the request, the better the agent can choose which files to read.

• "AI分析我的健康状况" (analyse my health status with AI)
• "我的健康有什么风险？" (what health risks do I have?)
• "我的睡眠怎么样？" (how is my sleep?)
• "运动对我的健康有什么影响？" (how does exercise affect my health?)
• "生成AI健康报告" (generate an AI health report)
• "我应该如何改善健康状况？" (how should I improve my health?)

What do example outputs look like, and what stays out of scope?

A typical answer describes trends in your files, notes outliers, and lists general risk factors with the caveat that the output is for reference only. Correlations are reported as associations, not causes. A low nutrient intake against a reference amount does not by itself prove a deficiency, and a sleep questionnaire does not diagnose a sleep disorder.

The skill will not calculate a disease probability unless a specific, validated model applies to your group and its required inputs are present. Otherwise it describes risk factors only. Its ADA diabetes screening score is not a personal 10-year onset probability.

• Good fit: a structured summary of your trends to bring to a doctor
• Good fit: exploring correlations in your own lifestyle data
• Not a fit: a diagnosis, medication guidance or disease probability
• Not a fit: emergency symptoms. Follow local emergency procedures instead of waiting for a report

What safety and privacy points should you know?

The skill runs no scripts by itself and makes no network calls, according to the catalog listing. Its only optional external dependency is the HTML report script, which is absent here. Your files are still read into the host model's context, so the skill cannot guarantee that data stays offline or is not processed externally.

Use it only with data you are willing to share with the model you are using. Do not use it on shared or cloud-connected setups for sensitive records unless you have authorised that. Keep private records in a private location, and escape untrusted text in any HTML output.

• Do not use on shared or cloud-connected setups for sensitive records without authorisation
• Level 3 advice (medical) includes a disclaimer and should be confirmed by a doctor
• High-risk results should prompt you to consult a doctor
• Emergency symptoms: follow local emergency procedures first

What tips help you get useful results?

Keep your data files consistent and dated, so trends are comparable. Small samples make correlations and outlier checks unreliable, so a few weeks of data will give limited conclusions. Bring the report to your doctor as a discussion aid, not a verdict.

Ask one question at a time, and ask the agent to state what data it used and what it could not check.

• Use consistent units and time periods across trackers
• Provide more data points before reading much into a trend
• Verify numbers against your original lab reports
• Ask what was missing from the analysis