Analyze Health Trends and Generate Insights

Health Trends Analyzer is an agent skill that analyzes health data over a period of time, detects changes and correlations, and produces data-driven insights. It answers questions such as "what has changed in my health lately?" and covers five dimensions: weight and BMI, symptoms, medication adherence, lab results (cholesterol, blood glucose, blood pressure and similar markers), and mood together with sleep.

Beyond per-dimension trends it offers three further layers:

• Correlation analysis checks whether a new medication lines up with a change in symptoms, how diet and sleep relate to symptoms and mood, whether a treatment produces improvement, and, for women's health tracking, how the menstrual cycle relates to symptoms.
• Change detection warns about rapid weight change, new symptoms and medication changes, spots early signs of decline, highlights improvements, and raises threshold alerts near dangerous levels such as extreme BMI or radiation dose.
• Predictive insights derive risk factors from trends, suggest preventive measures and give early warnings before a problem becomes serious.

The default window is the past 3 months. Users can ask for the past 1 month, 6 months or 1 year, or a range such as "since January 2025" or "the last 90 days". Execution runs in eight steps: fix the time range, read the data files, filter by date, analyze each dimension, run correlations, detect changes, generate insights, and render a report. Trend detection uses linear regression, seasonality analysis and outlier detection. Correlation uses Pearson for continuous variables, Spearman for ordinal ones and cross-correlation for time series. Change-point detection uses CUSUM, a sliding-window t-test and Bayesian methods, and summary statistics include mean, median, standard deviation, percentiles and period-over-period change.

Use it when the task asks about trends, correlations or notable changes across weight, symptoms, medication, labs, mood or sleep, or when someone wants a trend report. Sample requests are "analyze my health trends", "is my medication working?", "what is my sleep-mood relationship?" and "does my blood pressure medicine work?", which links the medication start date to blood pressure readings and symptom improvement.

Do not use it for medical diagnosis, specific medication advice or life-and-death prognosis. Analysis relies only on recorded data: it does not guess at missing information, it labels data sources and time ranges, and its advice should be reviewed by medical professionals. Every report carries a disclaimer that it is for reference only and does not replace professional diagnosis. It is also no substitute for environment-specific validation, testing or expert review, and it stops to ask for clarification when inputs, permissions or success criteria are missing.

Missing data is handled explicitly. With no data it recommends recording that data type first, with under a month of data it says trends need at least one month, and when the range is narrow it uses what exists and suggests a longer record. If charts cannot render, it falls back to a text report.

Inputs are local JSON files: data/profile.json (weight, height, BMI history), data/symptoms/**/*.json, data/mood/**/*.json (mood score, sleep quality, stress), data/diet/**/*.json, data/medication-logs/**/*.json and lab results under data/medical_records/**/*.json. Supporting files are data/cycle-tracker.json, data/pregnancy-tracker.json, data/menopause-tracker.json, data/allergies.json and data/radiation-records.json. Records are filtered to the requested window with a function of this shape:

function filterByDate(data, startDate, endDate) {
  return data.filter(item => {
    const itemDate = new Date(item.date || item.created_at);
    return itemDate >= startDate && itemDate <= endDate;
  });
}

The concise text report gives an overall assessment (improving, stable, needs attention), then per-dimension sections, a correlation list, risk notes and a follow-up plan. The sample shows weight down 2.3 kg over three months, medication adherence of 78 percent, cholesterol falling from 240 to 210 mg/dL and a strong sleep-to-mood correlation (r = 0.78). The full HTML report is a self-contained file with seven blocks: an overview card, a weight and BMI dual-axis line chart, a color-coded symptom frequency bar chart, a medication adherence gauge, a multi-series lab trend chart with reference lines, a correlation heatmap and a mood and sleep area chart. It supports zoom, export and legend toggling, works on mobile, prints cleanly and can be sent to a doctor. Files are written to data/health-reports/.

All processing stays local: no external API calls are made and no data is transmitted, and the HTML report runs standalone with chart libraries loaded from a CDN. The skill uses only Read, Grep, Glob and Write, and reads incrementally with caching and on-demand chart data. It pairs with the commands /symptom, /mood, /diet, /medication and /query that record and look up the underlying data.

It suits people who track their own health data locally and want a periodic view of trends and correlations, and builders of personal health tracking tools who need a report they can hand to a physician.