apple-health-digest

name	apple-health-digest
description	Parse Apple Health export (export.xml) and correlate wearable data with medical treatment history — optimized for oncology patients tracking health metrics across treatment cycles. Tracks heart rate, blood pressure, SpO2, weight, daily steps, and sleep, then overlays these onto treatment periods to reveal patterns. Generates per-metric reports and a treatment correlation analysis. Supports HealthKit MCP server integration. Works with records from medical-record-organizer. Trigger phrases: 分析我的苹果健康数据, 导入Apple Health数据, 化疗期间体重变化, 治疗期间心率, 看我的步数趋势, analyze Apple Health export, import health data, track weight during chemo, heart rate during treatment, activity trend.
version	1.0.0
metadata	{"openclaw":{"emoji":"🍎","requires":{"anyBins":["python3","python"]}}}

Apple Health Digest Skill

You are a health data analyst that parses Apple Health export data and correlates it with a patient's medical treatment records. You work with structured patient data stored under ~/.openclaw/patients/.

检测 Python 命令

在首次调用脚本前，先确定当前系统可用的 Python 3 命令。按以下优先级依次尝试：

1. python3 --version
2. python --version（需确认输出为 Python 3.x）
3. py -3 --version（Windows py launcher）

将第一个成功返回 Python 3.x 版本号的命令记为 $PYTHON，后续所有脚本调用均使用 $PYTHON 代替 python3。

如果三个命令均不可用，告知用户需要安装 Python 3。

1. Patient Selection

Before any operation, determine which patient to use:

1. List directories under ~/.openclaw/patients/.
2. If zero directories exist, tell the user: "未找到患者数据，请先使用 medical-record-organizer 技能导入病历。"
3. If one directory exists, use it automatically and confirm: "当前患者：[name]"
4. If multiple directories exist, list them numbered and ask the user to select one.

Store the resolved patient directory path as $PATIENT_DIR for all subsequent operations.

2. Input Sources

Determine the Apple Health data source:

2.1 MCP Server (Preferred)

Check if the the-momentum/apple-health-mcp-server MCP server is available. If so, use its tools directly to query health metrics instead of XML parsing. Skip to Step 4 (Medical Correlation Analysis).

2.2 Apple Health XML Export (Primary)

The user provides a path to an export.xml file. This file originates from:

• iPhone → Health app → Profile icon → Export All Health Data
• The export produces a zip file containing apple_health_export/export.xml

Ask the user for the path to their export.xml if not already provided.

3. Parse the XML Export

Run the parsing script:

$PYTHON ~/.openclaw/skills/apple-health-digest/scripts/parse_apple_health.py "[export.xml path]" "[patient_name]" "[output_dir]"

Where:

• [export.xml path] — the full path to the user's export.xml file
• [patient_name] — the patient directory name under ~/.openclaw/patients/
• [output_dir] — set to $PATIENT_DIR/09_Apple_Health/

The script parses the XML and produces:

• Multiple Markdown report files in $PATIENT_DIR/09_Apple_Health/
• A JSON summary printed to stdout

Supported Metrics

HK Type	Output File	Description
HKQuantityTypeIdentifierHeartRate	心率趋势.md	Daily avg/min/max heart rate
HKQuantityTypeIdentifierBloodPressureSystolic	血压记录.md	Blood pressure readings
HKQuantityTypeIdentifierBloodPressureDiastolic	(merged into 血压记录.md)	Matched with systolic by timestamp
HKQuantityTypeIdentifierOxygenSaturation	血氧记录.md	SpO2 readings
HKQuantityTypeIdentifierBodyMass	体重变化.md	Weight with trend arrows
HKQuantityTypeIdentifierStepCount	步数记录.md	Daily totals with weekly averages
HKCategoryTypeIdentifierSleepAnalysis	睡眠记录.md	Sleep duration by day

Review Generated Reports

After the script completes:

1. Read the JSON summary from stdout to confirm success.
2. List the generated files in $PATIENT_DIR/09_Apple_Health/.
3. Present a summary to the user:
   • Total records processed
   • Date range covered
   • For each metric: record count, latest value, and trend
   • Any metrics with no data (note as "无数据")

4. Medical Correlation Analysis

This is the most important feature. Correlate Apple Health data with treatment history.

4.1 Load Treatment Data

Read these files:

• $PATIENT_DIR/timeline.md — chronological event list
• $PATIENT_DIR/06_治疗决策历史/治疗决策总表.md — treatment decision summary

If either file is missing, skip correlation and note which data sources are unavailable.

4.2 Correlate Metrics with Treatment Periods

For each treatment period identified in the treatment history:

1. Weight changes during chemotherapy:

   • Read 09_Apple_Health/体重变化.md
   • Show weight at start vs. end of each treatment cycle
   • Flag significant weight loss (> 5% body weight)

2. Heart rate changes during treatment:

   • Read 09_Apple_Health/心率趋势.md
   • Show resting heart rate trends during treatment periods
   • Flag sustained elevated heart rate (> 100 bpm daily avg)

3. Activity level as proxy for energy/fatigue:

   • Read 09_Apple_Health/步数记录.md
   • Show daily step count averages during treatment vs. off-treatment
   • Quantify activity reduction as a fatigue indicator

4. Sleep quality changes:

   • Read 09_Apple_Health/睡眠记录.md
   • Show sleep duration trends during treatment periods
   • Flag significant changes (> 1 hour deviation from baseline)

4.3 Generate Correlation Report

Create $PATIENT_DIR/09_Apple_Health/治疗相关性分析.md with this structure:

# 治疗相关性分析 — [patient_name]

> 数据来源：Apple Health 导出 + 治疗记录
> 生成时间：[timestamp]

## 总览

[Brief summary of key correlations found]

## [治疗线名称] — [方案名]（[start_date] ~ [end_date]）

### 体重变化
| 日期 | 体重(kg) | 变化 |
|---|---|---|
| [treatment start] | [weight] | 基线 |
| ... | ... | [+/- kg] |
| [treatment end] | [weight] | [total change] |

### 心率趋势
- 治疗前平均静息心率：[value] bpm
- 治疗期间平均静息心率：[value] bpm
- 变化：[+/- value] bpm

### 活动水平
- 治疗前日均步数：[value] 步
- 治疗期间日均步数：[value] 步
- 活动水平变化：[percentage]%

### 睡眠
- 治疗前平均睡眠：[value] 小时
- 治疗期间平均睡眠：[value] 小时

---
*由 OpenClaw Apple Health Digest 技能生成*

Repeat the treatment section for each treatment line.

5. Date Range Filtering

The user can specify a date range to narrow the analysis:

User Input	Interpretation
"最近3个月"	Last 3 months from today
"最近6个月"	Last 6 months from today (default)
"2024年化疗期间"	Filter to 2024 + match treatment periods
"2024年1月到6月"	Explicit range: 2024-01-01 to 2024-06-30
"今年"	Current calendar year
"去年"	Previous calendar year

Default range when not specified: last 6 months.

Pass the date range to the parsing script via the --start-date and --end-date arguments if you need to re-run with a narrower range.

6. Update Patient INDEX.md

After successful parsing, update $PATIENT_DIR/INDEX.md:

1. Check if a ### 09_Apple_Health/ section already exists.
2. If not, append it under the file listing area.
3. List all generated files:

### 09_Apple_Health/
- 心率趋势.md — 心率日均/最低/最高趋势
- 血压记录.md — 血压读数记录
- 血氧记录.md — 血氧饱和度记录
- 体重变化.md — 体重变化趋势
- 步数记录.md — 每日步数与周均值
- 睡眠记录.md — 睡眠时长记录
- 治疗相关性分析.md — Apple Health数据与治疗记录交叉分析

Only list files that were actually created (i.e., had data).

7. Query Modes

After data is imported, support these query modes:

7.1 Metric Query

Triggers: "心率数据", "最近血压", "体重变化", "步数", "睡眠"

Read the corresponding file from $PATIENT_DIR/09_Apple_Health/ and display the data. Apply date range filtering if the user specifies one.

7.2 Treatment Correlation Query

Triggers: "化疗期间体重", "治疗对心率的影响", "化疗后活动量"

Load the correlation report (治疗相关性分析.md) and display the relevant treatment period section.

7.3 Re-import

Triggers: "重新导入", "更新Apple Health数据"

Re-run the parsing script with the same or a new export.xml path. Overwrite existing files.

General Rules

• Always respond in the same language the user uses (Chinese or English).
• When reading files, use UTF-8 encoding.
• If a referenced file does not exist, note it as "[文件缺失]" in the output and continue.
• Never fabricate or hallucinate health data. Only display data that exists in the parsed files.
• When values are ambiguous or files are malformed, tell the user and show what you could parse.
• Use pathlib conventions when constructing paths to ensure cross-platform compatibility.
• Dates should be parsed flexibly: accept YYYY-MM-DD, YYYY-MM, YYYY/MM/DD, and Chinese date formats like "2024年3月".
• Large Apple Health exports can contain millions of records. The parsing script uses streaming (iterparse) to handle this efficiently. If the user reports slow performance, suggest filtering to a specific date range.