apple-health-analyst
Analyze Apple Health export ZIP. Run local prepare to generate structured insights, then produce complementary health and training reports with long-term context.
RuochenLyu · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add RuochenLyu/apple-health-analyst --skill apple-health-analyst技能说明
name: apple-health-analyst description: Analyze Apple Health export ZIP. Run local prepare to generate structured insights, then produce complementary health and training reports with long-term context.
Apple Health Advisor
Use this skill when a user wants to analyze an Apple Health export ZIP. The ZIP is too large to fit directly into context, so the skill uses a local CLI pipeline to parse and structure the data first.
This is one skill that ships two complementary reports:
- Default (recommended): generate BOTH reports — health report + training report, rendered into the same
output/folder and cross-linked via an in-page link in the topbar.prepareruns once, thenrenderruns twice. - Explicit health-only: user says "只要健康报告" / "health report only" / similar → skip training render.
- Explicit training-only: user says "只要运动报告" / "training report only" / similar → skip health render. Naming a sport without saying "only" still produces both reports and prioritizes that sport in the training narrative.
Language Detection
Detect the user's language from their message:
- If the user writes in Chinese, use
--lang zh - For all other languages, use
--lang en
The narrative language must match the language declared in insights.json:
- Health report:
narrativeContext.language - Training report:
training.narrativeContext.language
Intent Routing
Default: generate both health + training reports. Only drop one when the user is explicit.
- Default (both) examples:
Analyze my Apple Health export帮我分析 Apple Health 导出Generate a report from my Apple Health export
- Health-only examples (skip training render):
Only generate the health report只要健康报告不用生成运动报告
- Training-only examples (skip health render):
Only generate the training report只要运动报告只生成运动报告,重点分析拳击Only generate the training report for running and cycling
- Named-sport examples that still default to both:
重点分析拳击训练状态分析跑步和骑行训练趋势
Ambiguous-but-lean-training keywords still default to both reports (the training report alone is rarely enough context). The keywords below only matter as hints — they do NOT suppress the health report unless the user also says "只" / "only":
training,workout,运动,训练,专项- named sports such as
boxing,running,cycling,walking,hiking,strength training,拳击,跑步,骑行,力量训练
Your Role
Two roles share the same pipeline:
- Health management advisor:
- Integrate sleep, recovery, activity, and body metrics into an overall health view
- Prioritize cross-metric reasoning over metric-by-metric reporting
- Training status advisor:
- Judge load, recovery support, consistency, and sport-specific trends
- Give actionable training-management advice without pretending to be Garmin or a coach writing a periodized plan
Workflow
- Confirm the input is an official Apple Health export ZIP and that the main XML has
HealthDataas its root node. - Run local
prepareonce with the correct--lang, producingsummary.jsonandinsights.json. Raw ZIP/XML parsing stays local. If the active agent uses a hosted model for the narrative, the structured JSON it reads may be processed by that model provider; follow the provider's data controls and the repository privacy notice. - Read
summary.json, theninsights.json. - Decide which reports to produce (default = both unless the user is explicit; see Intent Routing).
- For each selected report, write the narrative JSON:
- health:
report.llm.json - training:
training.report.llm.json
- health:
- Run
renderfor each selected report:- health: default
render - training:
render --type training
- health: default
- Both HTML reports share the same
output/folder. The topbar carries a cross-link between them, so the user can jump back and forth. File names are fixed (report.html↔training.report.html) — do not rename.
insights.json Keys You Must Use
Shared
| Key | What it contains |
|---|---|
metadata | tool, version, language, schemaVersion, generatedAt |
historicalContext | Recent 30d, baseline 90d, trailing 180d, all-time context |
charts[] | Health chart groups |
crossMetric | Cross-metric health reasoning |
riskFlags[] | Health risks with evidence |
notableChanges[] | Significant changes |
dataGaps[] | Missing or sparse data warnings |
sourceConfidence[] | Device/source reliability signals |
Health report
| Key | What it contains |
|---|---|
analysis.sleep | Sleep duration, stages, timing, regularity |
analysis.recovery | RHR, HRV, blood oxygen, respiratory rate, VO2 max |
analysis.activity | Active energy, exercise minutes, stand hours, workouts |
analysis.bodyComposition | Weight, body fat % |
analysis.menstrualCycle | Cycle analysis if present |
narrativeContext | Health-report audience, goal, schema version, boundaries |
Training report
| Key | What it contains |
|---|---|
training.summary | Training state, readiness, recent load, recovery support, primary sport |
training.summary.trainingLoad | MET-minute EWMA load snapshot (42-day CTL baseline, 7-day ATL recent load, TSB load balance) + relative 30-day & 90-day changes; null when < 28 days of data or < 6 load-bearing workouts |
training.sports[] | Top sports (dormant ones filtered, topSportCount configurable via --top-sports, default 5) with recent/baseline/trailing/all-time windows, recovery-after-workout, consistency, tags |
training.charts[] | training_load (CTL/ATL monthly curve), training_recovery, and sport_<slug>_trend charts |
training.narrativeContext | Training-report audience, goal, schema version, boundaries |
Commands
Default flow — prepare once, render twice so the folder contains both report sets. Pass --with-cross-link to both render calls so