[healthsync] Add natural language query support for health data

I've been tracking with Apple Watch since 2023. last year I noticed my VO2 max had peaked around March 2024 and then quietly declined. I was still working out consistently. the Health app confirmed the chart and offered nothing else. I wanted to just ask. not write a correlation query by hand. not export to a spreadsheet. just ask. so I built healthsync. it parses your Apple Health export into local SQLite. then: ```bash healthsync skills install ``` that command embeds the full DB schema, every table name, date formats, and example SQL into Claude Code's skills directory. one command. after that, Claude already knows your entire database layout. you open Claude Code and ask in plain English. I asked: "why did my VO2 max drop last year?" Claude pulled the VO2 max records, pulled the workout records, ran the correlation, and told me: my average workout duration had dropped from 39 minutes to 22 minutes right around the same time VO2 max started falling. not fewer workouts — just shorter sessions. a gradual drift that's invisible in the moment. I wouldn't have written that query on my own. I didn't even know what to look for. **still 100% local** your health data stays on your machine. Claude Code runs locally. the skill is a text file in `~/.claude/skills/healthsync/`. **a few other things I found asking questions** sleep averages look fine in aggregate. filtered for nights after late evening meetings (I work with a US team, calls run late), REM was consistently lower. the average completely hides that pattern. I found it by asking "does my sleep quality change after late meetings?" for step counts and active energy, Watch and iPhone both write records for the same time intervals. the raw export includes both. `--total` applies source-priority dedup (Watch > iPhone > other) before aggregating. my raw step totals were 1.83x inflated before dedup. blood pressure records in the raw export come as separate systolic and diastolic entries. not paired. healthsync stages them by source and timestamp, emits a paired row only when both arrive. if you've tried to analyze a BP trend from the raw XML you were working with fragmented data. **if you'd rather just query directly** ```bash healthsync query vo2-max --format table healthsync query workouts --format table healthsync query sleep --format json healthsync query steps --total ``` the database is plain SQLite at `~/.healthsync/healthsync.db`. open with `sqlite3` or any DB tool. tool: https://github.com/BRO3886/healthsync (MIT, local, no cloud) docs: https://healthsync.sidv.dev what questions have you wanted to ask your health data but couldn't?