← Library · AI-OS cookbook · 08
A private health space in your second brain, plus an AI that explains findings in plain language, preps every doctor visit and tracks what actually helps — grounded only in your own notes. This is organisation and preparation — not a substitute for professional medical advice. Decisions belong with your doctor.
Lab reports buried in email attachments, symptoms half-remembered, medication details in your head, every doctor visit improvised in the waiting room. No record of what actually helped — each episode starts from zero.
One private folder. One living note per topic with a symptom log, open questions for the doctor and explicit red flags. An AI that reads all of it, preps every visit — and never pretends to be the doctor.
A _health/ folder in your notes vault. Small on purpose: a hub plus a handful of living notes. The structure is what makes it AI-readable.
One entry point. It states the working principles and links every open topic. The four principles that run the real one: cause before symptom (chase the why, not just relief) · open questions per appointment (no visit without a written list) · track effect (log what you tried and whether it worked) · technique matters (how you take or apply something is where treatments quietly fail).
Every active topic gets one note, dated in the title, that grows over time instead of spawning fragments. The skeleton that has proven itself:
One note for what you currently take: preparation, dose, technique notes (the part everyone gets wrong), and open medication questions for the next visit. The sidekick checks new suggestions against it before saying anything.
Every note carries the same small set of keys: title, type, status, created, updated, tags — and source on notes that came from a document. The type values all share a health-* prefix; that prefix does double duty in the privacy section below.
Documents are where health data goes to die. The fix: every report becomes a markdown note with the same frontmatter, so the AI can read your history the way it reads any other file.
Lab PDF, visit summary, wearable export — hand it to the sidekick inside your vault. It extracts the structure, not just the text.
Frontmatter with type: health-lab-style typing, source and dates — then the panel as a section, marker by marker, plus a results section that waits to be filled when values arrive. A lab note can exist before the blood draw: the panel you agreed on, the additions you requested, the empty table.
Because every lab note has the same shape, "compare this against last year" is reading two files — not hunting through attachments. The same pattern extends to wearable exports: a periodic export saved as markdown with the same keys keeps sleep, training and recovery greppable next to the rest.
Done when: no health document exists only as a PDFThe sidekick never free-styles from general knowledge alone. Every answer is grounded in the space — your history, your log, your open questions.
| You ask | It reads | You get |
|---|---|---|
| "Explain this in plain language" | The lab or topic note | A calm, jargon-free interpretation — and context that takes the scare out of scary words |
| "Prep my appointment" | Topic note + open questions + log | A one-page brief: history, what changed since last visit, the questions to ask — ordered by importance |
| "What should I focus on this week?" | Symptom & effect log + plan | A movement / recovery / technique focus drawn from what measurably helped — not generic advice |
| "What's still open?" | All active topics | Follow-ups: results to enter, questions still unanswered, re-checks due |
| "Is this all connected?" | Timelines across notes | One connected picture instead of a pile of separate complaints — episodes lined up as a sequence |
"You are my health sidekick. Ground every answer only in my _health/ notes — never invent values or history. Explain in plain language, prep my visits, suggest evidence-based self-measures, track follow-ups. End every health answer with: not a substitute for professional medical advice — decisions belong with my doctor. If anything I describe matches a red-flag section, tell me to see a doctor. Do not coach me past a red flag."
Date, symptom, what you did, what happened. Ten seconds per entry, into the topic note's log. This is the raw material everything else runs on.
It reads the new entries, updates the dated re-assessment in each active note (the real notes carry these as dated sections — the picture refined after observation), and rebuilds the open-questions list. Improving topics get archived off the hub; stalling ones get flagged.
History one-pager plus the question list, generated from the note. You walk in prepared instead of reconstructing three months from memory in a waiting room.
What the doctor said, what changed in the therapy, which results to enter when they arrive, when to re-check. The loop closes; the note stays the single source of truth.
Done when: a doctor visit starts from a written brief, not from memoryThis system organises, explains and prepares. It is not a substitute for professional medical advice, diagnosis or treatment — decisions belong with your doctor. The sidekick's job is to make you the best-prepared patient in the practice, never to replace the person across the desk.
Every health answer ends with the line above — baked into the sidekick's instruction, not added when someone remembers. The convention in the source system: no health output without it.
Every topic note carries an explicit red flags → see a doctor section. When something matches, the sidekick's only move is "book the appointment" — it does not coach past the line.
The open-questions list is the product. The sidekick sharpens what to ask and keeps the history straight; the answering happens in the consultation room.
Evidence-based, conservative, reversible — movement, recovery, technique. Anything beyond that belongs on the open-questions list, not in a chat reply.
Health data is the most personal data you have. This setup treats privacy as architecture, not as a setting.
The space lives in your private notes vault — local files you own. Never in shared docs, team spaces, or anything another person might open.
The type: health-* frontmatter prefix — and the deliberate absence of the flags your export or enrichment pipelines key on — means every automated job that syncs, exports or summarises your vault skips this folder by design.
Always-on assistants that serve other contexts get a pointer at most ("a health space exists, read it only in private sessions"). The sidekick reads the folder live, in a session that belongs to you alone.
Folder layout, note skeletons, frontmatter keys, the loop — all of it can be documented and handed on, as this page proves. Markers, doses, diagnoses and names never leave the vault.
Copy the bootstrap promptThe button below puts it on your clipboard.
Paste it into Claude CodeOpened in your notes vault, on your own machine.
Answer its questionsIt scaffolds the space, ingests your first lab PDF, and installs the standing rules — disclaimer and red-flag routing included.
Runs in your vault, on your machine. Nothing leaves.