jarvis-import

Files (1)

• SKILL.md

SKILL.md (first lines)

# Bring existing files into Jarvis

## When to Use
- Onboarding step "Existing files".
- The user asks to import, bring in, or scan folders.

## Rules
- **Originals are never moved, renamed, edited or deleted.** Import means copy or reference.
- Only folders the user picked are opened. Nothing else on the Mac is read.
- The scan reads names, sizes and dates only. Say so when offering it.
- Finance and health folders are listed separately, unticked by default, and go only to
  `life/finance/` or `life/health/` (private, local) if the user explicitly chooses them.
- Code repos and photo/video folders are not imported by default; offer "reference only".
- Big imports cost tokens. Give a rough size up front (number of documents) and import in batches.

## Procedure
Client and company folders import into their **space** (`jarvis_space` create first): originals go in the
space's `raw/`, notes become pages inside it. Never mix two clients' files in one space.

1. **Scan.** Use a recent scan if one exists (`inbox/.import-scan.json` in the vault, from setup, less
   than 7 days old). Otherwise ask to scan, then run with the terminal tool:
   `python3 <Jarvis repo>/bin/scan_folders.py --out <vault>/inbox/.import-scan.json`
   (the Jarvis repo path: `jarvis_status`). macOS may ask for access to Desktop, Downloads or iCloud
   Drive; tell the user to allow it or skip that folder.
2. **Show what was found**, grouped, numbered, one line each: name, where, documents count, last
   modified. Groups: Work (client, project, work, other with documents), Private (finance, health),
   Skipped by default (code, media). Keep it to the 20 most relevant; offer "show the rest".
3. **Ask** with your question tool:
   - "Which should I bring in?" choices: `All work folders (recommended)`, `Let me pick by number`,
     `Include private ones too`, `Skip for now`. "Let me pick": ask for numbers (open-ended).
   - Then, one batch: "Copy documents into the vault, or just reference them where they are?"
     `Copy documents (recommended)`, `Reference only`, `Skip for now`.
4. **Map each chosen folder to a vault destination** using the work templates in onboarding
   `references/templates.md` (e.g. a client folder → `work/<company>/clients/<slug>/`, a project →
   `work/<company>/projects/<slug>/` or a life area). Show the mapping in one short list and confirm.
   Create a client/project page from `_templates/` if it doesn't exist.
5. **Record** in `me/onboarding.json` → `imports`: one entry per folder
   `{source, destination, mode: copy|reference, kind, documents, status: queued, done: 0}`.
6. **Import in batches** of up to 20 documents:
   - `copy`: copy supported documents (≤ 25 MB each) to `raw/imports/<slug>/…` keeping sub-paths,
     then run `brain-ingest` on them so the right pages get updated with sources.
   - `reference`: don't copy; read at most the 5 most useful-looking documents (briefs, contracts,
     proposals, READMEs) to write the destination page, and list the original folder path on it.
   - Update `done`/`status` after each batch; `log.md`: `## [date] import | <folder>: N documents`.
7. If more than one batch remains, ask whether to continue now or overnight. Overnight: create a
   cron job (02:30 daily) that loads this skill and continues queued imports, then removes itself
   when nothing is queued. Report progress in the morning briefing.
8. Finish with a one-line summary per folder and where it landed.

## Pitfalls
- Importing duplicates on a re-run: skip files already in `raw/imports/` with the same size and name.
- Treating file names as facts. Names only guide the mapping; facts come from reading documents.
- Copying a huge folder in one go. Always batch.

## Verification
Every `imports` entry is `done` or has a clear status; copied files exist under `raw/imports/`; the
destination pages list their sources; originals are untouched.

From JohannsenLum/jarvis, MIT.

More from JohannsenLum/jarvis