inbox-to-mindmap

inbox-to-mindmap

A folder is a staging area, not a filing cabinet. The user drops documents into ~/second_brain/inbox/; this skill reads them, puts the information into the graph, and archives the file to ~/second_brain/raw/. The documents themselves never become nodes — each one survives only as a file:// provenance URL (to its archived location) on the claims it produced.

That distinction is the whole point. A node called "Q3 architecture review deck" is a filename with extra steps: to use it you have to re-open the deck. A node called "Hybrid BM25 + dense retrieval beat dense-only by 14 points on internal eval" is knowledge — it answers a question on its own, links to other ideas, and gains confidence when a second document says the same thing. Nodes shaped like documents make a graph that mirrors a directory; nodes shaped like claims make a graph that thinks.

The binding ingest protocol is AGENTS.md. The rules that matter most here are repeated below.

Layout

Thing	Default	Override
Inbox folder	~/second_brain/inbox/	$MINDGAP_INBOX, or a path the user names
Archive	~/second_brain/raw/	—
Ledger	~/.mindgap/inbox-ledger.json	--ledger

The inbox is a drop zone, not a home: once a file's claims have landed (or it is marked skipped), it MOVES to the archive — ~/second_brain/raw/YYYY/<topic>/ (S9.5). Files are never renamed beyond collision handling and never deleted; the archive is immutable ground truth. The ledger records sha256 → what was ingested, and dedup is by content hash, so moves are safe: a second run over the emptied inbox does nothing, and an edited re-drop returns as changed and gets re-read. If the inbox does not exist, say so and offer to create it rather than inventing a different path.

Pipeline

S1 — Scan (recursive)

python3 <skill>/scripts/scan_inbox.py [DIR] > /tmp/inbox.json

Walks the whole tree, at any depth. Every pending file comes back with status (new, changed, done, unsupported, skipped-large), its rel path, its folder, and the folder's segments. The same records are also grouped under folders, each with its context_files listed separately.