Ingest

Ingest Skill

Process the inbox. Turn raw dumps into structured vault pages.

When to Run

• User says "ingest", "process inbox", "process the inbox"
• After dropping files into inbox/

Procedure

Step 1 — Scan Inbox

List all files in inbox/ (ignore inbox/_done/ and hidden files). If empty, report "Inbox clear" and stop.

Step 2 — Process Each File (sequentially, one at a time)

For each inbox item:

1. Read it — understand the content, language, structure
2. Classify it — what type of knowledge is this?
   • Task/action item → route to tasks/ with task frontmatter
   • Knowledge/concept/entity → route to wiki/ with appropriate type
   • Multiple topics in one note → split into separate files
3. Duplicate check — read wiki/index.md, look for existing pages
   • Match found → merge new info into existing page
   • No match → create new page using _templates/
4. Structure it — apply frontmatter, clean prose, add wikilinks
5. Preserve the original — move raw file to inbox/_done/ (never modify originals)

Step 3 — Update the Trio

After processing all items:

• wiki/index.md — add one-line entries for any new pages
• wiki/log.md — append ingest summary
• wiki/hot.md — refresh with current vault state

Step 4 — Report

INGEST REPORT
-------------
Items processed: X
New pages created: X
Existing pages updated: X
Tasks created: X
Originals archived to inbox/_done/: X

Rules

• Process one file at a time — no parallel, no batching across files
• Never modify the original inbox file — copy to _done/ as-is
• If a note has both tasks and knowledge, create both a task file and a wiki page
• Contradictions with existing pages: add a callout to BOTH pages, don't silently overwrite
• If input is in another language, write the wiki page in the vault's language but preserve key terms