document-ingest

name	document-ingest
description	Upload, vision-parse, and persist a document with multi-level summaries.
inputs	[{"name":"file_path","type":"string","required":true,"description":"Local or remote path to the source file."},{"name":"filename","type":"string","required":true,"description":"Human-readable filename stored in the Document record."},{"name":"collection_id","type":"string","required":false,"description":"Optional folder/collection to assign the document to on creation."}]
outputs	[{"name":"document_id","type":"string","description":"UUID of the newly created Document row."},{"name":"page_count","type":"integer","description":"Number of pages parsed from the document."}]
tools	["vision-parse","db-query"]

document-ingest

When to use

Pick this skill when the user uploads a file or references a path that has not yet been imported into the system. Use it before any classify, summarize, or extract operation because those skills assume the document already exists in the store.

How it works

• Convert the source file to images via vision-parse (DOCX → PDF → per-page images).
• Send page images in batches to the vision model; receive a structured JSON outline.
• Write one Document row, one Chapter row per chapter, one Section per section, and one Page per page to the DB via db-query.
• Chunk page text and write Chunk rows; optionally embed and store vectors.
• Compute top-down Document.summary, Chapter.summary, and Section.summary.

Audit

Emits one ExecutionHistory row with event_type=DOCUMENT_IMPORTED and actor_type=agent after the Document row is committed.