categorize-inbox-batch

Categorize Inbox Batch

End-to-end triage workflow for the Paperless inbox: per document, set correspondent + document type + tags, then remove the Eingang tag.

When to Use

Trigger when the user wants to process inbox documents — common phrasings: "Eingang durchgehen", "Eingang aufräumen", "Dokumente kategorisieren", "triage". Typically follows a check-inbox summary.

How to Use

For each document the user wants to process:

1. Identify the correspondent — if obvious from the title, use find_correspondent_by_name with name__icontains to find the existing ID. If no match, ask the user before creating a new correspondent (anti-duplicate discipline — see Open Loops).
2. Identify the document type — same pattern with find_document_type_by_name.
3. Tags — ask the user which tags apply (or suggest based on prior similar documents in the same correspondent).
4. Apply — preferred sequence on a small batch (1-5 docs):
   • bulk_set_correspondent with [doc_id] + correspondent_id
   • bulk_set_document_type with [doc_id] + document_type_id
   • bulk_add_tag for each tag (or bulk_modify_tags if multiple tags + removals at once)
   • Last: bulk_remove_tag with the Eingang-tag ID to release from inbox
5. After each document, confirm what was applied. Don't batch all documents into one giant operation without per-doc confirmation — gives the user a place to redirect.

Anti-Patterns

• Bulk-Sync vermeiden — kleine Schritte (see KI-OS feedback memory). Don't process 20+ documents in one operation without intermediate cleanup-stops.
• Vor jedem Neu-Anlegen suchen. Before creating a new correspondent or tag, use the find_*_by_name tool with name__icontains to check if one already exists. Server-side filter, not client-side grep (this caught the "Stadtwerke" duplicate, 2026-05-22).
• PUT, not PATCH — but this MCP wraps bulk_edit which handles the method-choice automatically. Mentioned only as a CLAUDE.md cross-reference for context.

Failure Modes

• bulk_set_correspondent with an invalid correspondent_id returns a 400; defensive: validate ID exists via find_correspondent_by_name first.
• find_*_by_name with no match returns an empty list — handle this branch explicitly (don't pass None as ID to a bulk_set call).