generate-from-collection

name	generate-from-collection
description	Generate a single Document (Word/PowerPoint) from all source documents in a collection, with attributions back to each source
model	deepseek/deepseek-v4-pro
tier	pro
inputs	[{"name":"collection_id","type":"str","required":false,"description":"Prisma Collection.id — pulls all member documents"},{"name":"source_document_ids","type":"list[str]","required":false,"description":"Explicit document_ids (alternative to collection_id)"},{"name":"brief","type":"str","required":true},{"name":"format","type":"str","required":false,"description":"docx | pptx (default: docx)"},{"name":"style","type":"str","required":false},{"name":"audience","type":"str","required":false},{"name":"tone","type":"str","required":false},{"name":"slide_count","type":"int","required":false,"description":"Only used when format=pptx"}]
outputs	[{"name":"file_path","type":"str"},{"name":"file_url","type":"str"},{"name":"title","type":"str"},{"name":"model","type":"str"},{"name":"format","type":"str"},{"name":"source_document_count","type":"int"},{"name":"source_document_ids","type":"list[str]"}]

generate-from-collection

The "wrap a folder" entry point for OfficePlane content generation. Picks up every document that belongs to a Collection (or an explicit list of document IDs) and synthesises them into one combined Word or PowerPoint output with per-node attributions pointing back to whichever source each piece of content was drawn from.

Why this skill exists

generate-docx and generate-pptx both require the caller to supply source_document_ids manually. When users have already organised their source material into a Collection (a folder-like grouping created by the collection-manage skill or the ECM UI), they should be able to say "generate a deck from this collection" without enumerating every member ID. generate-from-collection wraps that lookup and then does the same LLM-grounded generation, with all attributions intact.

How attributions are emitted

Each major section in the output document includes one or more attribution objects of the form:

{
  "node_id": "<generated-node-id>",
  "document_id": "<source-uuid>",
  "section_id": "<source-section-uuid>",
  "page_numbers": [1, 2]
}

A single section may attribute to multiple sources if it synthesises content from more than one input document. The attributions are persisted to the Derivation table by persist_derivations_from_document so they can be retrieved via the lineage API (GET /api/lineage/workspace/{id}).

Inputs

Name	Required	Notes
collection_id	one of these	Prisma Collection.id
source_document_ids	one of these	Explicit list of Document IDs
brief	yes	What the combined document should cover
format	no	docx (default) or pptx
style	no	e.g. professional, clinical, casual
audience	no	Target reader
tone	no	e.g. neutral, authoritative, warm
slide_count	no	Target slides when format=pptx (default 10)

Example invocation

curl -X POST http://localhost:8001/api/jobs/invoke/generate-from-collection \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": {
      "collection_id": "col-abc123",
      "brief": "Executive overview of our AI compliance training materials",
      "format": "pptx",
      "slide_count": 8,
      "audience": "C-suite",
      "tone": "authoritative"
    }
  }'

Or with explicit IDs instead of a collection:

curl -X POST http://localhost:8001/api/jobs/invoke/generate-from-collection \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": {
      "source_document_ids": ["d0a5322e-e48b-4fcc-974c-cc246fcac65b", "..."],
      "brief": "Consolidated training guide",
      "format": "docx"
    }
  }'

Output

{
  "skill": "generate-from-collection",
  "output": {
    "file_path": "/data/workspaces/<job_id>/output.pptx",
    "file_url": "/data/workspaces/<job_id>/output.pptx",
    "title": "AI Compliance Training Overview",
    "model": "deepseek/deepseek-v4-flash",
    "format": "pptx",
    "source_document_count": 4,
    "source_document_ids": ["uuid1", "uuid2", "uuid3", "uuid4"],
    "slide_count": 8
  }
}

View source trail

After generation the lineage endpoint shows which nodes came from which sources:

GET /api/lineage/workspace/<job_id>