CASE-UCO-SDK: Skill for Cursor

Image To CASE/UCO

Workflow

1. Identify the image artifact:

   • Record file path, file name, extension, MIME type, byte size, dimensions, and SHA256.
   • Model the source image as ObservableObject with FileFacet and ContentDataFacet.
   • Use ContentDataFacet(hash=[Hash(hash_method=["SHA256"], hash_value=...)]).

2. Extract text:

   • Prefer a real OCR engine when available (tesseract, cloud OCR, or an application OCR service).
   • If OCR is unavailable and the image is visible in the prompt, use vision-assisted transcription and mark that method in the graph.
   • Preserve raw OCR text with ExtractedStringsFacet(strings=[ExtractedString(...)]) on the source image observable.

3. Analyze the content:

   • Parse domain facts from OCR text, such as receipt totals, timestamps, merchant names, addresses, line items, payment methods, identifiers, or account fragments.
   • Keep uncertain values in annotations or descriptions; do not overstate them as verified facts.
   • Redact or preserve masked values as they appear. Never infer full payment card numbers, personal identifiers, or secrets from partial values.

4. Map to ontologies with the CASE/UCO MCP:

   • Always read the MCP tool schema before calling a tool.
   • Use search_classes, find_classes_for_domain, guide_mapping, and get_class_details to select classes.
   • Use existing UCO/CASE/CAC classes first. If no adequate class exists, model source evidence plus extracted attributes using Event, Annotation, Dictionary, or a documented local extension rather than inventing unvalidated terms.

5. Serialize:

   • Use CASEGraph and generated classes where possible.
   • Use UUID-ending IRIs so case_validate does not emit identifier info findings.
   • For receipt-like images, a good baseline is:
     • Source image: ObservableObject + FileFacet + ContentDataFacet + ExtractedStringsFacet
     • Merchant: Organization
     • Address: Location + SimpleAddressFacet
     • Payment card fragment: PaymentCard plus a conservative description or AccountFacet(account_identifier=masked_value)
     • Transaction facts: Event with event_context references and event_attribute=[ProperDictionary(...)]
     • Processing provenance: InvestigativeAction with instrument=[Tool(...)], object=[source_image], and result=[derived_event]