Repository navigation
Add legacy XLS cell-level provenance using xlrd and the shared faithful grid renderer: coordinates, values/types, formats, merged references, dates, hidden content and available notes/links. Formula results are cached values only; XLS images/charts remain explicitly outside extraction coverage.
Initialize the OCR SDK only on the first real image request. Pure native text spreadsheets require no model/key even with OCR enabled. Missing credentials for real OCR become explicit file errors.
Add Agent ingestion documentation explaining Markdown/JSON roles, hash verification, merge semantics and business-audit boundaries. Validation: 34 tests with both supported OCR interfaces; real XLS sample matches 249 nonempty cells across 33 records with zero model requests.