Food Label Compliance Review and Risk Cases

Problem

Food label design often has subtle violations: insufficient character height, incorrect ingredient order, missing allergen emphasis, or incomplete SC numbers. Manual review against standards is slow and can miss layout, color, logo, and typography issues. This skill reviews food label images or PDFs and returns findings, fixes, penalty references, and case examples.

How it works

• Input: label files such as PNG, JPG, or PDF; supports single-page or multi-page text extraction and visual inspection.
• Dual extraction: with multiple models, OCR extracts text and a multimodal model checks fonts, layout, graphics, and colors; with one model, the multimodal model handles both.
• 58 checkpoints: covers product name, net quantity, ingredients, nutrition facts, production information, special foods, special claims, allergens, transgenic labels, and related requirements.
• Output: default Markdown table; optional interactive HTML report rendered with jinja2, including filtering by compliance status.
• Case lookup: searches risk categories for common errors, standard clauses, and enforcement references.

Boundaries

It supports review and reference work; it does not replace final regulatory determinations. Ambiguous scans, small text, front/back labels, and low-quality images can affect extraction. Pair OCR results with visual checks. When standards such as GB 7718 or GB 28050 change, use the latest official text as the authority.