Image Table Data Extractor

Problem

Many test records arrive as images: screenshots of tables with headers, cells, and status labels such as pass or 通过. Copying them into spreadsheet tools can require manual realignment, while generic OCR often flattens the table into plain text and loses column structure. This skill targets that workflow: turning tabular content in images into editable, inspectable data.

How It Works

It prioritizes multimodal understanding over generic text recognition. The model first interprets visible text and table structure in the image. If that model cannot recognize the content, it switches to another multimodal image model. The documentation explicitly discourages plain OCR because OCR may fail to preserve table formatting. After extraction, it applies narrow filtering: only cells explicitly marked pass or 通过 are removed. Empty cells, headers, and all other visible data are kept unchanged. The result is returned in two matching formats: a Markdown table for direct viewing and CSV for import. The CSV output follows standard rules for comma separation, quoted fields, and escaped double quotes.

Boundaries

This is useful when a test image already contains a visible table and the goal is to extract that table. It does not infer missing cells, add unobserved data, or create headers that are not present. The output preserves the original visible content and structure, making it suitable for validation before downstream analysis.