excel-parser_skill

Overview

This skill provides smart Excel/CSV parsing that inspects file structure and routes processing to the optimal strategy. It balances speed and accuracy by using a lightweight metadata scout to choose between a high-speed Pandas mode and a semantic HTML mode for complex layouts.

How this skill works

The skill first performs a fast metadata scan (merged cells, row counts, empty interruptions) without loading full data. It computes a complexity score and recommends either Pandas Standard Mode for simple or large tables, or HTML Semantic Mode for small, irregular reports. The chosen path then executes a tailored extraction workflow to maximize correctness and performance.

When to use it

• Processing Excel/CSV files with unknown or varying structure
• Handling files that may contain merged cells, multi-level headers, or irregular layouts
• Optimizing between throughput (large tables) and extraction accuracy (complex reports)
• Automating pipelines where some sheets are simple tables and others are human-formatted reports
• Avoid when file structure is already documented and fixed

Best practices

• Always run the scout metadata analysis before full processing to avoid wasted effort
• Respect the row-count rule: avoid HTML mode for sheets >1000 rows to prevent token overflow
• Default to Pandas mode for unknown files; it fails fast and is efficient for standard tables
• Cache analysis results per file or workbook when processing multiple sheets
• Never modify the original workbook during scanning or extraction

Example use cases

• Ingesting mixed-source reporting files where some sheets are simple exports and others are formatted summaries
• Batch-processing large CSV exports where speed is critical and structure is tabular
• Extracting hierarchical headers and data from small financial reports with merged cells
• Pre-processing spreadsheets for downstream ML or BI pipelines that require clean tabular inputs
• Splitting complex sheets into semantic segments for accurate downstream NLU