CSV Data Cleaner Skill

What this skill does

Every CSV export from a real system has the same problems — inconsistent date formats, stray whitespace, duplicate rows from a retried sync, phone numbers in four different formats — and cleaning it by hand in a spreadsheet is slow and easy to get subtly wrong. This skill takes a messy CSV and fixes the common issues automatically, then reports exactly what it changed so nothing is silently altered without a record.

How it works

1. Issue detection — scans every column for inconsistent formatting (mixed date formats, inconsistent casing, stray whitespace, mixed phone/currency formats), exact and near-duplicate rows, and empty-vs-null inconsistencies.
2. Normalization — standardizes each column to a single consistent format (choosing the most common existing format as the target unless you specify one), trims whitespace, and removes exact duplicate rows.
3. Change report — outputs the cleaned CSV alongside a plain-English change log: how many rows were deduplicated, which columns were reformatted and how, and any rows flagged as ambiguous for manual review rather than auto-corrected.

Requirements

• Works on Claude AI (Claude Code, or claude.ai with code execution enabled) — cleaning runs via code execution (Python/pandas) so it can handle large files without loading everything into the model's context at once.

Setup

1. Unzip the package into your Skills directory, keeping the folder structure intact.
2. Attach or point it at your CSV file, specifying any target formats you want enforced (e.g. force all dates to YYYY-MM-DD).
3. Review the change report before treating the cleaned file as final — ambiguous rows are flagged, not silently guessed.

What you get

• A cleaned CSV with inconsistent dates, casing, whitespace, and formats standardized automatically
• Duplicate and near-duplicate rows removed, with counts reported
• A plain-English change log of exactly what was altered, so nothing changes silently
• Ambiguous rows flagged for your manual review rather than auto-corrected
• A worked example cleaning a sample messy export