CSV Data Doctor

CSV exports are rarely clean. Dates come in five formats, prices carry currency symbols, the same city is spelled three ways, and duplicates hide behind stray spaces. Fixing it by hand is slow, and asking an AI to "just clean it" can change things you didn't want changed.

This skill gives your AI a careful procedure. A profiler script reports, per column, the inferred type, blanks and null-like values, distinct counts, min and max, top values, mixed date formats and day/month order, stray spaces, case variants, formatted numbers and IQR outliers. For the whole file it reports rows with the wrong number of fields, empty rows, exact and likely duplicates, and repeated IDs. The AI explains the findings in plain words, asks you about the things only you can decide, and proposes a cleaning plan. Nothing changes until you say yes.

A second script applies the agreed plan: trim, null tokens, ISO dates, plain numbers, consistent spelling, value mapping, fill or drop blanks, dedupe, fix row length and rename columns. It has a dry run, checks the whole plan first, refuses to overwrite any file, and logs every change with examples.

Both scripts are Node.js with no dependencies and an RFC 4180 CSV parser that handles quoted commas and line breaks. Examples include a messy orders file, its profile, the agreed plan, the log and the cleaned result. UTF-8 files up to 200 MB. Outliers and unclear values are flagged for you, never changed.