CSV Cleaner by langbot-team

# CSV Cleaner

Profile and clean tabular data (CSV/TSV/Excel) with pandas.

## What it does

- Profiles a file: row/col counts, dtypes, null counts, duplicates, examples.
- Cleans explicitly: trim whitespace, dedupe, snake_case headers, drop empty
  columns, coerce column types.
- Always writes to a new file; never mutates the original.

## Files

| File | Purpose |
|------|---------|
| `SKILL.md` | Instructions the agent receives on activation |
| `scripts/clean.py` | Profile and clean a CSV/TSV/Excel file |
| `references/recipes.md` | Common cleaning recipes |

## Requirements

Installs `pandas` (and `openpyxl` for Excel) on first use.

## License

Apache-2.0.


---
name: csv-cleaner
display_name: CSV Cleaner
description: "Profile and clean tabular data (CSV/TSV/Excel): inspect columns, types, nulls, and duplicates, then trim whitespace, fix types, dedupe, and standardize headers. Use when the user has a messy spreadsheet or CSV and wants it profiled or cleaned before analysis. Do NOT use for charting (use a plotting skill) or for writing complex documents."
license: Apache-2.0
---

# CSV Cleaner

Profile a tabular file and apply safe, explicit cleaning steps with pandas.

## When to use

The user has a CSV/TSV/Excel file that is messy — inconsistent headers, stray
whitespace, wrong types, nulls, or duplicate rows — and wants it profiled or
cleaned.

## Execution steps

1. **Profile first**: `python scripts/clean.py profile data.csv`. It reports row/col
   counts, per-column dtype, null counts, duplicate-row count, and example values.
   Always profile before changing anything.
2. **Decide the cleaning steps** with the user based on the profile. See
   `references/recipes.md` for common fixes.
3. **Clean**: `python scripts/clean.py clean data.csv -o clean.csv` with flags:
   - `--strip` trim whitespace in string cells
   - `--dedupe` drop exact duplicate rows
   - `--snake-headers` normalize headers to snake_case
   - `--drop-empty-cols` drop all-null columns
   - `--coerce col:type` coerce a column (`int`, `float`, `datetime`, `str`)
4. **Report** what changed (rows before/after, columns affected) and the output path.

## Rules

- Profile before cleaning; never mutate blindly.
- Every transformation must be explicit and reported — no silent dropping.
- Preserve the original file; always write to a new `-o` output.

## Available resources

- `scripts/clean.py` — profile and clean a CSV/TSV/Excel file (installs pandas).
- `references/recipes.md` — common cleaning recipes and when to apply them.