# Clean up a messy spreadsheet

> NoInfra brief · Agent: OpenClaw

## Objective
Turn a chaotic sheet (duplicates, broken formatting, junk) into clean, usable data.

## Ask me first (don't assume)
- What does "clean" look like — the target columns/schema?
- Which column is the unique key/ID?
- Duplicate rule: keep first, keep latest, or merge?
- Any columns to ignore or values that are valid-but-weird?

## Integrations (install if needed)
- Google Sheets or file upload (CSV/XLSX) — to read/write the data. Ask me to connect Drive if the file is there.

## Steps
1. Profile the sheet: row/column count, data types, and a list of issues.
2. Propose a target schema and the dedupe rule; wait for my confirmation.
3. Normalize columns, types, dates, and casing.
4. Dedupe by the key per the agreed rule.
5. Flag rows you can't safely fix instead of deleting them.
6. Output the cleaned file + a change log.

## Deliverables
- Cleaned sheet, change log, list of rows needing my review.

## Done when
- Output matches the schema with no silent data loss.

## Rules
- Never delete rows silently — log every change. Test on the first 20 rows before bulk edits.