How do I install the Data Cleaning skill?

Run npx skillmds@latest add haifai-ai/data-cleaning in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Data Cleaning safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Data Cleaning?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Data Cleaning free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Data Cleaning?

Haifai-AI (@haifai-ai) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Data Cleaning by haifai-ai — Agent Skill

Name: Data Cleaning
Author: Haifai-AI

Data Cleaning

Clean, normalize, and reconcile messy spreadsheets or delimited files — duplicates, inconsistent labels, mixed types, broken dates — then hand back a tidy workbook plus an audit trail of every change. Use when data looks messy before analysis.

Haifai-AI 55cdfbd 2.4 KB Updated Sep 17, 2026

File contents

Data cleaning

Cleaning is a mapping from "untrustworthy" to "auditable". Never silently fix: every transformation must be visible in the delivered file.

Workflow

Profile before touching anything. xlsx_read / csv_read the source. Enumerate problems explicitly (column by column): blanks, duplicates, near-duplicate labels ("N. Region" vs "North"), numbers stored as text, impossible values (negative quantities, future dates), ragged rows.
Propose the rule set in one short message when the fixes are judgment calls ("merge 'US'/'USA'/'United States' into US?" — keep going with your best rule; flag it in the log sheet).
Produce one clean workbook with a pandas + openpyxl script run via $DSH_OFFICE_PYTHON:
- Clean — tidy table: one header row, typed columns (number_formats), no merged cells inside the data area, total_row only when meaningful.
- Audit — every change applied: original value → new value → rule name → affected row references. This sheet is what makes the work trustworthy. Style per excel-analysis (banner, header fill, number formats, banded rows).
Verify: re-read the cleaned file; assert row counts, spot-check five transformed cells against the audit entries, confirm totals match source sums where applicable.

Standard rules

Trim whitespace; collapse internal double spaces; unify casing of categorical labels (Title Case for names, UPPER for codes).
Dates → ISO yyyy-mm-dd column format; store real dates, never strings.
Numbers stay numbers: strip currency symbols/percent signs into the number_formats layer instead of baking them into strings.
Duplicates: define the key first (whole-row vs business key), keep first, record dropped keys in Audit.
Missing values: prefer explicit placeholders the analyst can filter ("(unknown)"), documented in Audit; never invent zeros for measures.

Handoff

Report: rows in → rows kept/dropped, distinct rules applied (count each), and where the audit lives. Offer to continue straight into analysis (excel-analysis) on the cleaned file.

---name: data-cleaningdescription: Clean, normalize, and reconcile messy spreadsheets or delimited files — duplicates, inconsistent labels, mixed types, broken dates — then hand back a tidy workbook plus an audit trail of every change. Use when data looks messy before analysis.---# Data cleaningCleaning is a mapping from "untrustworthy" to "auditable". Never silentlyfix: every transformation must be visible in the delivered file.## Workflow1. **Profile before touching anything.** `xlsx_read` / `csv_read` the source. Enumerate problems explicitly (column by column): blanks, duplicates, near-duplicate labels ("N. Region" vs "North"), numbers stored as text, impossible values (negative quantities, future dates), ragged rows.2. **Propose the rule set** in one short message when the fixes are judgment calls ("merge 'US'/'USA'/'United States' into US?" — keep going with your best rule; flag it in the log sheet).3. **Produce one clean workbook** with a pandas + openpyxl script run via `$DSH_OFFICE_PYTHON`: - `Clean` — tidy table: one header row, typed columns (`number_formats`), no merged cells inside the data area, `total_row` only when meaningful. - `Audit` — every change applied: original value → new value → rule name → affected row references. This sheet is what makes the work trustworthy. Style per excel-analysis (banner, header fill, number formats, banded rows).4. **Verify**: re-read the cleaned file; assert row counts, spot-check five transformed cells against the audit entries, confirm totals match source sums where applicable.## Standard rules- Trim whitespace; collapse internal double spaces; unify casing of categorical labels (Title Case for names, UPPER for codes).- Dates → ISO `yyyy-mm-dd` column format; store real dates, never strings.- Numbers stay numbers: strip currency symbols/percent signs into the `number_formats` layer instead of baking them into strings.- Duplicates: define the key first (whole-row vs business key), keep first, record dropped keys in Audit.- Missing values: prefer explicit placeholders the analyst can filter ("(unknown)"), documented in Audit; never invent zeros for measures.## HandoffReport: rows in → rows kept/dropped, distinct rules applied (count each),and where the audit lives. Offer to continue straight into analysis(`excel-analysis`) on the cleaned file.

Frequently asked questions