- Home
- Skills
- Data & Databases
- Spreadsheet Data Auditor
Works with the AI Agents you already use
Spreadsheet Data Auditor
Point it at a spreadsheet and get a data-quality report plus an annotated copy with every problem cell highlighted and commented, checked against your own rules
$5
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
Spreadsheet Data Auditor
Example session with this skill installed
Audit orders.csv for data-quality problems. order_id should be unique, status must be one of pending/shipped/delivered/cancelled, quantity should be 1-20, and no order date can be after 2026-10-20. Paste the report into your reply too.
- Read your context and instructions
- Compiled the spreadsheet data auditor
- Generated the data export
I audited orders.csv (12 orders, 7 columns) with your four rules plus the default checks. The tally is
5 high, 2 medium and 6 low finding groups, and the data-quality score is
35/100. Every one of your rules turned up at least one violation.
Breaks your rules (high severity)
Duplicate order_id: A-1003 appears twice, at rows 4 and 8. The two rows match on everything except the date: row 4 says 2026-13-05, which can't be a real date (there is no month 13), and row 8 says 2026-03-05. So row 4 is probably a mistyped copy of row 8, not a separate order.
Order dates: row 11 (A-1009) is dated 2027-01-01, which is after your 2026-10-20 cutoff. Row 4's 2026-13-05 can't be read as a date. That one bad cell is flagged twice, by type_mismatch and by invalid_dates, so it accounts for two of the high findings.
- Quantity outside 1–20: row 6 (
A-1005) has-2and row 9 (A-1007) has40.
Status not in the allowed list: row 6 has shiped, a misspelling of shipped. Row 3's Shipped passes, because the allowed-values check ignores case. It is flagged as a casing problem instead.
Worth fixing (medium and low)
- Row 5 (
A-1004) has no customer email. - Row 7's price is stored as the text
"$1,299.00", and row 9's country is" de", with a leading space and in lowercase. - The 5 price outliers are mostly normal price variation (9.50, 249.00, 1,299.00). The one to check is row 12,
A-1010, with a unit price of0.
Not checked, but visible in the data: row 9's date is 07/03/2026, while every other row uses YYYY-MM-DD. It passed because it falls inside the window whether it means 7 March or 3 July, but someone has to decide which it is. Row 3's email lee.ortiz@example has no domain ending (like .com). I didn't run an email-format check because you didn't ask for one.
Files
outputs/audit_report.md: the full report, pasted below.outputs/orders_audited.xlsx: a copy of your data with every flagged cell coloured by severity (red for high, amber for medium, grey for low). Each flagged cell has a comment naming the check it failed, and anAudit Summarysheet comes first. Your original orders.csv is unchanged.
Data Audit — orders.csv
Rows: 12 · Columns: 7 · Sheets: 1 · Distinct issue types: 11
Summary
- 🔴 High: 5 🟡 Medium: 2 ⚪ Low: 6
- Overall data-quality score: 35/100
- Top issues:
- 🔴
order_date— 2 unparseable or out-of-window dates. - 🔴
quantity— 2 values outside [1, 20]. - 🔴
order_date— 1 cells aren't valid dates in a mostly-date column. - 🔴
status— 1 values outside the allow-list ['pending', 'shipped', 'delivered', 'cancelled']. - 🔴
order_id— 1 repeated values in a column marked unique.
- 🔴
Findings by column
customer_email (text, 8.3% missing)
- missing_values [medium] — 1 affected. 1 empty cells (8.3%). Example rows: 5
order_date (date, 0.0% missing)
- type_mismatch [high] — 1 affected. 1 cells aren't valid dates in a mostly-date column. Example rows: 4
- invalid_dates [high] — 2 affected. 2 unparseable or out-of-window dates. Example rows: 4, 11
status (text, 0.0% missing)
allowed_values [high] — 1 affected. 1 values outside the allow-list ['pending', 'shipped', 'delivered', 'cancelled']. Example rows: 6
inconsistent_categories [medium] — 5 affected. Near-duplicate labels, e.g. 'shipped' vs 'shiped'. Example rows: 2, 3, 5, 6, 13
- casing [low] — 4 affected. Same category in mixed casing: {'shipped', 'Shipped'}. Example rows: 2, 5, 13, 3
quantity (number, 0.0% missing)
- out_of_range [high] — 2 affected. 2 values outside [1, 20]. Example rows: 6, 9
- outliers [low] — 1 affected. 1 statistical outliers (outside -2.00–6.00). Example rows: 9
unit_price (number, 0.0% missing)
- number_format [low] — 1 affected. 1 numbers carry currency/thousands symbols. Example rows: 7
- outliers [low] — 5 affected. 5 statistical outliers (outside 13.97–30.02). Example rows: 3, 4, 7, 8, 12
country (text, 0.0% missing)
- whitespace [low] — 1 affected. 1 cells with stray leading/trailing/double spaces. Example rows: 9
- casing [low] — 4 affected. Same category in mixed casing: {'de', 'DE'}. Example rows: 9, 2, 5, 10
order_id (text, 0.0% missing)
- duplicate_keys [high] — 1 affected. 1 repeated values in a column marked unique. Example rows: 8
Reproduce / reconfigure
Config used for this audit
{
"disable": [],
"params": {
"missing_values": {
"max_missing_pct": 5
},
"type_mismatch": {
"min_dominant_pct": 90
},
"outliers": {
"method": "iqr",
"iqr_k": 1.5
},
"inconsistent_categories": {
"similarity": 0.9
}
},
"columns": {
"order_id": {
"unique": true
},
"status": {
"allowed_values": [
"pending",
"shipped",
"delivered",
"cancelled"
]
},
"quantity": {
"out_of_range": {
"min": 1,
"max": 20
}
},
"order_date": {
"invalid_dates": {
"before": "2026-10-20"
}
}
},
"foreign_keys": []
}
Score = 100 − min(100, 100 × Σ(weight × affected cells) / total cells), weights: high 5, medium 2, low 0.5.
Row numbers are spreadsheet rows: the header is row 1, so the first order (A-1001) is row 2.
Do you want me to tighten anything else? I could add an email-format rule on customer_email, an allowed list of country codes, or turn off the price outlier check, which is mostly noise here.
orders.csv
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Audits a spreadsheet (.xlsx, .xlsm, .csv, .tsv) for data-quality problems and hands back two things: a readable audit report and an annotated copy of the workbook, with problem cells coloured by severity and commented with the check they failed. It runs fourteen checks out of the box: missing values, duplicate rows and keys, type mismatches, numbers stored as text, out-of-range values, invalid dates, outliers, near-duplicate categories, stray whitespace, mixed casing, regex rules, allow-lists and broken references between columns.
What it does
Say what the data should look like in plain words ("order_id is unique", "status is one of pending, shipped, delivered", "quantity 1 to 20", "no dates after today") and it turns that into rules. With no rules it runs the defaults, so "audit this" works on its own. The chat answer leads with the worst problems by row number, then the report lists findings by column with example rows, a data-quality score, and the exact config used, so you can rerun it.
Try it first
Give it a CSV or workbook and one or two rules. You should get a short summary, audit_report.md, and <name>_audited.xlsx with an Audit Summary sheet in front and the flagged cells highlighted.
What it needs
Bash and Python 3 with pandas and openpyxl (pip install pandas openpyxl). Three bundled scripts do the work: audit, report and annotate. No network access and no API keys.
It audits one file at a time; merging or deduplicating across several files is a different job. Outlier and near-duplicate checks are statistical, so some of what they flag is normal variation, and the report says which findings to look at first. Dates are parsed from common formats, so an ambiguous date like 07/03/2026 passes if it fits the window either way; it is mentioned, not decided. It doesn't edit your data: the annotated copy is a separate file and the original is untouched. Very large sheets are slower, because every cell is profiled.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 28 days ago
Requires