yprincedec / my-skill

name: data-quality-csv-auditor
description: Audits CSV datasets for missing values, duplicate rows, inconsistent data types, invalid values, and potential outliers. Use when a data analyst needs a reliable CSV quality review, a transparent cleaning plan, or reproducible validation checks before analysis.
license: MIT
metadata:
llmskillhub:
version: 0.1.0
categories: [data-analysis]
capabilities:
network: false
filesystem: read
shell: false
secrets: []

Data Quality CSV Auditor

Purpose

Help data analysts inspect CSV datasets before analysis. Produce an evidence-based quality report and a safe cleaning plan.

When to Use

Use this skill when a user wants to:

• Find missing values or duplicate rows.
• Check column types and inconsistent formats.
• Identify invalid values and potential outliers.
• Prepare data for exploratory data analysis.
• Improve dataset quality before machine learning.

Workflow

1. Understand the Dataset

• Identify the dataset's purpose and intended analysis.
• Inspect column names, row count, and a small sample when available.
• Ask about ambiguous columns instead of guessing.
• Never claim to have inspected an inaccessible file.

2. Audit Data Quality

Check for:

• Missing values by column and percentage.
• Exact duplicate rows and possible duplicate identifiers.
• Inconsistent data types, whitespace, and category labels.
• Invalid values based on known rules or user-provided constraints.
• Potential outliers using appropriate statistical summaries.
• Suspicious dates and inconsistent units.

Distinguish confirmed errors from possible issues. An unusual value is not automatically wrong.

3. Recommend Cleaning Actions

For each issue, report:

• The detected problem.
• The affected column or records.
• Evidence supporting the finding.
• A recommended action and its reason.
• Any assumptions or risks.

Never delete rows, impute values, or overwrite original data without authorization. Preserve the original dataset and recommend working on a separate copy.

4. Generate a Quality Report

Include:

1. Dataset overview.
2. Missing-value summary.
3. Duplicate and consistency checks.
4. Invalid values and potential outliers.
5. Prioritized cleaning recommendations.
6. Limitations and items requiring human review.

Calculate counts and percentages only when the necessary data is available.

5. Validate Changes

After authorized cleaning:

• Compare original and cleaned row counts.
• Recheck missing values and duplicates.
• Verify relevant data types and constraints.
• Report changes and any remaining issues.

Never claim a test passed unless it was actually performed.

Safety and Reliability

• Do not invent statistics or test results.
• Avoid exposing personal or confidential data.
• Do not send dataset contents to external services.
• Explain assumptions and ask questions when needed.
• Prefer reversible and reproducible transformations.
• If tools are unavailable, provide instructions or code without claiming execution.

Expected Outcome

A clear data-quality report, a prioritized cleaning plan, and validation steps to help users prepare reliable datasets for analysis.