data-exploration_skill

Overview

This skill profiles and explores datasets to reveal their shape, quality, and patterns before analysis. It provides a concrete, repeatable methodology for structural discovery, column-level profiling, relationship detection, and data quality scoring. Use it to reduce surprises and make informed decisions about cleaning, modeling, and analysis.

How this skill works

The skill inspects table-level metadata (row/column counts, grain, keys, update cadence) and classifies columns by role (identifier, dimension, metric, temporal, text, boolean, structural). It computes per-column statistics (nulls, cardinality, top values, numeric percentiles, string lengths, date ranges) and derives relationship signals (foreign key candidates, correlations, redundancies). Finally, it applies quality checks (completeness, consistency, accuracy, timeliness) and produces documentation templates and recommended queries for schema discovery.

When to use it

• On first encounter with a new dataset or table
• Before building models or reports to assess fitness for purpose
• When investigating unexplained results or suspected data issues
• To prioritize cleaning and imputation work based on completeness and accuracy
• When tracing lineage and dependencies for impact analysis

Best practices

• Start with table-level questions: grain, primary key, row counts, last update
• Classify every column by role to guide downstream analysis and aggregation
• Compute null rates and cardinality before running heavy transformations
• Flag high-impact quality issues: business-rule violations, placeholder values, type inconsistencies
• Document schema, known issues, and common query patterns for team reuse

Example use cases

• Profiling an event log to find timestamp gaps, session outliers, and hot event types
• Assessing a customer table for completeness of contact info and suspicious default values
• Comparing revenue and order metrics across segments to spot skewed distributions or outliers
• Identifying foreign-key relationships and redundant columns before designing joins
• Generating a dataset summary and schema doc to onboard analysts and data engineers