data-analysis

Use this skill when asked to explore, analyse, or visualise any dataset, CSV, table, or collection of numbers.

Analysis Workflow

Always follow this sequence:

Inventory - How many rows? How many columns? What are the column names and types?
Quality check - Missing values, duplicates, outliers, type mismatches.
Descriptive stats - Mean, median, std, min, max, percentiles for numeric columns.
Distribution - Are numeric columns skewed? Are categorical columns imbalanced?
Relationships - Correlations, group comparisons, time trends if a date column exists.
Key findings - The 3-5 most actionable insights from the above.

Never skip steps. Never jump to "Key findings" without the groundwork.

Every insight must include a number. "Higher" without a figure is not an insight.
Flag data quality issues before drawing conclusions - dirty data produces wrong insights.
Distinguish correlation from causation explicitly when it matters.
If asked to plot or chart, use the scripts in scripts/ - do not write raw matplotlib from scratch.
If a column looks like a date but is typed as a string, note it and parse it before time-series analysis.

Pre-built analysis scripts live in scripts/. Read them before writing any analysis code.

Script	Purpose
`scripts/profile.py`	Full dataset profile: types, nulls, stats, top values
`scripts/correlations.py`	Pearson + Spearman correlation matrix with heatmap
`scripts/time_series.py`	Date-aware trend analysis, resampling, rolling averages

Usage: copy the relevant script, adapt the INPUT_FILE and column name variables at the top, then run.

Use this skill when asked to explore, analyse, or visualise any dataset, CSV, table, or collection of numbers.

Always follow this sequence:

Inventory - How many rows? How many columns? What are the column names and types?
Quality check - Missing values, duplicates, outliers, type mismatches.
Descriptive stats - Mean, median, std, min, max, percentiles for numeric columns.
Distribution - Are numeric columns skewed? Are categorical columns imbalanced?
Relationships - Correlations, group comparisons, time trends if a date column exists.
Key findings - The 3-5 most actionable insights from the above.

Never skip steps. Never jump to "Key findings" without the groundwork.

Every insight must include a number. "Higher" without a figure is not an insight.
Flag data quality issues before drawing conclusions - dirty data produces wrong insights.
Distinguish correlation from causation explicitly when it matters.
If asked to plot or chart, use the scripts in scripts/ - do not write raw matplotlib from scratch.
If a column looks like a date but is typed as a string, note it and parse it before time-series analysis.

Pre-built analysis scripts live in scripts/. Read them before writing any analysis code.

Script	Purpose
`scripts/profile.py`	Full dataset profile: types, nulls, stats, top values
`scripts/correlations.py`	Pearson + Spearman correlation matrix with heatmap
`scripts/time_series.py`	Date-aware trend analysis, resampling, rolling averages

Usage: copy the relevant script, adapt the INPUT_FILE and column name variables at the top, then run.