csv-analyzer（CSV 分析器）

CSV Analyzer

Overview

Comprehensive CSV data analysis and visualization engine. Run the script, then use this guide to interpret results and provide insights to users.

Quick Start

cd ~/.claude/skills/csv-analyzer/scripts
export $(grep -v '^#' /path/to/project/.env | xargs 2>/dev/null)
python3 analyze_csv.py /path/to/data.csv

Chart Selection Decision Tree

IMPORTANT: Choose charts based on what the user needs to understand:

What is the user trying to understand?
│
├── "What does my data look like?" (Overview)
│   └── Run with defaults → overview_dashboard.png
│
├── "Is my data clean?" (Quality)
│   └── Check: quality_score, missing_values, duplicates
│   └── Show: missing_values.png if problems exist
│
├── "What's the distribution?" (Single Variable)
│   ├── Numeric → numeric_distributions.png (histogram + KDE)
│   ├── Categorical → categorical_distributions.png (bar chart)
│   └── Time-based → time_series.png
│
├── "Are there outliers?" (Anomalies)
│   └── box_plots.png → points beyond whiskers are outliers
│
├── "How are variables related?" (Relationships)
│   ├── 2 numeric vars → correlation_heatmap.png
│   ├── 2-6 numeric vars → pairplot.png (scatter matrix)
│   ├── Numeric vs Categorical → violin_plot.png
│   └── All numeric → correlation_heatmap.png
│
└── "Can I predict X from Y?" (Predictive)
    └── correlation_heatmap.png → |r| > 0.5 suggests predictive power

How to Interpret Results (For Claude)

Quality Score Interpretation

Correlation Interpretation

Sign matters:

• Positive: "As X increases, Y tends to increase"
• Negative: "As X increases, Y tends to decrease"

Skewness Interpretation

Outlier Assessment

When reporting outliers:

• Few outliers (<1%): "A few extreme values that may warrant investigation"
• Moderate outliers (1-5%): "Notable outliers - check if they're errors or genuine extremes"
• Many outliers (>5%): "High outlier rate suggests either data issues or a non-normal distribution"

Insight Generation Framework

After running analysis, provide insights in this order:

1. Data Overview (Always)

"Your dataset has [rows] records and [cols] columns:
- [n] numeric columns: [list top 3]
- [n] categorical columns: [list top 3]
- Data quality score: [score]/100 ([grade])"

2. Key Findings (Pick most relevant)

If quality issues exist:

"I noticed some data quality concerns:
- [X]% missing values in [column] - [recommend: drop/impute/investigate]
- [N] duplicate rows detected - [recommend: keep first/remove all/investigate]"

If strong correlations found:

"Interesting relationships I found:
- [col1] and [col2] are strongly correlated (r=[value]) - [interpretation]
- This suggests [actionable insight]"

If outliers detected:

"I detected outliers in [columns]:
- [column]: [n] values beyond normal range ([min outlier] to [max outlier])
- These could be [data errors / genuine extremes / worth investigating]"

If skewed distributions:

"[Column] has a [right/left]-skewed distribution:
- Most values cluster around [median]
- But there are extreme values up to [max]
- For modeling, consider [log transform / robust methods]"

3. Recommendations (Based on findings)

Multi-Chart Dashboard Requests

When user asks for a "dashboard" or "comprehensive view":

# Generate all visualizations
python3 analyze_csv.py data.csv --format html --max-charts 10

Then present charts in this order:

1. overview_dashboard.png - "Here's your data at a glance"
2. correlation_heatmap.png - "Key relationships between variables"
3. numeric_distributions.png - "How your numeric data is distributed"
4. box_plots.png - "Outlier analysis"
5. categorical_distributions.png - "Category breakdowns" (if applicable)

Command Reference

Basic Analysis

python3 analyze_csv.py data.csv

Full Report with All Charts

python3 analyze_csv.py data.csv --format markdown --max-charts 10

Quick Analysis (No Charts)

python3 analyze_csv.py data.csv --no-charts

Large Files (>100MB)

python3 analyze_csv.py huge.csv --sample 50000

Specific Date Columns

python3 analyze_csv.py data.csv --date-columns created_at updated_at

JSON for Programmatic Use

python3 analyze_csv.py data.csv --format json --no-charts

Custom Output Location

python3 analyze_csv.py data.csv --output-dir /path/to/project/.tmp/analysis

Chart Descriptions (For Explaining to Users)

Common User Questions → Actions

Output Locations

Charts are saved to:

• Default: ~/.claude/skills/csv-analyzer/scripts/.tmp/csv_analysis/
• Custom: Use --output-dir /path/to/project/.tmp/analysis

Always copy charts to user's project.tmp for visibility:

cp ~/.claude/skills/csv-analyzer/scripts/.tmp/csv_analysis/*.png /path/to/project/.tmp/csv_analysis/

Cost

Free - runs entirely locally using pandas, matplotlib, seaborn, scipy.

Dependencies

pip install pandas matplotlib seaborn scipy numpy