Data Analysis

Overview

Analyzes user-provided Excel (.xlsx/.xls) or CSV files using DuckDB — an in-process analytical SQL engine. Supports schema inspection, SQL querying, statistical summaries, and result export.

> Poirot note: The original deer-flow skill uses a bundled > scripts/analyze.py helper. Poirot doesn't bundle that script, so this > version uses bash with python3 + duckdb directly. Install duckdb first: > pip install duckdb.

When to Use

• User uploads Excel/CSV files and wants analysis
• User wants statistics, summaries, pivot tables, or SQL queries on data
• User wants to filter, join, or aggregate structured data

Prerequisites

bash
# Install duckdb if not present
pip install duckdb openpyxl

Workflow

Step 1: Inspect File Structure

bash
python3 -c "
import duckdb
con = duckdb.connect()
# For CSV
result = con.execute(\"DESCRIBE SELECT * FROM read_csv_auto('data.csv')\").fetchall()
for col in result:
    print(f'{col[0]:30s} {col[1]}')

# For Excel (each sheet = a table)
result = con.execute(\"SELECT * FROM st_read('data.xlsx', layer='Sheet1') LIMIT 0\").fetchall()

# Row count
count = con.execute(\"SELECT COUNT(*) FROM read_csv_auto('data.csv')\").fetchone()[0]
print(f'Rows: {count}')
"

Step 2: Statistical Summary

bash
python3 -c "
import duckdb
con = duckdb.connect()
# Describe statistics
print(con.execute(\"SUMMARIZE SELECT * FROM read_csv_auto('data.csv')\").df().to_string())
"

Step 3: SQL Queries

bash
python3 -c "
import duckdb
con = duckdb.connect()

# Aggregation
result = con.execute('''
    SELECT category, COUNT(*) as count, AVG(price) as avg_price
    FROM read_csv_auto('data.csv')
    GROUP BY category
    ORDER BY count DESC
''').fetchall()
for row in result:
    print(row)

# Join two files
result = con.execute('''
    SELECT a.id, a.name, b.amount
    FROM read_csv_auto('orders.csv') a
    JOIN read_csv_auto('payments.csv') b ON a.id = b.order_id
''').fetchall()
"

Step 4: Export Results

bash
python3 -c "
import duckdb
con = duckdb.connect()
# Export to CSV
con.execute(\"COPY (SELECT * FROM read_csv_auto('data.csv') WHERE amount > 100) TO 'filtered.csv' (HEADER, DELIMITER ',')\")
# Export to JSON
con.execute(\"COPY (SELECT * FROM read_csv_auto('data.csv')) TO 'output.json' (FORMAT JSON)\")
"

Common Patterns

Pivot table

sql
SELECT
    product,
    SUM(CASE WHEN month = 'Jan' THEN amount ELSE 0 END) AS jan,
    SUM(CASE WHEN month = 'Feb' THEN amount ELSE 0 END) AS feb,
    SUM(CASE WHEN month = 'Mar' THEN amount ELSE 0 END) AS mar
FROM read_csv_auto('sales.csv')
GROUP BY product

Percentiles

sql
SELECT
    percentile_cont(0.5) WITHIN GROUP (ORDER BY price) AS median,
    percentile_cont(0.95) WITHIN GROUP (ORDER BY price) AS p95
FROM read_csv_auto('data.csv')

Multi-sheet Excel

bash
python3 -c "
import duckdb
con = duckdb.connect()
# List sheets
sheets = con.execute(\"SELECT table_name FROM st_geometry_tables()\").fetchall()
# Query specific sheet
result = con.execute(\"SELECT * FROM st_read('data.xlsx', layer='Sheet2') LIMIT 10\").fetchall()
"

Pitfalls

• DuckDB not installed: pip install duckdb openpyxl first
• Large files: DuckDB handles large files well, but SUMMARIZE on very

large datasets may be slow. Sample first: SELECT * FROM ... TABLESAMPLE 10%

• Encoding: CSV with non-UTF-8 encoding may fail. Specify encoding in

read_csv_auto options.

• Date parsing: DuckDB auto-detects dates, but ambiguous formats may need

explicit strptime parsing.

• Excel formulas: st_read reads cell values, not formula results. Use

openpyxl directly if you need computed values.