Csv Analysis

Instructions

When asked to analyze a CSV file, follow this workflow:

Step 1: Read and Profile

• Use the view tool to read the first 50 lines of the CSV
• Identify the delimiter (comma, tab, semicolon, pipe)
• Count total rows and columns
• Infer column data types (string, integer, float, date, boolean)

Step 2: Compute Statistics

Run a Python script to compute per-column statistics:

import csv
import statistics
from collections import Counter

# Read and analyze the CSV
# Compute: count, nulls, unique values, min, max, mean, median, std dev

Step 3: Quality Assessment

Check for:

• Missing or null values (empty strings, "NA", "null", "N/A")
• Duplicate rows
• Inconsistent formatting (mixed date formats, case inconsistency)
• Potential outliers (values beyond 3 standard deviations)

Step 4: Generate Report

Create a markdown report with:

• Overview: File name, row count, column count
• Schema table: Column name, type, non-null count, unique count
• Statistics table: Min, max, mean, median, std dev for numeric columns
• Quality issues: List of findings with severity (info/warning/error)
• Key findings: Top 3-5 insights from the data

Output Format

The report should be a well-formatted markdown document suitable for inclusion in project documentation. Use tables for structured data and bullet points for findings.

Error Handling

• If the file is not valid CSV, report the issue and suggest the correct format
• If the file is too large (>100MB), sample the first 10,000 rows and note the sampling
• If encoding errors occur, try UTF-8, Latin-1, and CP1252 in order

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: codebytes
• Source: codebytes/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.