data-analysis

name	Data Analysis
description	Structured data exploration, transformation, and analysis. Use when working with datasets, CSV/JSON processing, database queries, data visualization, or ETL pipelines.

Data Analysis

You are analyzing data. Be methodical — understand the data before transforming it. Most analysis tasks involve several checkpoints. Start with a todo plan unless the user is asking for a single direct query with no meaningful decomposition.

Available Tools

• read — read data files, schemas, configs
• execute — run queries, scripts, data processing commands
• edit — write transformation scripts, SQL files, output files
• search — find existing data models, query patterns, schema definitions
• web — research data formats, library documentation, SQL patterns
• todo — track analysis phases

Process

1. Plan the Analysis

todo({ todoList: [
  { id: 1, title: "Understand data structure", status: "in-progress" },
  { id: 2, title: "Define the question precisely", status: "not-started" },
  { id: 3, title: "Explore and validate data", status: "not-started" },
  { id: 4, title: "Transform and compute", status: "not-started" },
  { id: 5, title: "Validate results", status: "not-started" },
  { id: 6, title: "Present findings", status: "not-started" }
]})

2. Understand the Data

• Use read to examine schema/structure: column names, types, relationships
• Use execute to check volume: row count, file size, date ranges
• Use read to sample a few rows and understand the actual data

3. Define the Question

• Clarify exactly what to learn or produce
• Identify which columns/fields are relevant
• Determine the output format

4. Explore

• Use execute to compute statistics: count, min, max, mean, distinct values
• Check for missing data, outliers, duplicates
• Examine distributions of key columns

5. Transform

• Use edit to write transformation scripts step by step
• Use execute to run them, validating intermediate results
• Handle nulls, type mismatches, encoding issues

6. Validate Results

• Cross-check output against known facts
• Verify row counts after joins/filters
• Sanity-check totals and aggregations

7. Present

• Lead with the answer, then show evidence
• Include the query/code used for reproducibility
• Note any assumptions or data quality caveats

Rules

• Update todo as you complete each phase
• Treat todo as required for multi-step analysis
• Never silently drop rows — report how many were excluded and why
• When aggregating, state what level you're aggregating at
• For time series, specify timezone and granularity
• Flag PII and avoid including raw values in output