Use Case: Data Analysis Workflow Design

Introduction

Data analysis work is a complex process involving multiple steps: collection, transformation, analysis, and report generation. With Claude Code Workflow Studio, you can visually design such complex workflows and execute them repeatedly in Claude Code.

Challenges

Typical challenges data analysts face:

• Repeating the same steps: Weekly analysis of project data, etc.
• Multiple options: Statistical analysis, visualization, machine learning - changing processes based on objectives
• Error-prone: Skipping steps or reading wrong files in manual operations
• Undocumented: Analysis procedures not shared across the team

Solution: Visual Workflow Design

Design analysis workflows in Claude Code Workflow Studio to achieve:

• Visual design: Grasp the entire flow at a glance
• Reusable: Execute repeatedly once designed
• Team sharing: Easily share with JSON files
• Flexible branching: Switch processes based on objectives

Workflow Design Example

Overall Structure

A data analysis workflow consists of the following steps:

[Start]
   ↓
[Data Collector] ← Collect data files
   ↓
[Data Validator] ← Validate data
   ↓
[Choose Analysis Type] ← User selects
   ↓
[Statistical Analysis] or [Data Visualization]
   ↓
[Report Generator] ← Generate report from results
   ↓
[End]

Step 1: Data Collector

Node Configuration

Node Type: Sub-Agent

Settings:

• Name: Data Collector
• Prompt:

Find and load the following files from the project's data directory:
- metrics.csv: Development metrics
- logs.json: Application logs
- performance.csv: Performance data

Check the content of each file and report the data format and count.

• Tools: Read, Glob
• Model: Sonnet

Design Points

• Tools: Only allow file search (Glob) and reading (Read)
• Clear instructions: Specify target files concretely
• Validation: Check data format and count to prevent errors in next steps

Step 2: Data Validator

Node Configuration

Node Type: Sub-Agent

Settings:

• Name: Data Validator
• Prompt:

Validate the collected data:
1. Confirm required columns exist
2. Calculate missing value percentages
3. Verify data types are as expected
4. Detect outliers and anomalies

If problems exist, report details and auto-correct minor issues.

• Tools: Read
• Model: Sonnet

Design Points

• Error detection: Check beforehand to prevent issues in subsequent analysis
• Auto-correction: Automatically handle minor issues (type conversion, etc.)
• Clear reporting: Specifically report problem areas

Step 3: Choose Analysis Type

Node Configuration

Node Type: AskUserQuestion

Settings:

• Question: Which analysis do you want to run?
• Header: Analysis Type
• Options:
  • Option 1:
    • Label: Statistical Analysis
    • Description: Calculate statistical metrics like mean, median, correlation coefficients
  • Option 2:
    • Label: Data Visualization
    • Description: Generate time series graphs and histograms
  • Option 3:
    • Label: Run Both
    • Description: Execute both statistical analysis and visualization

Design Points

• Clear choices: Explain each option’s content in detail
• Flexibility: Switch processing based on user’s objective
• Multi-select support: multiSelect: false (exclusive selection)

Step 4: Statistical Analyzer

Node Configuration

Node Type: Sub-Agent

Settings:

• Name: Statistical Analyzer
• Prompt:

Execute statistical analysis on validated data:

1. Basic statistics
   - Mean, median, standard deviation
   - Minimum, maximum, quartiles

2. Correlation analysis
   - Correlation coefficients between variables
   - Identify strong correlations

3. Trend analysis
   - Trends in time series data
   - Detect periodicity and seasonality

4. Anomaly detection
   - Identify outliers
   - Detect abnormal patterns

Summarize results in Markdown format.

• Tools: Read
• Model: Opus (for complex analysis)

Design Points

• Opus model: Use Opus for advanced analysis
• Systematic analysis: Specify analysis items step by step
• Structured output: Readable in Markdown format

Step 5: Data Visualizer

Node Configuration

Node Type: Sub-Agent

Settings:

• Name: Data Visualizer
• Prompt:

Visualize the validated data:

1. Time series graphs
   - Each metric's trend
   - Period: Past 30 days

2. Histograms
   - Distribution of each variable

3. Scatter plots
   - Relationships between variables

4. Heat maps
   - Visualize correlation matrices

Express using Mermaid or text-based graphs.

• Tools: Read
• Model: Sonnet

Design Points

• Text-based: Use text-representable diagrams like Mermaid
• Multiple perspectives: Analyze from multiple angles with different graph types
• Clear period specification: Explicitly state time series graph ranges

Step 6: Report Generator

Node Configuration

Node Type: Sub-Agent

Settings:

• Name: Report Generator
• Prompt:

Integrate analysis results and create a final report:

## Data Analysis Report
### Execution Date: {{timestamp}}

### 1. Data Overview
- Summary of collected data
- Data quality assessment

### 2. Analysis Results
- Statistical analysis results (if executed)
- Visualization results (if executed)

### 3. Key Findings
- Important insights (3-5 items)
- Anomalies or trends to watch

### 4. Recommended Actions
- Specific recommendations based on analysis results

Save the report in the reports/ directory.

• Tools: Read, Write
• Model: Sonnet

Design Points

• Write permission: To save report files
• Structured: Easy-to-read report composition
• Actionable: Include recommendations

Node Connections

Basic Flow

Data Collector → Data Validator → Choose Analysis Type

Branching Logic

When “Statistical Analysis” is selected:

Choose Analysis Type → Statistical Analyzer → Report Generator

When “Data Visualization” is selected:

Choose Analysis Type → Data Visualizer → Report Generator

When “Run Both” is selected: Execute both nodes, then proceed to Report Generator (controlled by Branch node)

Save and Export

Save

1. Enter workflow name: data-analysis-pipeline
2. Click Save button
3. Saved to .vscode/workflows/data-analysis-pipeline.json

Export

1. Click Export button
2. Generated files:
   • .claude/agents/Data_Collector.md
   • .claude/agents/Data_Validator.md
   • .claude/agents/Statistical_Analyzer.md
   • .claude/agents/Data_Visualizer.md
   • .claude/agents/Report_Generator.md
   • .claude/commands/data-analysis-pipeline.md

How to Run

Execute the following command in Claude Code:

/data-analysis-pipeline

The workflow starts and each step executes in sequence.

Best Practices

1. Error Handling Design

Consider handling when data is not found:

• Specify in Data Collector “report to user if files not found”
• Add Branch node for file existence check

2. Model Selection

• Haiku: Simple data collection and validation
• Sonnet: Standard analysis and report generation
• Opus: Complex statistical analysis and prediction

3. Clear Prompts

Good example:

Read all CSV files in the data directory,
and report the number of rows and column names for each file.

Bad example:

Read data and analyze it.

4. Minimize Tool Permissions

Grant only the minimum necessary tools to each node:

• Data Collector: Read, Glob only
• Report Generator: Read, Write only

Application Examples

Weekly Report Auto-Generation

Add “scheduled execution” to the designed workflow to enable automatic report generation every week (combined with Claude Code features).

A/B Test Analysis

Replace “Choose Analysis Type” with “Choose Test Group” node and design group-specific analysis flows.

Anomaly Detection Alerts

Add “Alert Checker” node after Report Generator to send notifications when specific thresholds are exceeded.

Summary

Claude Code Workflow Studio lets you visually design complex data analysis workflows.

Key Takeaways:

• ✅ Divide into steps progressively
• ✅ Flexible branching with user selection
• ✅ Set clear prompts for each node
• ✅ Minimize tool permissions to necessary only

Next, see another use case in Code Review Workflow Design.