data-dictionary

Data Dictionary: Automated Maintenance & Enhancement

Automate the update and maintenance of data dictionaries in Treasure Data - with 80-90% automated column descriptions that you can review, edit, and enhance. From schema extraction through TD write-back, this skill handles the heavy lifting while keeping you in control.

Usage

/data-dictionary [segment-name]

Example:

/data-dictionary customer_360

If segment name not provided, will list available Parent Segments.

Workflow Overview

This skill orchestrates 6 sequential stages:

1. Context Gathering - Collect business context (industry, company, use cases)
2. Schema Extraction - Extract schema from TD Parent Segment with sample data
3. Description Generation - Generate AI-powered column descriptions
4. Review Export - Export to CSV for human review
5. Validation - Validate edited CSV and import back to JSON
6. Write-back - Publish descriptions to Treasure Data

Each stage creates artifacts used by subsequent stages. Workflow supports resume from partial runs.

Stage 1: Context Gathering

Purpose: Collect business context to inform description generation.

Process: Ask 4 context questions ONE at a time (conversational, not all at once):

1. "What industry is this data from?" (e.g., e-commerce, SaaS, healthcare)
2. "Describe your company/business model briefly"
3. "What is this segment used for?" (default: customer segmentation)
4. "What are the primary use cases?" (e.g., marketing campaigns, analytics, personalization)

Output: Context stored in memory for Stage 3.

Progress:

✅ Stage 1 complete: Context gathered
➡️  Next: Stage 2 (Schema Extraction)

Stage 2: Schema Extraction

Purpose: Extract schema from TD Parent Segment with sample data for PII detection.

Command:

node src/index.js extract $SEGMENT --sample

Where $SEGMENT is the segment name from $ARGUMENTS[0].

Output:

✅ Stage 2 complete: Schema extracted
📁 Artifacts created:
   - ./schemas/{segment}.json
➡️  Next: Stage 3 (Description Generation)