data-dictionary

data-dictionary

A Claude Code skill that generates a comprehensive markdown data dictionary from any database — combining schema definitions with live SQL statistics, null rates, sample rows, and a Mermaid ER diagram.

What It Does

• Auto-detects schema source: Prisma, SQL DDL, Django models, SQLAlchemy, or live DB introspection (no schema file needed)
• Multi-database: SQLite, PostgreSQL, MySQL, DuckDB
• Rich statistics per table: row counts, null rates, numeric min/max/mean, categorical distributions, date ranges
• Sample rows with automatic PII column redaction (email, birth_date, token, etc.)
• Mermaid erDiagram rendered natively on GitHub
• Quick Reference summary table at the top for fast navigation
• Update mode (--update): refreshes stats while preserving hand-written descriptions, detects new/removed tables and columns

Install

# Clone the claude-skills repo and symlink this skill
git clone https://github.com/biomystery/claude-skills ~/.claude/skills-repo
ln -s ~/.claude/skills-repo/data-dictionary ~/.claude/skills/data-dictionary

Or symlink just this skill:

git clone https://github.com/biomystery/claude-skills /tmp/claude-skills
ln -s /tmp/claude-skills/data-dictionary ~/.claude/skills/data-dictionary

Claude Code will auto-detect it on next launch.

Usage

Generate (first time)

/data-dictionary

Claude will ask two quick questions (glossary terms, column exclusions), then auto-detect your schema and database, query live statistics, and write docs/data_dictionary.md.

Custom output path or language

/data-dictionary --output reports/schema_ref.md
/data-dictionary --lang zh
/data-dictionary --output docs/db.md --lang es

Update existing dictionary

/data-dictionary --update

Refreshes all statistics and detects schema changes, while preserving any descriptions you've hand-edited. Prints a structured diff summary:

✅ Data dictionary updated: docs/data_dictionary.md

Schema changes detected:
  🆕 New tables (1):    audit_log
  📝 Changed tables (1): users (+1 col)

Statistics refreshed:
  • 18 tables — row counts updated
  • 94 columns — null rates refreshed

Output

A single markdown file (default: docs/data_dictionary.md) containing:

• Header with DB engine, schema source, and git commit hash
• Quick Reference table (all tables with row counts)
• Per-table sections: columns, statistics, sample data
• Enumerations section
• Mermaid ER diagram
• Data Quality Notes (flags empty tables, high null rates, type mismatches)
• Glossary (if provided)

Supported Schema Sources

Schema	Detection
Prisma	prisma/schema.prisma or **/*.prisma
SQL DDL	schema.sql or files with CREATE TABLE
Django	**/models.py
SQLAlchemy	Python files with Column(, relationship(
None	Falls back to PRAGMA table_info / information_schema

Requirements

• Claude Code (latest version)
• Database CLI accessible in PATH:
  • SQLite: sqlite3 (built into macOS)
  • PostgreSQL: psql
  • MySQL: mysql
  • DuckDB: duckdb

Skill Structure

data-dictionary/
├── SKILL.md    ← Claude Code skill definition
└── README.md