---

name: unified-sql description: Query databases and explore data. Use when the user mentions database, query, SQL, table, schema, data exploration, "check the data", "look at the database", "what's in the table", "show me records", "find in database", PostgreSQL, MySQL, SQLite, cross-database joins, data validation, or export query results. Also triggers on: exploring schemas, sampling data, running analytics queries, checking data quality, or any task involving database operations.

Unified SQL

Query and analyze data across PostgreSQL, MySQL, and SQLite databases using DuckDB as a unified query engine.

---

⚠️ MANDATORY FIRST STEP: List Available Databases

YOU MUST ALWAYS RUN THIS FIRST before attempting ANY database operation.

DO NOT skip this step. DO NOT assume database names. DO NOT run queries until you see the list of available databases.

# ALWAYS run this first - find and list configured databases
CREDS=".claude/data-analyze/credentials.json"
if [ -f "./$CREDS" ]; then
  CREDS_FILE="./$CREDS"
  echo "Using project credentials: $CREDS_FILE"
elif [ -f "$HOME/$CREDS" ]; then
  CREDS_FILE="$HOME/$CREDS"
  echo "Using user credentials: $CREDS_FILE"
else
  echo "❌ No credentials file found!"
  echo "Create credentials at: ./.claude/data-analyze/credentials.json (project) or ~/.claude/data-analyze/credentials.json (user)"
  exit 1
fi

echo "Available databases:"
jq -r '.databases[] | "  - \(.name) (\(.type))"' "$CREDS_FILE"

ONLY PROCEED with queries after seeing the list of available databases above.

---

Quick Start

Schema Exploration

# List all tables
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/schema_explorer.py --name prod_db --list-tables

# Describe a specific table
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/schema_explorer.py --name prod_db --describe users

# Sample data from a table
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/schema_explorer.py --name prod_db --sample orders --limit 10

Simple Queries

python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --name prod_db \
  --query "SELECT * FROM prod_db.users WHERE created_at >= '2024-01-01' LIMIT 10"

Cross-Database Queries

python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --names prod_db,sales_db \
  --query "SELECT u.email, o.order_id FROM prod_db.users u JOIN sales_db.orders o ON u.id = o.user_id"

Core Capabilities

1. Database Schema Exploration

Examine table structures, columns, and data types without writing SQL.

Available operations:

• --list-tables: Show all tables in database
• --describe TABLE: Show column names, types, nullability
• --sample TABLE: Preview rows from table
• --stats TABLE: Show row counts and column statistics

Example workflow:

# Step 1: List tables
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/schema_explorer.py --name prod_db --list-tables

# Step 2: Examine specific table
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/schema_explorer.py --name prod_db --describe users

# Step 3: Sample data
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/schema_explorer.py --name prod_db --sample users --limit 5

2. Cross-Database Joins

Join tables across different database systems in a single query.

Pattern:

-- Databases are referenced by their credential names
-- For example, if you have "users_db" and "orders_db" in credentials.json

SELECT
    u.column,
    o.column
FROM users_db.table1 u
JOIN orders_db.table2 o ON u.id = o.foreign_id

Example:

python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --names users_db,orders_db \
  --query "
    SELECT
      u.email,
      COUNT(o.order_id) as total_orders
    FROM users_db.users u
    LEFT JOIN orders_db.orders o ON u.id = o.user_id
    GROUP BY u.email
    ORDER BY total_orders DESC
  "

3. Query Analysis

Analyze query performance and execution plans.

Using EXPLAIN:

python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --name prod_db \
  --query "EXPLAIN SELECT * FROM prod_db.large_table WHERE created_at >= '2024-01-01'"

4. Data Export

Export query results in multiple formats.

Supported formats: table, json, csv, markdown

# Export to JSON
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --name prod_db \
  --query "SELECT * FROM prod_db.users" \
  --format json > output.json

# Export to CSV
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --name prod_db \
  --query "SELECT * FROM prod_db.analytics" \
  --format csv > output.csv

Credential Management

Credential File Structure

Credentials are stored in .claude/data-analyze/credentials.json and searched in order:

1. Project: ./.claude/data-analyze/credentials.json
2. User: ~/.claude/data-analyze/credentials.json

See ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/credentials.example.json for the template.

Key points:

• Each database needs a unique name (this is what you reference in queries)
• Supported types: postgres, mysql, sqlite
• PostgreSQL/MySQL: requires host, port, database, user, password
• SQLite: requires path to database file

Show available databases:

# Find and use credentials file (Project → User)
CREDS=".claude/data-analyze/credentials.json"
if [ -f "./$CREDS" ]; then
  CREDS_FILE="./$CREDS"
elif [ -f "$HOME/$CREDS" ]; then
  CREDS_FILE="$HOME/$CREDS"
else
  echo "No credentials file found" >&2
  exit 1
fi

jq -r '.databases[].name' "$CREDS_FILE"
jq -r '.databases[] | "\(.name): \(.type)"' "$CREDS_FILE"

Security Best Practices

1. Never commit credentials - .claude/ is typically gitignored
2. Use read-only accounts - Grant minimal permissions for analytics queries
3. Credential file locations:
   • Project: ./.claude/data-analyze/credentials.json
   • User: ~/.claude/data-analyze/credentials.json
4. File permissions - Restrict access: chmod 600 ~/.claude/data-analyze/credentials.json
5. Reference by name only - Scripts read credentials automatically, just use --name

Security Notes

SQL Injection Protection:

• Table and schema names are validated (alphanumeric and underscores only)
• Credential names are validated before use
• By default, write operations (DROP, DELETE, INSERT, UPDATE, etc.) are blocked
• Use --allow-writes flag only when write operations are explicitly needed

Read-Only Mode (Default):

The federated query tool runs in read-only mode by default. Dangerous SQL operations are blocked:

# This will be blocked by default
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --name prod_db \
  --query "DELETE FROM prod_db.users WHERE id = 1"
# Error: Dangerous SQL operation 'DELETE' detected. Use --allow-writes to enable.

# To allow write operations (use with caution)
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --name prod_db \
  --query "DELETE FROM prod_db.users WHERE id = 1" \
  --allow-writes

Error Message Sanitization:

• Passwords and connection details are automatically redacted from error messages
• Sensitive information will not leak through error output

Usage Patterns

Single database:

# Scripts find credentials automatically (Project → User)
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/schema_explorer.py --name kolverse --list-tables

Multiple databases (federated query):

# Reference multiple databases by name (comma-separated)
python ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/scripts/federated_query.py \
  --names kolverse,analytics_db \
  --query "SELECT * FROM kolverse.users u JOIN analytics_db.metrics m ON u.id = m.user_id"

Available Extensions

DuckDB supports many database extensions:

Pre-configured in scripts:

• postgres - PostgreSQL databases
• mysql - MySQL/MariaDB databases
• sqlite - SQLite file databases

Other useful extensions:

• httpfs - Query remote files (S3, HTTP)
• parquet - Parquet file support
• json - JSON/NDJSON support
• icu - Advanced string operations

See references/extensions.md for detailed extension documentation and usage examples.

Troubleshooting

Extension Not Found

# Install missing extension
con.execute("INSTALL postgres")
con.execute("LOAD postgres")

Query Timeout

• Add LIMIT clause to large queries
• Use WHERE filters to reduce data scanned
• Consider materializing intermediate results

Memory Issues

• Process data in batches
• Use streaming results: con.execute(query).fetch_df_chunk()
• Increase DuckDB memory limit: con.execute("SET memory_limit='4GB'")

Resources

Scripts

• scripts/federated_query.py - Main federated query tool (supports --name, --names)
• scripts/schema_explorer.py - Database schema exploration tool (supports --name)
• scripts/credential_manager.py - Credential loading and validation

Credentials

• .claude/data-analyze/credentials.json - Your database credentials (Project or User scope)
• ${CLAUDE_PLUGIN_ROOT}/skills/unified-sql/credentials.example.json - Template for credentials file

References

• references/extensions.md - DuckDB extension documentation
• references/connection_examples.md - Connection string examples and patterns
• references/query_patterns.md - Common query patterns and optimizations