Write Query

/write-query - Write Optimized SQL

If you see unfamiliar placeholders or need to check which tools are connected, see CONNECTORS.md (opens in a new tab).

Write a SQL query from a natural language description, optimized for your specific SQL dialect and following best practices.

Usage

/write-query <description of what data you need>

Workflow

1. Understand the Request

Parse the user’s description to identify:

• Output columns: What fields should the result include?
• Filters: What conditions limit the data (time ranges, segments, statuses)?
• Aggregations: Are there GROUP BY operations, counts, sums, averages?
• Joins: Does this require combining multiple tables?
• Ordering: How should results be sorted?
• Limits: Is there a top-N or sample requirement?

2. Determine SQL Dialect

If the user’s SQL dialect is not already known, ask which they use:

• PostgreSQL (including Aurora, RDS, Supabase, Neon)
• Snowflake
• BigQuery (Google Cloud)
• Redshift (Amazon)
• Databricks SQL
• MySQL (including Aurora MySQL, PlanetScale)
• SQL Server (Microsoft)
• DuckDB
• SQLite
• Other (ask for specifics)

Remember the dialect for future queries in the same session.

3. Discover Schema (If Warehouse Connected)

If a data warehouse MCP server is connected:

1. Search for relevant tables based on the user’s description
2. Inspect column names, types, and relationships
3. Check for partitioning or clustering keys that affect performance
4. Look for pre-built views or materialized views that might simplify the query

4. Write the Query

Follow these best practices:

Structure:

• Use CTEs (WITH clauses) for readability when queries have multiple logical steps
• One CTE per logical transformation or data source
• Name CTEs descriptively (e.g., daily_signups, active_users, revenue_by_product)

Performance:

• Never use SELECT * in production queries – specify only needed columns
• Filter early (push WHERE clauses as close to the base tables as possible)
• Use partition filters when available (especially date partitions)
• Prefer EXISTS over IN for subqueries with large result sets
• Use appropriate JOIN types (don’t use LEFT JOIN when INNER JOIN is correct)
• Avoid correlated subqueries when a JOIN or window function works
• Be mindful of exploding joins (many-to-many)

Readability:

• Add comments explaining the “why” for non-obvious logic
• Use consistent indentation and formatting
• Alias tables with meaningful short names (not just a, b, c)
• Put each major clause on its own line

Dialect-specific optimizations:

• Apply dialect-specific syntax and functions (see sql-queries skill for details)
• Use dialect-appropriate date functions, string functions, and window syntax
• Note any dialect-specific performance features (e.g., Snowflake clustering, BigQuery partitioning)

5. Present the Query

Provide:

1. The complete query in a SQL code block with syntax highlighting
2. Brief explanation of what each CTE or section does
3. Performance notes if relevant (expected cost, partition usage, potential bottlenecks)
4. Modification suggestions – how to adjust for common variations (different time range, different granularity, additional filters)

6. Offer to Execute

If a data warehouse is connected, offer to run the query and analyze the results. If the user wants to run it themselves, the query is ready to copy-paste.

Examples

Simple aggregation:

/write-query Count of orders by status for the last 30 days

Complex analysis:

/write-query Cohort retention analysis -- group users by their signup month, then show what percentage are still active (had at least one event) at 1, 3, 6, and 12 months after signup

Performance-critical:

/write-query We have a 500M row events table partitioned by date. Find the top 100 users by event count in the last 7 days with their most recent event type.

Tips

• Mention your SQL dialect upfront to get the right syntax immediately
• If you know the table names, include them – otherwise Claude will help you find them
• Specify if you need the query to be idempotent (safe to re-run) or one-time
• For recurring queries, mention if it should be parameterized for date ranges