Sql Query Generation

SQL Query Generation

This skill enables an AI agent to translate natural language questions into correct, efficient SQL queries. The agent maps user intent to the appropriate query constructs — joins, aggregations, window functions, CTEs, and subqueries — while respecting the target database schema. It also analyzes query performance with EXPLAIN plans and recommends optimizations such as indexing, predicate pushdown, and query restructuring.

Workflow

1. Parse the natural language request. Extract the analytical intent: what metric is being asked for, which entities are involved, what filters apply, and how results should be ordered or grouped. Distinguish between requests for aggregated summaries versus row-level detail.

1. Map to the database schema. Identify the relevant tables and columns from the schema. Resolve ambiguous references (e.g., "sales" could mean the orders table or the revenue column). Determine the join path between tables using foreign key relationships, avoiding unnecessary joins that inflate result sets.

1. Select the appropriate query constructs. Choose between simple aggregation, window functions, CTEs, or subqueries based on complexity. Use CTEs for multi-step calculations to improve readability. Use window functions for running totals, rankings, and comparisons within partitions. Prefer explicit JOINs over implicit comma-separated joins.

1. Generate the SQL query. Write syntactically correct SQL with consistent formatting: uppercase keywords, lowercase identifiers, aliased tables, and indented clauses. Include comments for complex logic. Always specify column aliases for computed expressions.

1. Validate and optimize. Run EXPLAIN (or EXPLAIN ANALYZE) on the generated query to inspect the execution plan. Look for full table scans, hash joins on large tables, and sort operations on unindexed columns. Recommend indexes or query rewrites when the estimated cost is high.

1. Return results with explanation. Present the query alongside a plain-language explanation of what it does, the expected output format, and any assumptions made about the schema or data.

Supported Technologies

• PostgreSQL — CTEs, window functions, LATERAL joins, EXPLAIN ANALYZE
• MySQL — common table expressions (8.0+), window functions (8.0+), EXPLAIN FORMAT=JSON
• SQLite — lightweight queries, window functions (3.25+)
• SQL Server — T-SQL extensions, cross apply, query store analysis
• BigQuery — STRUCT/ARRAY types, QUALIFY clause, slot-based cost estimation

Usage

Provide the agent with a natural language question and the database schema (as DDL, a schema diagram, or a list of tables and columns). Optionally specify the target SQL dialect, any performance constraints, and whether you want the EXPLAIN analysis included.

Examples

Example 1: Complex analytics query from natural language

Request: "Show the top 5 customers by total spending in the last 12 months, along with their most frequently ordered product category and how their spending compares to the previous 12-month period."

WITH current_period AS (
    SELECT
        c.customer_id,
        c.customer_name,
        SUM(oi.quantity * oi.unit_price) AS total_spent,
        COUNT(DISTINCT o.order_id) AS order_count
    FROM customers c
    JOIN orders o ON c.customer_id = o.customer_id
    JOIN order_items oi ON o.order_id = oi.order_id
    WHERE o.order_date >= CURRENT_DATE - INTERVAL '12 months'
    GROUP BY c.customer_id, c.customer_name
),
previous_period AS (
    SELECT
        c.customer_id,
        SUM(oi.quantity * oi.unit_price) AS prev_total_spent
    FROM customers c
    JOIN orders o ON c.customer_id = o.customer_id
    JOIN order_items oi ON o.order_id = oi.order_id
    WHERE o.order_date >= CURRENT_DATE - INTERVAL '24 months'
      AND o.order_date = CURRENT_DATE - INTERVAL '12 months'
    GROUP BY c.customer_id, pc.category_name
    ORDER BY c.customer_id, COUNT(*) DESC
)
SELECT
    cp.customer_name,
    cp.total_spent,
    cp.order_count,
    tc.favorite_category,
    pp.prev_total_spent,
    ROUND(
        (cp.total_spent - COALESCE(pp.prev_total_spent, 0))
        / NULLIF(pp.prev_total_spent, 0) * 100, 1
    ) AS spending_change_pct
FROM current_period cp
LEFT JOIN previous_period pp ON cp.customer_id = pp.customer_id
LEFT JOIN top_categories tc ON cp.customer_id = tc.customer_id
ORDER BY cp.total_spent DESC
LIMIT 5;

-- Expected output:
-- customer_name | total_spent | order_count | favorite_category | prev_total_spent | spending_change_pct
-- Acme Corp     | 284,500.00  | 47          | Electronics       | 198,200.00       | 43.5
-- GlobalTech    | 231,800.00  | 38          | Software          | 245,100.00       | -5.4
-- ...

Example 2: Optimizing a slow query with EXPLAIN analysis

Original slow query (takes 12.4 seconds on 5M rows):

SELECT product_name, SUM(quantity * unit_price) AS revenue
FROM order_items oi, products p, orders o
WHERE oi.product_id = p.product_id
  AND oi.order_id = o.order_id
  AND o.order_date BETWEEN '2024-01-01' AND '2024-12-31'
GROUP BY product_name
ORDER BY revenue DESC;

EXPLAIN ANALYZE output (problem indicators):

Seq Scan on orders o  (cost=0.00..98456.00 rows=1245000)
  Filter: (order_date >= '2024-01-01' AND order_date = '2024-01-01' AND < '2025-01-01'` instead of `BETWEEN`, which includes the end boundary's midnight.
- **Very large result sets.** Always include `LIMIT` in exploratory queries. For production queries, add pagination with `OFFSET`/`FETCH` or keyset pagination for better performance on deep pages.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [seb1n](https://github.com/seb1n)
- **Source:** [seb1n/awesome-ai-agent-skills](https://github.com/seb1n/awesome-ai-agent-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.