snowflake-sql - SKILL.md Agent Skill

name: snowflake-sql description: "Snowflake-specific SQL patterns: QUALIFY for window filtering, LATERAL FLATTEN for arrays, semi-structured VARIANT data, ILIKE for case-insensitive matching, date functions, and time travel." type: skill

Snowflake SQL Skill

1. Window Function Filtering — Use QUALIFY

Instead of wrapping in a subquery, use QUALIFY:

-- Find the latest record per customer
SELECT customer_id, order_date, amount
FROM orders
QUALIFY ROW_NUMBER() OVER (PARTITION BY customer_id ORDER BY order_date DESC) = 1;

-- Top 5 products by sales
SELECT product_id, total_sales
FROM sales_summary
QUALIFY DENSE_RANK() OVER (ORDER BY total_sales DESC) <= 5;

2. Case-Insensitive Matching — Use ILIKE

-- Case-insensitive LIKE
WHERE product_name ILIKE '%widget%'

-- Case-insensitive equality
WHERE UPPER(status) = 'ACTIVE'
-- or
WHERE status ILIKE 'active'

3. Arrays and Semi-Structured Data — LATERAL FLATTEN

-- Explode an ARRAY column
SELECT t.id, f.value AS item
FROM table t,
LATERAL FLATTEN(input => t.array_col) f;

-- Access VARIANT fields
SELECT col:field_name::STRING AS field_value
FROM table;

-- Parse JSON string
SELECT PARSE_JSON(json_col):key::STRING AS val
FROM table;

4. Date Functions

-- Add/subtract time
DATEADD(day, 7, order_date)          -- 7 days from order_date
DATEADD(month, -1, current_date())   -- 1 month ago

-- Difference between dates
DATEDIFF(day, start_date, end_date)  -- days between dates
DATEDIFF(month, start_date, end_date)

-- Truncate to period
DATE_TRUNC('month', event_ts)
DATE_TRUNC('year', event_ts)

-- Current timestamp
CURRENT_TIMESTAMP()
CURRENT_DATE()

5. String Functions

SPLIT_PART(col, '/', 1)              -- split by delimiter, get Nth part
REGEXP_SUBSTR(col, '[0-9]+')         -- first match of regex
TRIM(col)                            -- remove leading/trailing whitespace
LTRIM(col, '0')                      -- remove leading zeros
UPPER(col) / LOWER(col)
CONCAT(col1, '-', col2)              -- or col1 || '-' || col2

6. Null-Safe Equality

-- NULL-safe: TRUE when both are NULL or both equal
col1 IS NOT DISTINCT FROM col2

-- COALESCE for default values
COALESCE(col, 'unknown')

7. Time Travel (querying historical data)

-- Query table as it was 1 hour ago
SELECT * FROM my_table AT (OFFSET => -3600);

-- Query at a specific timestamp
SELECT * FROM my_table AT (TIMESTAMP => '2024-01-01'::TIMESTAMP);

8. Common Anti-Patterns to Avoid

Do NOT use = NULL — use IS NULL
Do NOT use <> for NULL comparison — use IS NOT NULL
Prefer QUALIFY over subquery wrapping for window filters
When accessing VARIANT fields, always cast: col:field::STRING

9. Benchmark Patterns

Numeric precision: Snowflake returns DECIMAL/NUMBER with configurable precision. Do NOT cast to FLOAT unless needed — precision loss fails exact-match evaluation.
IDENTIFIER case: Snowflake upper-cases identifiers by default. Use double-quotes "lower_case_col" when column names are lowercase in source. Always check with describe_table.
LISTAGG: Use LISTAGG(col, ',') WITHIN GROUP (ORDER BY col) for string aggregation (not GROUP_CONCAT).
TRY_CAST / TRY_TO_NUMBER: Use for safe type conversion that returns NULL instead of error.
OBJECT_KEYS / ARRAY_SIZE: Useful for introspecting semi-structured data before querying.