Skip to content
FunCoding

sql-correctness

SQL evaluation criteria for Databricks. Load when the trace contains execute_sql tool calls or SQL code in responses. Covers syntax validity, Unity Catalog patterns, and Databricks-specific SQL features.

Star
1,942
Category
Data
Last commit
Aug 13, 2026

.test/eval-criteria/sql-correctness/SKILL.md

View on GitHub

SQL Correctness Rubric

When evaluating SQL in agent traces, check these dimensions:

1. Unity Catalog Namespace

  • Must use 3-level namespace: catalog.schema.table
  • Never use unqualified table names
  • Catalog/schema should match the user's context (not hardcoded main.default unless appropriate)

2. Modern DDL Syntax

  • Use CREATE OR REPLACE instead of DROP IF EXISTS + CREATE
  • Use ALTER TABLE ... SET TBLPROPERTIES for table properties
  • Use COMMENT ON for documentation

3. Tool Selection

  • Must use mcp__databricks__execute_sql for SQL execution
  • Must NOT use Bash with databricks sql CLI as a workaround
  • Must NOT use notebook execution for simple queries

4. Databricks SQL Features

  • Use Delta-specific syntax where appropriate (MERGE INTO, OPTIMIZE, VACUUM)
  • Use IDENTIFIER() function for dynamic table references
  • Use SELECT * FROM read_files() for file ingestion, not COPY INTO (unless streaming)

5. Syntax Validity

  • SQL must be syntactically valid for Databricks SQL (Spark SQL dialect)
  • String literals use single quotes, identifiers use backticks if needed
  • Semicolons at statement boundaries

6. Safety

  • No string interpolation for user-provided values in SQL
  • Use parameterized queries where applicable
  • No DROP operations unless explicitly requested

See detailed patterns for specific syntax examples.

Similar Skills

agentic-os

AFaffaan-m/ECC278k4,046 new stars in the last 7 days

health

GAgarrytan/gstack136k883 new stars in the last 7 days

deprecation-and-migration

ADaddyosmani/agent-skills106k3,628 new stars in the last 7 days

Official

dashboard

NEnexu-io/open-design101k1,126 new stars in the last 7 days

Official

flowai-live-dashboard-template

NEnexu-io/open-design101k1,126 new stars in the last 7 days

handoff

THthedotmack/claude-mem100k3,024 new stars in the last 7 days