semantic-view-debug

Provide the failing natural language question, and the Skill generates SQL with the current semantic view, analyzes gaps like missing joins or metrics, and applies approved fixes to the semantic view YAML. It then re-generates and executes the SQL to validate the fix.

Run the debug workflow after completing semantic view setup: diagnose the issue by generating SQL, perform root cause analysis mapping SQL errors to semantic view gaps, then apply and validate optimizations. Each optimization iteration requires explicit user approval.

Common gaps include missing relationships between tables causing absent JOINs, missing metric definitions causing wrong aggregations, poorly described dimensions causing missing columns, and missing named filters causing incorrect WHERE clauses.

Yes, it executes both the optimized generated SQL and any user-provided ground truth SQL, comparing row counts and columns side by side. Without ground truth, it asks the user to review results before declaring success.

You must first complete the semantic-view-setup skill, which loads the setup instructions, creates a workspace directory, and familiarizes you with the semantic view tools. Debugging cannot start without this initialization.