{
    "action": "tool_call",
    "reasoning": "Use the profiling metadata to get null counts for all columns across the 5 tables in one query, keeping only columns with nulls and computing percent of rows.",
    "tool": "query_sql",
    "args": {
        "sql": "SELECT c.table_name, c.column_name, c.null_count, t.row_count, ROUND(100.0 * c.null_count / t.row_count, 1) AS pct_missing FROM analytics.column_overview c JOIN analytics.table_overview t ON c.table_name = t.table_name WHERE c.null_count > 0 ORDER BY pct_missing DESC, c.table_name, c.column_name",
        "purpose": "Null counts and % missing per column per table",
        "source": "raw"
    }
}