chat2duckdb

Register the CSV as a table using read_csv_auto, then execute standard DuckDB SQL against it. This Skill wraps that flow in a script: pass --file_path with --mode query and your --sql statement to get results with execution time and row counts.

Run the analyzer in describe mode, which reports row counts, column types, numeric statistics like mean and median, categorical value distributions, date ranges, and missing-value percentages. This helps you write accurate queries on the first attempt.

Yes. Excel files are read via pandas and registered into DuckDB, with a specific sheet selectable via --excel_sheet. Chinese column names are automatically wrapped in double quotes during query correction so DuckDB can parse them.

Common failures include trailing semicolons, Chinese punctuation, and misspelled column names. The correction engine fixes syntax patterns, converts Chinese punctuation, and matches wrong column names to the closest real column using prefix matching and edit distance up to 2.

Use --sample_fraction to validate queries on a fraction of the data first, add LIMIT to cap result rows, and prefer aggregate queries over full-table scans. The connection sets a 4GB memory limit by default.

Yes. Export results directly to CSV, Excel, JSON, or Parquet via --output, or use --persist_db_path with --persist_table to write tables into a DuckDB database file for cross-batch accumulation and multi-table joins later.