Duckdb Analyst

DuckDB Analyst

Conventions

• Data lives in ./data/*.parquet (a read-only mount). Never write there.
• Use DuckDB — either the duckdb CLI or python -c "import duckdb; …".
• Always print the SQL you ran — analysts trust answers only when they see the query.
• Cap queries to LIMIT 1000 by default.
• Save intermediate results as ./cache-<slug>.parquet in the workdir.

Workflow

1. ls ./data to discover files
2. duckdb -c "DESCRIBE SELECT * FROM read_parquet('./data/X.parquet') LIMIT 0" to learn schema
3. Write the query and run it via duckdb -c "..." or a short Python script
4. If the user wants a chart, follow the matplotlib-charts skill next

Guardrails

• If the underlying file is larger than ~10 GB, warn the user before scanning it
• Never suggest or run DELETE / UPDATE / DROP — DuckDB on parquet can't anyway, but the LLM must still not suggest it

Example shell

duckdb -c "SELECT product, SUM(revenue) AS total
           FROM read_parquet('./data/sales.parquet')
           GROUP BY product ORDER BY total DESC LIMIT 1000"