Duckdb

The DuckDB skill enables AI agents to execute SQL queries directly against local data files, allowing for instant analysis of CSVs, JSON, Parquet, and other formats without requiring a separate database server. It acts as an in-process analytical database that handles complex joins, aggregations, and time-series calculations efficiently within the agent's execution environment.

When to use it

• Analyzing large datasets locally where setting up a persistent database cluster is impractical or too slow.
• Performing ad-hoc exploratory data analysis on CSV or JSON files uploaded during a session.
• Generating dynamic reports by joining multiple local data sources in real-time.
• Validating data quality and performing aggregations before feeding results into downstream AI workflows.

Key capabilities

• Reads diverse file formats including CSV, JSON, Parquet, SQLite, and Avro directly from the filesystem.
• Executes full SQL dialects supporting complex joins, window functions, and recursive queries.
• Operates as a zero-configuration, in-process engine requiring no external server installation.
• Handles both single-threaded and parallel execution modes for varying data sizes.

Example prompts

• "Load the sales_2023.csv file from my project directory and calculate the total revenue by region."
• "Join the customers.json and orders.parquet files to find the top 5 customers by purchase frequency."
• "Analyze the time-series data in events.log to identify any spikes in error rates over the last week."

Tips & gotchas

Ensure the AI agent has read permissions for the local file paths you specify, as DuckDB operates strictly on the local filesystem. While highly efficient for analytical workloads, avoid using this skill for high-concurrency transactional systems designed for persistent storage.