The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of data from external files and remote sources (HTTPS, S3), creating a surface where malicious instructions embedded in the data could attempt to influence the agent.
Ingestion points: Examples in SKILL.md demonstrate reading from file paths and URLs (e.g., FROM 'data/*.parquet', read_csv('f.csv', ...), FROM 'https://.../f.parquet').
Boundary markers: No specific instructions are provided to the agent to treat external data as untrusted or to ignore instructions found within processed files.
Capability inventory: The skill uses the duckdb CLI, which can perform file system operations (read/write) and network requests.
Sanitization: No data validation or sanitization steps are described for the content of ingested files.
[COMMAND_EXECUTION]: The skill directs the agent to interact with the system shell to execute the duckdb CLI tool for running SQL queries.
Evidence: Multiple shell command examples are provided in the "Running SQL from a shell" section, including the use of duckdb -c "..." and heredocs.

duckdb-sql