The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external and potentially untrusted sources such as CSV files, Parquet files, and cloud storage (S3, Azure, GCS) via functions like pl.read_csv, pl.scan_parquet, and database connectors.
Ingestion points: Functions like pl.read_csv, pl.read_parquet, pl.scan_csv, and pl.read_database mentioned in SKILL.md and references/io_guide.md enable loading external content into the agent's context.
Boundary markers: No explicit instructions or delimiters for the agent to distinguish between data content and instructions within these files are described.
Capability inventory: The skill documents capabilities to write data to the local filesystem (write_csv, sink_parquet) and external databases (write_database) as seen in references/io_guide.md and references/best_practices.md.
Sanitization: No data sanitization, validation logic, or schema enforcement recommendations are provided to mitigate the risk of processing malicious instructions embedded in the data.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the polars package using uv pip install polars. Polars is a widely recognized and reputable open-source library for data manipulation, presenting minimal risk as a dependency.

polars