The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of data from external and potentially untrusted sources (CSV, JSON, Parquet), which creates a surface for indirect prompt injection attacks. An attacker could embed malicious instructions in these files that an agent might inadvertently execute.
Ingestion points: Data reading and scanning operations like pl.read_csv, pl.scan_csv, pl.read_json, and pl.read_database located in references/io_guide.md.
Boundary markers: The documentation does not provide instructions for using delimiters or boundary markers to distinguish data from instructions.
Capability inventory: The skill enables capabilities such as writing files to the local system (write_parquet), executing database queries (read_database), and accessing cloud storage (AWS S3, Azure Blob, GCS).
Sanitization: No specific data validation or sanitization methods are described for the content being read from external files.
[CREDENTIALS_SAFE]: The documentation correctly uses placeholders (e.g., 'your_key', 'your_secret', 'user:pass') when demonstrating how to configure environment variables for cloud services and database connection strings.
[EXTERNAL_DOWNLOADS]: Recommends the installation of the official polars package and standard dependencies like sqlalchemy, pyarrow, and connectorx from legitimate registries.

polars