The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from various external sources, which presents a potential attack surface for indirect prompt injection if those sources contain adversarial instructions.\n
Ingestion points: The skill uses pl.read_csv, pl.scan_csv, pl.read_parquet, pl.scan_ndjson, and pl.read_database_uri to bring external content into the agent's context (found in SKILL.md and references/io_guide.md).\n
Boundary markers: The documentation advises on using "authorized destination," "data scope," and "least-privilege connections" to manage access (found in SKILL.md and references/io_guide.md).\n
Capability inventory: The skill possesses significant data manipulation and output capabilities, including write_csv, write_parquet, write_database, and integration with other libraries like Pandas and NumPy (found in references/io_guide.md and references/pandas_migration.md).\n
Sanitization: The guides recommend the use of "parameterized user values" for database interactions and "spreadsheet-safe export policy" for CSV outputs (found in references/io_guide.md).

polars