The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes metadata retrieved from external catalogs such as Unity Catalog or Polaris. This presents an ingestion surface for untrusted data (e.g., malicious strings in table/column metadata), which is inherent to the skill's intended purpose of table discovery and metadata inspection.
Ingestion points: Metadata is ingested from external catalogs via pyiceberg and databricks-sdk (SKILL.md).
Boundary markers: No explicit sanitization or delimiters are provided for the formatted markdown output.
Capability inventory: Limited to metadata retrieval and SQL statement generation.
Sanitization: No explicit sanitization or escaping of ingested metadata values is defined.
[EXTERNAL_DOWNLOADS]: The code examples utilize the pyiceberg and databricks-sdk libraries. These are official and well-known libraries for interacting with Apache Iceberg and Databricks platforms respectively.

table-inspection