The Agent Skills Directory

[PROMPT_INJECTION]: The utility script scripts/apply_constraints.py is susceptible to indirect prompt injection (specifically SQL injection) through the dynamic construction of DDL statements.
Ingestion points: The script retrieves values for catalog and gold_schema directly from Databricks widgets using dbutils.widgets.get().
Boundary markers: No delimiters or validation logic are present to ensure that the inputs match expected naming patterns or do not contain malicious SQL fragments.
Capability inventory: The skill possesses the capability to execute administrative database operations via spark.sql(), including ALTER TABLE and DROP CONSTRAINT.
Sanitization: The script uses Python f-strings to interpolate widget values directly into SQL strings without any escaping or validation (e.g., spark.sql(f"ALTER TABLE {fqn} ...")). An attacker controlling the widget parameters could potentially execute arbitrary SQL.
[COMMAND_EXECUTION]: The script scripts/apply_constraints.py performs dynamic execution of SQL commands generated at runtime. While these commands are intended for metadata management, the pattern of building executable strings from external input is a known security risk vector.

unity-catalog-constraints