The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python and SQL to dynamically construct and execute commands on a Databricks environment. Specifically, the GDPRHandler and DataRetentionManager classes in references/implementation.md use string interpolation to build SQL queries (e.g., DELETE FROM {table_name} WHERE {user_column} = '{user_id}'). While intended for data maintenance, this pattern can be risky if inputs like user_id or table metadata are not strictly validated.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It processes external data (database tags and table contents) to perform critical operations like data deletion.
Ingestion points: Reads table tags (data_classification, retention_days) and user record data from Databricks catalogs (found in SKILL.md and references/implementation.md).
Boundary markers: None identified. The skill lacks explicit instructions to ignore potentially malicious content within the metadata or data it processes.
Capability inventory: The skill uses spark.sql() to execute destructive operations (DELETE, VACUUM) and read operations across the entire catalog (found in references/implementation.md).
Sanitization: There is no evidence of input sanitization or parameterization for the dynamically constructed SQL strings beyond standard Spark SQL behavior.

databricks-data-handling