The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python f-strings to dynamically construct Spark SQL queries in several locations, including GDPRHandler.process_deletion_request, DataRetentionManager.apply_retention_policies, and generate_sar_report. This pattern is susceptible to SQL injection if inputs like user_id, user_column, or table_name are not properly sanitized.
[COMMAND_EXECUTION]: The skill requests Bash(databricks:*) permissions, allowing the agent to execute any Databricks CLI command, which represents a high level of privilege within the target environment.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted data (user-provided identifiers and table metadata) and interpolates it into executable SQL commands without sanitization or boundary markers.
Ingestion points: User-provided user_id in process_deletion_request and generate_sar_report; table and column tags fetched from system.information_schema in SKILL.md.
Boundary markers: None identified. SQL queries are constructed using direct string interpolation.
Capability inventory: Uses spark.sql() to execute DDL and DML operations (DELETE, INSERT, ALTER) and Bash(databricks:*) for environment management.
Sanitization: No evidence of input validation, escaping, or parameterized queries for the SQL execution.
[DATA_EXFILTRATION]: The generate_sar_report function in the examples allows for the extraction of all data associated with a user across all PII-tagged tables, converting it into a local Pandas DataFrame/dictionary. While intended for GDPR compliance, this provides a mechanism for significant data exposure if misused.
[EXTERNAL_DOWNLOADS]: The skill references official Databricks documentation for Delta Lake privacy and Unity Catalog security. These are well-known, trusted sources used for configuration guidance.

databricks-data-handling