The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides robust templates for handling authentication credentials such as API keys, service principal secrets, and passwords. It explicitly demonstrates safe practices for these values, including using Databricks Secrets (dbutils.secrets) and implementing redaction logic to ensure sensitive data is not leaked into logs or system representations.
[PROMPT_INJECTION]: The skill provides patterns for ingesting data from external APIs and databases, which represents a surface for indirect prompt injection. However, this is well-mitigated by the inclusion of production patterns that focus on structured data mapping and explicit input validation.
Ingestion points: External data sources accessed via the requests library or database-specific drivers within DataSourceReader implementation classes.
Boundary markers: The templates primarily process data into structured Spark Rows; while explicit text delimiters for the AI are not provided, the architectural pattern relies on schema enforcement.
Capability inventory: Network access via the requests library, file system write access for Dead Letter Queues (DLQ), and execution of standard Python build tools.
Sanitization: The references/production-patterns.md file provides specific validation logic (_validate_identifier) using regular expressions to sanitize table names and other identifiers, effectively mitigating the risk of injection when the generated code interacts with external systems.
[COMMAND_EXECUTION]: Instructions involve standard project initialization and testing commands using verified Python tools like uv, poetry, and pytest. There are no instances of arbitrary or dangerous command execution.

spark-python-data-source