The Agent Skills Directory

[PROMPT_INJECTION]: The skill describes a pattern for custom RAG pipelines (Pattern 2) where text extracted from unstructured documents is directly concatenated into a prompt for the ai_query function. This creates a surface for indirect prompt injection, as malicious instructions within a processed document could potentially influence the LLM's behavior.\n
Ingestion points: The read_files function targeting Databricks Unity Catalog Volumes (e.g., /Volumes/catalog/schema/volume/invoices/) as shown in SKILL.md.\n
Boundary markers: The example ai_query call in Pattern 2 lacks explicit delimiters or boundary markers to isolate the text variable from the system instructions.\n
Capability inventory: The skill utilizes ai_parse_document for extraction, ai_query for LLM invocation, and Spark SQL/Python for data processing.\n
Sanitization: There is no evidence of sanitization, filtering, or validation of the extracted document content before it is processed by the model.\n- [EXTERNAL_DOWNLOADS]: The skill references an official Databricks example repository on GitHub (github.com/databricks/bundle-examples). This reference is documented as a link to a well-known service associated with the author's platform and follows best practices for providing official examples.

databricks-parsing