The Agent Skills Directory

[SAFE]: The skill serves as a comprehensive technical guide for Databricks AI Functions. All referenced URLs point to official Databricks documentation (docs.databricks.com) or the official Databricks GitHub repository (github.com/databricks/bundle-examples), which are trusted sources for this vendor.
[SAFE]: The provided SQL and PySpark code snippets demonstrate standard usage of built-in platform features such as ai_query, ai_extract, and ai_parse_document. These operations are performed within the governed environment of Databricks Unity Catalog volumes and SQL warehouses.
[SAFE]: The skill includes instructions for using the Databricks CLI (serving-endpoints list) to discover available models. This is a legitimate administrative task within the platform's ecosystem and does not involve unauthorized command execution or privilege escalation.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external data (documents, user reviews) using LLMs, which inherently creates an attack surface for indirect prompt injection. However, the skill follows safe patterns by providing documentation on structured extraction and error handling, and this capability is the primary intended purpose of the platform's AI functions.
Ingestion points: Data is ingested via read_files from Unity Catalog Volumes or via spark.table operations in SKILL.md and reference files.
Boundary markers: Code examples do not explicitly define delimiters for untrusted content within prompts, which is typical for platform documentation.
Capability inventory: The skill utilizes ai_query and ai_gen for model inference and read_files for data ingestion.
Sanitization: No explicit sanitization of input data is demonstrated in the code examples.

databricks-ai-functions