The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides documentation and examples for the EXECUTE IMMEDIATE statement in references/sql-scripting.md. This feature allows for the construction and execution of dynamic SQL at runtime, which is a standard administrative and engineering capability within the Databricks platform.
[INDIRECT_PROMPT_INJECTION]: The skill documents surface areas for indirect prompt injection in references/ai-functions.md, specifically through the read_files function (which ingests data from cloud storage and volumes) and the http_request function (which processes external API responses). These functions could potentially ingest untrusted content that might influence an agent's behavior during data enrichment tasks.
Ingestion points: External data enters the context via read_files, http_request, and remote_query operations as documented in references/ai-functions.md.
Boundary markers: The provided documentation does not explicitly define prompt delimiters for these ingestion points, though it mentions the ai_mask function for PII redaction.
Capability inventory: The skill documents capabilities for network operations via http_request and file write operations via SQL INSERT and MERGE commands.
Sanitization: The skill highlights the ai_mask function as a mechanism for sanitizing sensitive data (PII) before further processing.

databricks-dbsql