The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill guides the agent in building pipelines that ingest data from external sources, presenting a standard attack surface for indirect prompt injection.\n- Ingestion points: The skill utilizes read_files, read_kafka, and other streaming sources to bring external data into the Databricks environment, as documented in SKILL.md and format-specific references like references/kafka.md.\n- Boundary markers: Safe development practices are encouraged through the use of explicit schemas and Expectations (e.g., @dp.expect_or_fail), which establish validation boundaries for incoming data.\n- Capability inventory: Capabilities include project deployment via the official databricks CLI and the execution of data processing tasks using Spark SQL and Python on Databricks clusters.\n- Sanitization: The skill demonstrates data sanitization by using from_json with strict schemas and implementing quarantine patterns for malformed data using the rescuedDataColumn feature.

databricks-pipelines