The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process free-form, untrusted text (such as customer reviews and support tickets) using Large Language Models (LLMs) via Databricks AI functions.
Ingestion points: The source_text column in silver_review_sentiment (defined in pipeline-templates.md) ingests data from external sources like customer_reviews.
Boundary markers: The SQL templates do not implement explicit boundary markers or instructions to the AI functions to ignore potential injections within the source text.
Capability inventory: The skill uses ai_analyze_sentiment, ai_classify, ai_extract, and ai_summarize to process data and writes the resulting structured data into Materialized Views in the Silver and Gold layers.
Sanitization: While the skill includes quality gates (e.g., LENGTH(TRIM(review_text)) < 5), these are intended for cost and data quality control rather than security sanitization against malicious instructions embedded in the text.
[EXTERNAL_DOWNLOADS]: The skill references official Databricks documentation for AI functions (docs.databricks.com). This is an informational reference to a well-known service and does not involve downloading executable code.

sentiment-analysis