The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external text data (e.g., from CUSTOMER_SUPPORT_TICKETS) for named entity recognition and classification, which is a standard surface for indirect prompt injection.\n
Ingestion points: Text data is ingested from Exasol database tables via SourceTableExtractor in references/text-ai-extension.md.\n
Boundary markers: The provided examples do not explicitly show delimiters or safety instructions for the ingested text.\n
Capability inventory: The skill executes Python code and SQL User Defined Functions (UDFs) within the database via the Extraction.run() method.\n
Sanitization: Sanitization of the input text is not detailed in the skill's reference scripts.\n- [EXTERNAL_DOWNLOADS]: The initialization process downloads machine learning models and language containers from external sources.\n
Evidence: initialize_text_ai_extension fetches default models from Hugging Face and a Script Language Container (SLC) for the Text AI Extension as documented in references/text-ai-extension.md.\n
Context: These downloads target well-known repositories and official vendor resources necessary for the extension's primary functionality.\n- [DYNAMIC_EXECUTION]: The skill deploys and executes user-defined functions (UDFs) and language containers within the Exasol database environment.\n
Evidence: initialize_text_ai_extension installs TXAIE scripts and language containers into the configured database schema.\n
Context: This is a standard procedure for enabling AI capabilities within the database and uses vendor-provided packages from exasol-labs.

exasol-text-ai