The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides architectures for Retrieval Augmented Generation (RAG) and natural language to SQL (NL2SQL) in features/dbms-vector.md and features/select-ai.md. These patterns process user-supplied input to influence model behavior.
Ingestion points: User questions enter the system via bind variables or JSON parameters (e.g., :user_question).
Boundary markers: The skill suggests prompt structures that limit model output to provided context (e.g., "Answer the question using only the context provided.").
Capability inventory: The skill documents extensive database capabilities including data modification (DML), structure changes (DDL), and network communication (UTL_HTTP).
Sanitization: Documentation explicitly instructs users to sanitize input at the application layer to prevent prompt injection.
[DYNAMIC_EXECUTION]: The skill teaches the use of dynamic SQL and JavaScript in the database. These are documented alongside robust security measures, specifically advising the use of bind variables and the DBMS_ASSERT package to validate SQL identifiers.
[EXTERNAL_DOWNLOADS]: The skill references downloads for software components (utPLSQL, Swagger UI) and container images (container-registry.oracle.com). These references are to well-known technology providers and official repositories and are documented neutrally.
[COMMAND_EXECUTION]: The skill contains administrative commands for database and container management. These commands are essential for the primary purpose of database operations and are accompanied by technical guidance and safety warnings.

db