The Agent Skills Directory

[SAFE]: The skill is a massive library of documentation and code examples for Oracle Database operations. It specifically includes defensive patterns for AI agents, such as intent disambiguation, destructive operation guards, and SQL injection avoidance.
[INDIRECT_PROMPT_INJECTION]: The skill supports mapping natural language to SQL, which creates an indirect prompt injection surface. The evidence chain is as follows: (1) Ingestion points: User prompts processed in agent/nl-to-sql-patterns.md and features/select-ai.md. (2) Boundary markers: Present in agent/intent-disambiguation.md, which instructs the agent to clarify ambiguous or destructive intents. (3) Capability inventory: SQL execution capabilities documented in appdev/python-oracledb.md and ords/ords-architecture.md. (4) Sanitization: Extensively covered in sql-dev/dynamic-sql.md and agent/safe-dml-patterns.md via DBMS_ASSERT and bind variables. This risk is inherent to the skill's primary purpose and is properly mitigated by the provided safety guidelines.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for pulling container images from container-registry.oracle.com and cloning software from github.com. These are trusted organizations and services, and the references are made in a neutral, instructional context.

db