The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates exposing database entities as tools to an AI agent using AddDbContextAITools and the Model Context Protocol (MCP).
Ingestion points: Database records from configured entities (e.g., Orders, Customers) are ingested into the agent context via tool execution results (SKILL.md).
Boundary markers: The skill documentation does not define specific output delimiters or instructions for the agent to treat database content as untrusted data.
Capability inventory: The agent is granted significant capabilities including querying, aggregating, inserting, updating, and deleting database records through automatically named tools like {entity}_query and {entity}_update (SKILL.md).
Sanitization: While the skill provides developer-level controls for security such as tenant-based scoping (.Where<ICurrentTenant>), field exclusion (.Ignore), and capability restrictions (AllowQuery), it remains vulnerable if the database contains content provided by untrusted third parties that the agent then processes.
[DYNAMIC_EXECUTION]: The skill utilizes native library loading for SQLite functionality.
Evidence: The documentation describes using SqliteVec.RegisterAutoExtension() and UseShinyVectors() to load the sqlite-vec native extension for vector search capabilities (SKILL.md).
Context: This is a standard procedure for extending SQLite functionality but involves dynamic loading of executable code at runtime.

shiny-efcore