The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes patterns for ingesting untrusted data through various HTTP and WebSocket interfaces (parameters, headers, and request bodies) as detailed in parameters.md, dtos.md, and websockets.md. This data is often passed to sensitive sinks such as database repositories (services-and-repos.md), file storage (stores-and-caching.md), or real-time message brokers (channels-and-sse.md). While the framework encourages strict schema validation using msgspec and DTOs, the processing of external data by the resulting applications provides a surface for indirect prompt injection if those application outputs are later interpreted by an LLM.
Ingestion points: Request handlers utilize FromPath, FromQuery, JSONBody, and MultipartBody to accept external data.
Boundary markers: The skill emphasizes the use of msgspec.Struct and DTO factories to enforce strict schema validation on all inputs.
Capability inventory: The skill enables the creation of applications with extensive capabilities including file system access (FileStore), database persistence (SQLAlchemyAsyncRepositoryService), and network broadcasting (ChannelsPlugin).
Sanitization: Input validation is handled via Litestar's SignatureModel and msgspec conversion, which ensures data conforms to defined types and constraints before reaching the business logic.

litestar