The Agent Skills Directory

[Indirect Prompt Injection] (SAFE): The skill documents an attack surface consisting of multiple HTTP and WebSocket ingestion points but mitigates the risk through strict schema enforcement.\n
Ingestion points: Request body, query strings, path parameters, and WebSocket messages defined across SKILL.md, core.md, and websocket.md.\n
Boundary markers: The skill consistently utilizes the Elysia t (TypeBox) utility to define structured data schemas, ensuring untrusted input is validated before being processed by the application logic.\n
Capability inventory: Handlers include capabilities for database querying via bun:sql and static file serving via @elysiajs/static.\n
Sanitization: Schemas enforce data types, lengths, and formats (e.g., format: 'email'), and the onError hook in validation.md demonstrates how to handle validation failures securely.\n- [EXTERNAL_DOWNLOADS] (SAFE): Installation steps refer only to standard, well-known packages from the npm registry using the bun add command.\n- [CREDENTIALS_UNSAFE] (SAFE): Code samples correctly demonstrate using environment variables (process.env.JWT_SECRET) for secrets, preventing the exposure of hardcoded credentials.

bun-elysia