The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates that process untrusted external data from user search inputs and WebSocket messages. This creates a potential surface for indirect prompt injection if the data processed by these reactive streams is subsequently used to influence agent instructions without sanitization.
Ingestion points: Search input events in references/search-with-debounce.md and WebSocket message events in references/websocket-with-reconnection.md.
Boundary markers: Not explicitly defined in code templates.
Capability inventory: Uses fetch for network requests and WebSocket for real-time communication.
Sanitization: Includes basic JSON parsing error handling and debounce logic to limit frequency of operations.

reactive-programming