The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements real-time communication patterns (WebSockets, SSE, Long Polling) which act as ingestion points for external, untrusted data, creating a surface for indirect prompt injection.
Ingestion points: The onmessage handler in the TypeScript ManagedWebSocket implementation and the handleConn function in the Go implementation process data frames from external clients. Additional surfaces include the Last-Event-ID header in SSE and query parameters in the long polling handler.
Boundary markers: The provided code examples do not include explicit prompt delimiters or instructions to the agent to ignore potentially malicious commands embedded within the received real-time data.
Capability inventory: The skill enables full-duplex network communication, HTTP streaming, and integration with Redis Pub/Sub systems.
Sanitization: Code examples demonstrate standard protocol handling (e.g., JSON parsing) but do not include content-level sanitization for preventing prompt injection attacks.
[EXTERNAL_DOWNLOADS]: The Go examples reference github.com/gorilla/websocket, which is a widely used and well-known library for WebSocket communication. The TypeScript examples utilize standard libraries including express and redis.
[SAFE]: The skill incorporates several security best practices, such as using short-lived tickets for authentication instead of long-lived tokens in query strings, setting connection limits to prevent resource exhaustion, and implementing backpressure strategies to handle slow consumers.

real-time-communication-patterns