The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides templates that ingest and process untrusted data from remote WebSocket clients, creating a surface for indirect prompt injection.
Ingestion points: Data is received via socket.on events in references/nodejs-websocket-server-socketio.md and the ws.iter_any() loop in references/python-websocket-server-aiohttp.md.
Boundary markers: The code snippets do not include delimiters or specific instructions to treat incoming data as untrusted.
Capability inventory: The implementations include capabilities for broadcasting data to other clients and performing database operations (Message.create, save_message).
Sanitization: No input validation or sanitization routines are included for the message payload fields.
[DATA_EXFILTRATION]: The Node.js implementation in references/nodejs-websocket-server-socketio.md uses a permissive CORS policy (origin: "*") which is insecure for production environments as it allows any origin to connect.
[DATA_EXFILTRATION]: The Python implementation in references/python-websocket-server-aiohttp.md transmits raw error strings (str(error)) to clients, which could expose internal stack traces or server logic details.

websocket-implementation