The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the development of systems that ingest and broadcast real-time data from external clients, which presents a surface for potential indirect prompt injection if untrusted data is processed by an agent or client application without strict sanitization.
Ingestion points: socket.on("message", ...) and socket.on("join-room", ...) handlers in SKILL.md and references/patterns.md.
Boundary markers: The primary implementation examples do not include specific delimiters or instructions for the agent to disregard potentially malicious instructions embedded within the message payload.
Capability inventory: The skill uses io.to(roomId).emit() to broadcast data, which could propagate untrusted content to all connected clients in a room.
Sanitization: The skill provides a dedicated security reference (references/security.md) describing sanitization with sanitize-html and validation with Joi, which mitigates the risk if followed by the user.
[SAFE]: No malicious patterns, hidden commands, or unauthorized data access were identified. The code examples demonstrate secure practices by using environment variables for sensitive credentials like JWT_SECRET and REDIS_URL rather than hardcoding them.

websocket-engineer