realtime-systems
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements communication protocols (WebSocket, SSE, and Socket.IO) that ingest untrusted data from external clients and broadcast it to other participants without explicit sanitization.
- Ingestion points: Data enters through
ws.on('message')handlers inSKILL.mdandtemplates/websocket-server.ts, and viareq.bodyin the SSE subscription and event handlers intemplates/sse-handler.ts. - Boundary markers: The templates lack instructions for the agent to treat incoming message content as untrusted or to use delimiters to prevent embedded instructions from influencing the system.
- Capability inventory: The skill's capabilities are limited to message routing and room management; no high-risk system calls (like
evalorsubprocess) are performed on the data. - Sanitization: There is no evidence of content filtering, escaping, or validation of the message payloads before they are distributed to other clients.
Audit Metadata