gpt-live-demo

No clear evidence of malware or deliberate sabotage is present. The code implements a voice-agent service, but it has a high-impact access-control and privacy weakness: unauthenticated network clients can query workspace memory, view conversation/task data, reveal the system prompt, create OpenAI sessions, and submit background tasks. Binding to 0.0.0.0 makes this especially dangerous unless an external trusted proxy enforces authentication and network restrictions. The memory endpoint and task APIs should be protected, and workspace file access should be narrowed and permission-controlled.

未发现该文档本身包含恶意代码或明显供应链攻击逻辑。主要安全问题是公开部署时 /api/session 无鉴权，可能导致 OpenAI quota 被滥用；应在公开前加入 Bearer token 或会话鉴权、速率限制和配额监控。由于实际 server.mjs、依赖清单和请求处理逻辑未提供，无法评估其真实网络调用和数据处理行为。