简体中文

Security fixes are evaluated for the latest published minor line.

Do not disclose a suspected vulnerability in a public issue.

Use GitHub private vulnerability reporting. Include the affected version, operating system, installation method, impact, and the smallest safe reproduction you can provide.

Remove API keys, access tokens, prompts, responses, and personal data before attaching configuration or logs. Acknowledge that model prompts can be sensitive even when credentials have been removed.

Relevant reports include credential exposure, history access, dashboard or proxy authorization bypass, cross-session data exposure, unsafe package loading, and reproducible boundary violations during model handoff.

Provider outages, provider model behavior, and an intentionally exposed local listener without external access controls are not product vulnerabilities by themselves.

Reports are reviewed on a best-effort basis. Fix timing depends on severity, reproducibility, and release constraints.

The proxy and dashboard bind to loopback by default and do not provide user login. If you expose either listener, place trusted authentication, TLS, and network access control in front of it.

Traffic history may contain complete prompts and responses. Protect the log directory with the same care as provider credentials.

安全修复以最新发布的次版本系列为评估对象。

不要在公开 Issue 中披露疑似安全漏洞。

请使用 GitHub 私密漏洞报告。 报告中应包含受影响版本、操作系统、安装方式、影响，以及可以安全提供的最小复现。

提交配置或日志前，必须删除 API Key、访问令牌、提示词、回答和个人数据。即使已经 删除凭据，模型提示词本身仍可能包含敏感信息。

相关范围包括凭据泄漏、历史数据越权、代理或管理界面认证绕过、跨会话数据泄漏、 不安全的软件包加载，以及跨模型接管时可复现的边界破坏。

服务商故障、模型自身行为，以及没有外部访问控制却被主动暴露的本地监听端口， 本身不属于产品漏洞。

所有报告按实际能力尽力处理。修复时间取决于严重程度、可复现性和发布约束。

代理和管理界面默认监听 loopback，并且不提供用户登录。需要对外暴露时，必须在 前方增加可信认证、TLS 和网络访问控制。

流量历史可能包含完整提示词与回答。日志目录需要获得与服务商凭据相同级别的保护。