jlcodes99/cockpit-tools v0.25.0 on GitHub

更新日志（中文）

Codex 模型供应商现支持完整管理工作流：Codex 模型供应商页新增单供应商多 API Key、可搜索的 API Key 与实例选择弹框、供应商搜索/筛选/排序、批量选择与删除、供应商服务面板、OAuth 绑定，以及与账号页卡片交互对齐的快速启用操作。
Codex 第三方 API Key 额度查询现支持 new-api 与 sub2api：Cockpit 会探测支持的额度接口，缓存已识别的供应商类型，保留历史额度数据，跟随现有配额刷新策略，并在账号卡片、首页卡片、模型供应商卡片、服务面板和 macOS 菜单栏中按供应商类型展示核心字段。
Codex 供应商协议选择改为显式配置：供应商添加默认使用 Responses 原生模式，仅已知 Chat Completions 供应商默认选择 Chat Completions；界面提供带说明的样式化协议选择器，并且只有 Chat Completions 供应商会走本地网关。
WebDAV 备份同步：设置页新增 WebDAV 备份同步配置，并补齐服务调用、翻译和数据传输支持，可用于同步 Cockpit 备份数据。感谢 @xdd666t。
Codex 唤醒与会话修复吸收社区 PR 改进：唤醒请求现在会注入官方 StartCascadeRequest.source 字段，Codex 可见性修复会在修复前协调 session_index.jsonl。感谢 @Slone123c 和 @andrew05060414。

Codex 模型供应商现可接入 deepseek-v4-pro 等国产 Chat Completions 模型：Responses 原生供应商保持直连，Chat Completions 供应商通过本地网关完成协议转换，并且模型目录与图片输入等网关相关配置只在选择该协议时显示。
Codex 模型供应商卡片与服务面板复用账号页配额展示风格：供应商卡片会保留历史额度数据，提供手动刷新入口，按 new-api 与 sub2api 分别展示关键字段，并把供应商详情收敛到单个可滚动服务面板中。
Codex Linux OAuth 登录更稳定：OAuth 回调处理避免重复完成，改善 Linux 登录流程。

Codex 模型供应商 OAuth 绑定现在会在启用供应商时生效：模型供应商的 OAuth 绑定会同步到实际用于启动的 API Key 账号，与账号页行为保持一致。
通过官方 Language Server 执行 Codex 唤醒不再因缺少请求来源失败：唤醒请求现在会注入上游服务需要的官方 StartCascadeRequest.source 字段。感谢 @Slone123c。
Codex 会话可见性修复现在会先协调 session_index.jsonl：修复流程会更新 session index，让隐藏或过期会话能够更可靠地修复。感谢 @andrew05060414。

Codex model providers now support a full provider-management workflow: the Codex model provider page adds multiple API Keys per provider, searchable API Key and instance pickers, provider search/filter/sort, bulk selection and deletion, provider service panels, OAuth binding, and quick enable actions that align with the account page card interactions.
Codex third-party API Key quota detection now supports new-api and sub2api providers: Cockpit detects supported quota endpoints, caches the detected provider type, keeps previous quota data visible, follows the existing quota refresh strategy, and renders provider-specific core metrics across account cards, dashboard cards, model provider cards, service panels, and the macOS menu bar.
Codex provider protocol selection is now explicit: provider setup defaults to Responses-native mode except for known Chat Completions providers, exposes a styled protocol selector with inline help, and only uses the local gateway for Chat Completions providers.
WebDAV backup synchronization: Settings now includes WebDAV backup sync configuration, service wiring, locale coverage, and data-transfer support for synchronizing Cockpit backup data. Thanks @xdd666t.
Codex wakeup and session-repair improvements from community PRs: wakeup requests now include the official StartCascadeRequest.source field, and Codex visibility repair reconciles session_index.jsonl before repair. Thanks @Slone123c and @andrew05060414.

Codex model providers can now connect Chat Completions models such as deepseek-v4-pro: Responses-native providers stay in direct mode, while Chat Completions providers use the local gateway for protocol conversion and only show gateway-related model catalog and image-input controls when that protocol is selected.
Codex model provider cards and service panels now reuse the account-page quota presentation: provider cards keep cached quota data visible, expose manual refresh controls, render new-api and sub2api fields with provider-specific layouts, and keep provider details in one scrollable service panel.
Codex OAuth login is more stable on Linux: OAuth callback handling avoids duplicate completion and improves the Linux login flow.

Codex provider OAuth binding now takes effect when enabling a model provider: model provider OAuth binding is synchronized to the actual API Key account used for launch, matching the account page behavior.
Codex wakeup through the official Language Server no longer fails because of a missing request source: wakeup requests now inject the official StartCascadeRequest.source field expected by the upstream service. Thanks @Slone123c.
Codex session visibility repair now reconciles session_index.jsonl before repairing visibility: the repair flow updates the session index so hidden or stale sessions can be repaired more reliably. Thanks @andrew05060414.