jlcodes99/cockpit-tools v1.1.5 on GitHub

更新日志（中文）

Codex API 服务支持保护绑定 OAuth 账号的额度：可分别为 5 小时和周窗口设置 1% 到 100% 的保留值；任一剩余额度达到保留值时，HTTP、WebSocket、内置网关、sidecar 和会话亲和路由只会从可用池中移除该绑定账号，配额快照缺失、过期、无效或刷新失败时按不可用处理。
OAuth 额度保护支持持续监控和状态展示：API 服务运行期间每分钟刷新绑定账号额度，并在成功使用后按节流规则补充刷新；sidecar 模式无需重启即可热更新动态配额快照，Codex 账号卡片会在接近或达到保留值时显示生效窗口、剩余额度和保留值。

公告弹框改为全应用生效：公告检查由常驻应用层 Host 负责，不再依赖仪表盘是否打开；后台定时、窗口聚焦和恢复可见时会按节流规则刷新，已有弹框关闭后再展示公告，页面切换也不会重复提示。
Codex 账号选择器与账号总览行为统一：账号总览会保存搜索状态，API 服务成员选择器和 OAuth 绑定选择器复用相同的搜索、套餐与有效性、标签、分组、排序方向和自定义顺序；缺少可用 refresh_token 的 OAuth 账号会保留在列表中并显示为不可选和对应说明，不再直接从选择器中消失。
Codex API Key 用量刷新改为跨页面集中管理：定时刷新会更新适用的 API Key 账号，仪表盘、账号总览和模型供应商管理共享同一份用量缓存；不支持用量查询的供应商会被记住以避免重复请求，手动刷新仍可强制重试。
桌面更新和发布资产改为精确匹配目标：Windows MSI 与 NSIS、macOS Apple Silicon 与 Intel，以及 Linux x86_64 和 ARM64 的 AppImage、DEB、RPM 分别使用独立签名清单和稳定文件名；全部目标构建完成前继续保留旧版 latest.json，避免老客户端读取到不完整的发布元数据。

修复 Codex 5.6 Responses Lite 请求兼容问题：gpt-5.6-sol、gpt-5.6-terra 和 gpt-5.6-luna 现在会声明并强制禁用并行工具调用，Responses Lite 请求头会在 /responses、/responses/compact、HTTP 和 WebSocket 链路中保留，非 Lite 模型也会保留显式设置的 parallel_tool_calls: false。
修复 Windows 官方 ChatGPT 启动路径迁移问题：发现逻辑现在优先选择 ChatGPT Store 包；检测到 ChatGPT 后会迁移已保存的官方 Codex Store 路径，排除仅名称命中的辅助程序，同时保留用户自定义可执行文件路径。
修复异常 Codex 配额响应被当作满额的问题：缺失或超出范围的 used_percent 以及账号准备失败现在会持久化为刷新错误，不再产生误导性的剩余额度或绕过额度保护。

Codex API Service can protect quota on its bound OAuth account: separate 1-100% reserves can be configured for the 5-hour and weekly windows; HTTP, WebSocket, embedded-gateway, sidecar, and session-affinity routing remove only the bound account from the eligible pool when either remaining quota reaches its reserve, while missing, stale, invalid, or failed quota snapshots fail closed.
OAuth quota protection is continuously monitored and visible: while API Service is running, the bound account quota refreshes every minute and after successful use with request throttling; sidecar mode hot-reloads the dynamic quota snapshot without restarting, and the Codex account card shows the effective window, remaining quota, and reserve when quota is near or below the configured threshold.

Announcement popups now work across the entire app: announcements are checked from a persistent app-level host instead of only when the dashboard is open, with throttled background, focus, and visibility refreshes; popups wait for existing dialogs to close and do not repeat when navigating between pages.
Codex account selectors now share the account-overview behavior: overview search state is persisted, and the API Service member picker plus OAuth binding selector reuse the same search, plan and validity, tag, group, sort direction, and custom-order rules; OAuth accounts without a usable refresh_token remain visible but disabled with an explanation instead of disappearing from the picker.
Codex API Key usage refresh is centralized across account views: scheduled refreshes update eligible API Key accounts, and cached usage is shared between the dashboard, account overview, and model provider manager; providers without usage-query support are remembered to avoid repeated requests, while manual refresh can force a retry.
Desktop updates and release assets now resolve the exact target: Windows MSI and NSIS, macOS Apple Silicon and Intel, and Linux AppImage, DEB, and RPM packages for x86_64 and ARM64 receive separate signed manifests and stable asset names; legacy latest.json remains available until all target builds finish so older clients do not receive incomplete release metadata.

Fixed Codex 5.6 Responses Lite request compatibility: gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna now advertise and enforce disabled parallel tool calls, the Responses Lite header is preserved across /responses, /responses/compact, HTTP, and WebSocket paths, and non-Lite models retain an explicit parallel_tool_calls: false setting.
Fixed Windows official ChatGPT launch-path migration: discovery now prefers the ChatGPT Store package over legacy Codex packages, migrates saved official Codex Store paths when ChatGPT is available, rejects keyword-matching helper executables, and preserves custom executable paths.
Fixed invalid Codex quota responses being treated as fully available: missing or out-of-range used_percent values and account-preparation failures are persisted as refresh errors instead of producing misleading remaining quota or bypassing quota protection.