更新日志（中文）
[1.3.45] - 2026-09-08
修复
- 修复 DCP 注入额度与 Cockpit 账号池状态不一致：额度池快照现在携带账号套餐信息；账号文件短暂刷新或读取失败时保留上一份有效额度，只有确认账号池确实为空时才显示
0/0，避免额度区域在正常数据和“暂无额度统计”之间闪烁。 - 修复 Windows 刷新 Codex 额度时的栈溢出闪退：降低额度刷新对主线程栈的占用，避免手动或自动刷新额度时主程序意外退出；保留历史额度和原有刷新后处理行为。
- 仅在“统计与日志”页签显示用量统计区域：服务总览、客户端 Key、账号池和模型与能力页签不再重复显示统计范围、指标卡和趋势图。
- 修复账号异常弹框恢复成功后看似没有反应：恢复结果会直接显示在当前弹框内；全部恢复成功后会清除旧 Sidecar 遗留的聚合账号池异常，单账号恢复仍保留其他账号的异常状态。
变更
- Codex 额度刷新优先响应手动操作并限制资源占用：手动与后台刷新各有独立通道，各最多执行 1 个账号、排队 16 个账号，同一账号的重复请求共享一次刷新；手动刷新会将本轮尚未开始的后台任务延后至下一轮，不因插队中断已开始的请求。排队和执行分别设置 2 分钟超时，任务结束后释放任务资源及等待者，并提供本地化的繁忙、超时和异常提示。
- 仅为已开启混合路由的实例运行后台监控：启用混合路由后启动监控，关闭后停止对应后台检查和自动恢复，不影响其他实例及独立的 API 服务。关闭路由不打断正在运行的 Codex 会话，配置变更在下次通过 Cockpit 启动时生效。
- 恢复 API 服务进程监控和受保护的自动恢复：服务启用时检测 Sidecar 意外退出，并在原有限制内尝试恢复；清理端口前校验 Sidecar 命令行和父进程归属，保留手动重启、请求重试和账号池故障切换。
- 恢复退出时的混合路由配置清理：退出 Cockpit 时恢复未运行的历史受管实例配置，保留仍由运行中 Codex 会话使用的配置；未使用混合路由的普通实例不参与对应的进程检查。
Changelog (English)
[1.3.45] - 2026-09-08
Fixed
- Keep DCP-injected quota data aligned with the Cockpit account pool: quota snapshots now carry account plan types, transient account-file refresh or read failures preserve the last valid quota, and
0/0is shown only when the account pool is confirmed empty. This prevents the injected quota panel from flickering between valid data and “no quota stats”. - Prevent Windows crashes from stack overflow during Codex quota refreshes: reduce main-thread stack usage during quota refreshes to prevent unexpected application exits during manual or automatic refreshes. Quota history and existing post-refresh behavior are preserved.
- Show the usage statistics area only on the Statistics & Logs tab: the Service Overview, Client Key, Account Pool, and Models & Capabilities tabs no longer repeat the range picker, metric cards, or trend chart.
- Show account recovery feedback inside the account-issues dialog: successful recovery now reports its result in the active dialog. Recovering the full collection also clears stale aggregate pool diagnostics left by older Sidecars, while single-account recovery preserves unrelated account issues.
Changed
- Prioritize manual Codex quota refreshes with bounded resource usage: manual and background refreshes have independent lanes, each with at most one active account and sixteen queued accounts; duplicate requests for the same account share one refresh. Manual refreshes defer not-yet-started background work from the current round until the next round without interrupting requests already running. Queue waits and execution each have a two-minute timeout, task resources and waiters are released on completion, and busy, timeout, and interruption messages are localized.
- Run mixed-routing monitoring only for enabled instances: enable monitoring when mixed routing is configured, stop its background checks and automatic recovery when disabled, and leave unrelated instances and the independent API Service untouched. Disabling a route does not interrupt an active Codex session; its configuration changes take effect on the next launch through Cockpit.
- Restore API Service process monitoring and guarded recovery: detect unexpected Sidecar exits and retry recovery within the existing limits while the service is enabled. Port cleanup checks the Sidecar command line and parent ownership before stopping a process; manual restart, request retries, and account-pool failover remain available.
- Restore mixed-routing configuration cleanup on exit: restore inactive previously managed profiles on Cockpit exit while preserving profiles still used by running Codex sessions. Ordinary profiles without mixed routing are not included in its process checks.