jlcodes99/cockpit-tools v0.21.0 on GitHub

更新日志（中文）

Codex 现已新增独立的“模型供应商”工作区，用于统一管理 API Key 账号的供应商与密钥复用：可集中维护兼容供应商与多个 API Key，在新增或编辑 Codex API Key 账号时直接复用，并可在账号页内把现有 API Key 账号快速切换到已保存的供应商与密钥。
界面语言现已新增印尼语（Bahasa Indonesia）：语言注册、设置页语言选择器与文档语言列表现已纳入印尼语。

Gemini CLI 启动现已支持配置默认终端，并可在启动弹框内直接拉起终端执行：可在设置页选择偏好的终端，切换实例后既可以复制启动命令，也可以直接从弹框内执行。
Codex 会话管理现已补充“一键修复历史可见性”能力：会按各实例根目录 config.toml 中的 model_provider 修复 rollout 文件与 state_5.sqlite 的 provider 元数据，并在写入前自动创建备份。
Windows 桌面端 WebSocket 现已允许来自 WSL 本地桥接网段的客户端连接：不再只接受 loopback，可直接接入运行在 WSL 内的本地插件或运行时客户端。

各平台本地账号持久化现已改为原子写入，并在可恢复的 JSON 解析失败时自动回滚备份：账号索引与详情文件会先写备份，再在检测到可恢复损坏时自动从 .bak 恢复，降低本地数据损坏风险。

Codex now includes a dedicated Model Providers workspace for API-key accounts: manage compatible providers and multiple API keys in one place, reuse them while adding or editing API-key accounts, and quick-switch existing API-key accounts to a saved provider/key pair directly from the account page.
Bahasa Indonesia is now available as a supported UI language: the locale registry, settings language picker, and documentation language list now include Indonesian.

Gemini CLI launch now supports a configurable default terminal plus direct in-terminal execution from the launch dialog: users can choose the preferred terminal in Settings, then copy the launch command or run it directly from the dialog after switching an instance.
Codex Session Manager now adds one-click historical visibility repair across instances: it repairs rollout files and state_5.sqlite provider metadata from each instance's root config.toml model_provider, and creates backups before writing.
Windows desktop WebSocket access now allows WSL-side clients through a detected local-network whitelist: Cockpit can now accept local plugin/runtime connections coming from WSL bridge networks instead of loopback only.

Local account persistence now uses atomic writes with backup-assisted recovery across providers: account index/detail JSON writes create backups first and can auto-restore from .bak files when a recoverable parse failure is detected, reducing local data corruption risk.