jlcodes99/cockpit-tools v1.3.59

Cockpit Tools v1.3.59

on GitHub

更新日志（中文）

[1.3.59] - 2026-09-23

新增

• 内置模型新增 GPT-6 Sol 与 GPT-6 Luna：账号模型清单、API 服务模型选择器与唤醒模型预设都会带出这两个官方模型，顺序与官方客户端一致（排在 GPT-6 Astra 之后、GPT-5.6 系列之前）。API 服务的费用估算按官方公开价目计入标准档与快速档，长上下文按与其它 6 代模型相同的倍数计算。模型清单仍跟随官方推荐集的用户会自动补齐这两个模型，自己改过清单的用户不受影响。
• 账号额度行新增「≈$」满额折算金额：5h / Weekly 额度行在「A $已花费」之后补一个 ≈$ 数值，按当前已用比例把本窗口费用折算成「满额约合多少」，方便判断这段额度的实际价值；窗口内没有费用或还没有用量时不显示，原有额度、进度条与重置时间都不受影响。
• DeepSeek 模型目录补齐 V2 协作能力声明：API 服务与 DeepSeek 目录在模板覆盖后仍保留声明，旧目录通过生成版本检查自动重建；GPT 与 Grok 的声明保持不变。
• 恢复鹈鹕测智：Codex 账号可再次向所选账号并行发起统一提示词测试，查看生成结果、HTML 预览与并排对比，重试失败账号，并按保留天数自动或手动清理测试记录与生成文件；测试请求重新走本地 API 服务的内部调度链路，与唤醒共用账号并发、Token 与额度冷却。

变更

• 内建模型显示名统一带 GPT- 前缀：唤醒与鹈鹕测智的模型预设、模型管理列表和受管模型目录现在统一显示为 GPT-6 Astra、GPT-6 Sol、GPT-6 Luna、GPT-5.6 Sol、GPT-5.6 Terra、GPT-5.6 Luna、GPT-5.5；升级后旧短名（6 Astra、5.6 Sol 等）会自动补上前缀，模型 ID 与你手动改过的展示名保持不变。
• 账号「重新授权」默认打开官方登录并显示账号信息：账号卡片、表格行与账号异常弹框里的重新授权入口现在直接进入「官方登录」，弹框顶部固定显示目标账号邮箱与备注入口，重新授权时也能直接查看或编辑备注。
• 移除账号风控检测（已失效）：账号总览不再提供「风控检测」入口与检测弹框，账号卡片/表格不再显示「疑似风控」「未知」等状态标签，API 服务请求日志也不再标注 state 长度与分级；唤醒与网关转发不再旁路记录 state 观测。已有本地观测数据不做删除，仅停止继续采集与展示。
• CodeBuddy 账号总览移除「本机会话文件」列表：会话查看统一走「会话」页签的会话管理器，总览页不再在账号列表下方显示本地 JSON/JSONL 文件列表。
• 账号总览移除 Codex 速度切换：账号卡片、清单行、账号操作菜单与 API 服务卡片不再显示「标准 / 快速」档位选择。
• macOS DMG 安装窗口改为两行布局：Cockpit Tools 与 Applications 别名保持同一行，拖拽安装的动线不变；说明文件改名为「安装与常见问题.txt」并下移到第二行，内容先讲安装步骤，再讲“应用已损坏”的处理，并把不需要密码的「系统设置 → 仍要打开」放在前面，命令行方式作为备选。

修复

• 修复重新授权后本地网关仍用旧凭据：重新授权或本机导入完成后会立刻把新 Token 写穿到 API 服务 sidecar，并在后台重置该账号的调度与失败状态，不再等到下一次整体重载才生效，界面也不会继续显示上一次的失败原因。
• 修复 API 服务账号「恢复」在凭据失效后没有反应：恢复前会先强制刷新该账号的 Token 并同步给本地网关；凭据已失效（上游 401 invalidated oauth token）时账号异常弹框改为提供「重新授权」，不再只提示已提交却无法恢复。
• 修复一次性历史清理在大历史库上的内存峰值：清理第三方推理历史时改为分批次扫描并逐批提交，回滚备份改为边扫描边落盘，内存占用不再随命中的行数线性增长；备份文件结构与覆盖范围不变，仍可手动回滚。

Changelog (English)

[1.3.59] - 2026-09-23

Added

• Two new built-in models: GPT-6 Sol and GPT-6 Luna: the account model catalog, the API-service model picker and wakeup model presets now include both official models, ordered like the official client (right after GPT-6 Astra and before the GPT-5.6 family). API-service cost estimates use the published per-token rates for Standard and Fast mode, with the same long-context multipliers as the other GPT-6 models. Users whose catalog still follows the official recommended set get both models added automatically, while catalogs you edited yourself are left untouched.

• Quota rows now show an estimated full-window cost: the 5h and Weekly quota rows add an ≈$ figure right after the A $ amount, scaling this window's cost by the usage percentage so you can judge what the quota is worth. Nothing is shown when the window has no cost or no usage yet, and the existing quota, progress bar and reset time are unchanged.

• DeepSeek models declare V2 collaboration capability in managed model catalogs: API-service and DeepSeek catalogs retain the declaration after template overrides. Existing catalogs rebuild through the generator-version check; GPT and Grok declarations are unchanged.

• Pelican testing is back: Codex accounts can again run one shared prompt across selected accounts in parallel, inspect the generated results, HTML previews and side-by-side comparison, retry failed accounts, and clean up test records and generated files automatically by retention days or manually. Test requests go through the local API Service internal scheduler again, sharing account concurrency, token handling and quota cooldowns with wakeups.

Changed

• Built-in model labels now always carry the GPT- prefix: wakeup and Pelican model presets, the model-management list and the managed model catalog now consistently show GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna and GPT-5.5. Upgrading adds the prefix to the old short labels (6 Astra, 5.6 Sol, …) automatically, while model IDs and any display name you edited yourself stay unchanged.

• Account re-authorization now opens the official login and keeps the account info visible: the re-authorize entries on account cards, table rows and the account-issue dialog go straight to the official login tab, which now always shows the target account email and its note button, so the note can be checked or edited while re-authorizing.

• Account risk detection has been removed: the accounts overview no longer offers the risk-check entry or its dialog, account cards and table rows no longer show status labels such as Suspected risk or Unknown, and the API Service request log no longer annotates state lengths or classes. Wakeups and gateway forwarding no longer record state observations on the side. Existing local observation data is not deleted; collection and display simply stop.

• The CodeBuddy accounts overview no longer lists local session files: session browsing stays in the Sessions tab, so the overview no longer shows the local JSON/JSONL file list under the account list.

• The Codex speed switch has been removed from the accounts overview: account cards, list rows, the account action menu and the API service card no longer show the Standard/Fast tier picker.

• The macOS DMG installer window now uses a two-row layout: Cockpit Tools and the Applications link stay on the same row so the drag-to-install gesture is unchanged, while the note is renamed to 安装与常见问题.txt and moved to a second row below them. The note covers installation first, then the "app is damaged" fix, recommending the no-password System Settings "Open Anyway" path with the Terminal command as the fallback.

Fixed

• Fixed the local gateway keeping stale credentials after re-authorization: a completed re-authorization or local import now writes the fresh token into the API-service sidecar immediately and resets that account's scheduler and failure state in the background, instead of waiting for the next full reload, and the UI no longer keeps showing the previous failure.

• Fixed API-service account recovery doing nothing once credentials are invalid: recovery now forces a token refresh and pushes the fresh credentials to the local gateway first, and accounts whose credentials are invalid (upstream 401 invalidated oauth token) now offer re-authorization in the account-issue dialog instead of only reporting that recovery was submitted.

• Fixed the memory peak of the one-time history cleanup on large history databases: the third-party reasoning cleanup now scans and commits in batches, and the rollback backup is streamed to disk while scanning, so memory no longer grows with the number of matched rows. The backup file layout and coverage are unchanged and remain manually restorable.