feat(desktop): 支持分别配置任务命名与推荐指令模型 - #3307
Conversation
Signed-off-by: david <david@xd.com>
Signed-off-by: DavidShen <david@xd.com>
反馈已处理已处理：
变更：
验证：
备注：
|
Signed-off-by: DavidShen <david@xd.com>
|
已处理本轮 Windows 单测失败并推送修复 根因：测试在首个业务帧完成后统计所有携带 修复：在发送首个可靠业务帧前截取并断言独立确认重试，按“首包已丢弃，送达数不超过总尝试次数减一”校验；业务累计 ACK 不再进入该计数。 本地验证：
已推送到 |
Signed-off-by: DavidShen <david@xd.com>
Signed-off-by: DavidShen <david@xd.com>
Signed-off-by: DavidShen <david@xd.com>
Signed-off-by: DavidShen <david@xd.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 77f9802b73
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Signed-off-by: DavidShen <david@xd.com>
反馈已处理已处理：
变更：
验证：
备注：
|
|
| Filename | Overview |
|---|---|
| apps/desktop/src/main/utility-model/auxiliary-model-settings-store.ts | 新增 owner-scoped 辅助模型偏好存储，并通过共享原子设置文件机制保存两个独立字段。 |
| apps/desktop/src/main/utility-model/oneShotCandidates.ts | 扩展一次性文本请求能力，以支持固定 provider、agent 与 model 的精确路由及发送前复核。 |
| apps/desktop/src/main/maker-host/auxiliary-title-one-shot.ts | 新增统一任务命名入口；自动模式保留历史行为，显式模式严格使用所选模型并 fail closed。 |
| apps/desktop/src/main/maker-ipc/promptPrediction.ts | 推荐指令生成接入独立辅助模型配置，同时保留会话修订、停止状态和发送前有效性检查。 |
| apps/desktop/src/main/maker-ipc/auxiliary-model-settings.ts | 新增可信 IPC 边界、运行期 payload 校验，以及可用和已保存但暂不可用模型的列表构建。 |
| apps/desktop/src/renderer/components/settings/AuxiliaryModelSection.tsx | 在个性化设置中提供两个相互独立的辅助模型选择器及加载、保存和错误状态。 |
| apps/desktop/src/preload/preload.ts | 为主 Renderer 暴露窄化的辅助模型设置读取与更新接口。 |
| apps/desktop/src/shared/auxiliaryModelSettings.ts | 定义 Main、Preload 与 Renderer 共用的辅助模型设置和选项数据契约。 |
Sequence Diagram
sequenceDiagram
participant U as 用户
participant R as 设置界面
participant P as Preload
participant M as Main IPC
participant S as Owner-scoped 设置存储
participant A as 辅助模型路由
participant V as 模型 Provider
U->>R: 分别选择命名/推荐模型
R->>P: 提交 catalog pin
P->>M: 调用受限 IPC
M->>M: 校验 sender、payload 与可路由模型
M->>S: 原子保存当前 owner 的选择
S-->>M: 返回最新设置
M-->>R: 返回设置与可用模型列表
R->>A: 后续触发命名或推荐
A->>S: 读取对应的独立选择
A->>A: 固定 provider/agent/model 并复核选择
A->>V: 发送精确模型请求
V-->>A: 返回结果或失败
A-->>R: 返回标题/推荐；显式路由失败时不切换 Provider
Reviews (3): Last reviewed commit: "fix(desktop): 禁用推荐指令模型的短预算推理" | Re-trigger Greptile
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f8b4f8fe98
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Signed-off-by: DavidShen <david@xd.com>
反馈已处理已处理：
变更：
验证：
备注：
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: df5f8dadca
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
MagicLizi
added
status:awaiting-bot-review
Signed-off-by: DavidShen <david@xd.com>
反馈已处理已处理：
变更：
验证：
备注：
|
2 participants
这次改了什么
摘要
在「设置 → 个性化」新增「辅助模型」配置，让用户可以分别选择自动任务命名和聊天输入框推荐指令所使用的模型。
自动命名改为通用能力，不再绑定 Claude Code：Claude Code、Codex、Pi、IM 任务及手动重新命名都走统一的辅助模型路由。用户显式选择模型后，系统严格使用对应的 provider / agent / model；路由不可用时直接失败，不静默切换到其它提供方。未配置时保持原有自动选择行为。
变更类型
feat新功能fix缺陷修复refactor/perf重构或性能优化docs/test/chore文档、测试或工程维护范围
UI 变化
docs/design-rules/DESIGN.md§4 Inputs & Forms：复用现有模型选择器与表单交互docs/design-rules/DESIGN.md§5 Spacing System：沿用设置页既有间距和布局层级docs/design-rules/DESIGN.md§10 Light / Dark Dual-Mode Delivery Gate：颜色全部使用语义 token，并完成双模式实机目检docs/design-rules/DESIGN.md§11 Voice & Content：说明文案保持简洁，并同步五种语言怎么验证的
自动验证
手工验证
--isolated=@worktree启动隔离 dev 沙盒并通过真实 UI 跳过登录未执行的验证
风险
风险分类
影响与回滚
提交前检查
git commit -s，见 DCO）关联 Issue