<!-- qwen-release-notes:v1 -->
## 亮点
- 新增 ACP 支持，可通过会话配置将推理强度从默认（Default）调整至最大（Max）。 ([#8526](https://github.com/QwenLM/qwen-code/pull/8526))
- 在 Web Shell 中点击上传或粘贴的图片，现在会在工件面板中打开预览，并支持下载。 ([#8930](https://github.com/QwenLM/qwen-code/pull/8930))
- 修复 AutoFix 超时问题：通过启用流式进度输出，以便在长时间运行的无头任务期间检测活跃工作。 ([#8895](https://github.com/QwenLM/qwen-code/pull/8895))
- 修复桌面端本地控制功能：通过允许在已验证接口上使用非私有 IPv4 路由，支持企业局域网地址。 ([#8866](https://github.com/QwenLM/qwen-code/pull/8866))
- 提升 Web Shell 响应速度：修复动画卡顿、滚动稳定性问题，并优化大文件的行数变更统计。 ([#8915](https://github.com/QwenLM/qwen-code/pull/8915), [#8914](https://github.com/QwenLM/qwen-code/pull/8914), [#8924](https://github.com/QwenLM/qwen-code/pull/8924))
- 修复 Windows 平台 Bug：桌面端应用因未处理原始路径前缀导致 Node 运行时无法启动。 ([#8936](https://github.com/QwenLM/qwen-code/pull/8936))
## 破坏性更新
无已知破坏性更新。
## 完整变更列表
### 新功能
- OpenAI API 日志清理现在适用于非交互式会话、无头调用和守护进程模式，以减少磁盘占用。 ([#8893](https://github.com/QwenLM/qwen-code/pull/8893)) by @doudouOUC
- Web Shell 会话列表现在跨组件共享缓存数据，以提升加载性能并减少网络请求。 ([#8891](https://github.com/QwenLM/qwen-code/pull/8891)) by @doudouOUC
- 新增 ACP 支持，允许客户端通过会话配置将推理强度选项从默认（Default）设置至最大（Max）。 ([#8526](https://github.com/QwenLM/qwen-code/pull/8526)) by @zjunothing
- 在 Web Shell 中点击上传或粘贴的图片，现在会在右侧工件面板中打开预览，并支持下载。 ([#8930](https://github.com/QwenLM/qwen-code/pull/8930)) by @ytahdn
### 缺陷修复
- 修复 AutoFix 超时问题：启用流式进度输出，使系统能在长时间运行的无头任务期间检测活跃工作。 ([#8895](https://github.com/QwenLM/qwen-code/pull/8895)) by @qqqys
- 更改 `@` 补全功能：使用左右方向键直接切换分类标签，替代原来的 `Ctrl+方向键` 或 `Ctrl+Tab`。 ([#8576](https://github.com/QwenLM/qwen-code/pull/8576)) by @LaZzyMan
- 修正圆帽标记的生命周期，防止出现矛盾的停止提示，并确保在收敛时标记能正确清除。 ([#8850](https://github.com/QwenLM/qwen-code/pull/8850)) by @wenshao
- 提升 Web Shell 子代理面板响应速度：修复动画卡顿、已用时间重置及输出滚动行为。 ([#8915](https://github.com/QwenLM/qwen-code/pull/8915)) by @ytahdn
- 修复会话加载超时 Bug：通过清除过期的目标标识，解决无法重试同一会话的问题。 ([#8883](https://github.com/QwenLM/qwen-code/pull/8883)) by @yiliang114
- 恢复 Live Host 版本镜像至阿里云 OSS：修正工作流权限和凭证传递。 ([#8917](https://github.com/QwenLM/qwen-code/pull/8917)) by @LaZzyMan
- 当状态栏已显示上下文信息时，UI 中不再重复显示上下文使用指示器。 ([#8749](https://github.com/QwenLM/qwen-code/pull/8749)) by @yiliang114
- 聊天输出中的超链接现在能正确在中文全角标点处截断，防止中文文本中的 URL 断裂。 ([#8755](https://github.com/QwenLM/qwen-code/pull/8755)) by @yiliang114
- Web Shell 现在能可靠地将中途消息与守护进程状态进行同步，防止刷新后重复提交。 ([#8798](https://github.com/QwenLM/qwen-code/pull/8798)) by @ytahdn
- 修复调整终端窗口大小或唤醒应用时出现的横幅重复和屏幕闪烁问题。 ([#8831](https://github.com/QwenLM/qwen-code/pull/8831)) by @chiga0
- 滚动长 Web Shell 记录时现在更加稳定，并在历史分页时保留可见消息。 ([#8914](https://github.com/QwenLM/qwen-code/pull/8914)) by @ytahdn
- 使 WebUI 跨会话切换具备事务性，防止在恢复会话失败或超时时发生数据丢失或状态损坏。 ([#8882](https://github.com/QwenLM/qwen-code/pull/8882)) by @doudouOUC
- 修复桌面端本地控制功能：通过允许在已验证的物理接口上使用非私有 IPv4 路由，支持企业局域网地址。 ([#8866](https://github.com/QwenLM/qwen-code/pull/8866)) by @yiliang114
- 修复 Web Shell：通过用线性时间算法替换之前的比较限制，正确显示大编辑文件的行数变更统计。 ([#8924](https://github.com/QwenLM/qwen-code/pull/8924)) by @ytahdn
- 修复延迟 MCP 工具在恢复会话中缺失的问题：在工具声明刷新时重新显示它们。 ([#8475](https://github.com/QwenLM/qwen-code/pull/8475)) by @zjunothing
- 修复 CLI VP 模式：按下 `Ctrl+S` 时完全展开静态历史记录项，确保长表格和输出不再被截断。 ([#8704](https://github.com/QwenLM/qwen-code/pull/8704)) by @zjunothing
- 自动会话标题现在排除钩子上下文，防止冗长的通用模块挤占用户的实际请求。 ([#8781](https://github.com/QwenLM/qwen-code/pull/8781)) by @zjunothing
- 修复 Windows 平台 Bug：桌面端应用因未处理原始路径前缀导致 Node 运行时无法启动。 ([#8936](https://github.com/QwenLM/qwen-code/pull/8936)) by @yiliang114
- 新增会话关闭的结构化错误码，以确保 Web UI 和 Java 客户端能可靠地处理重试。 ([#8884](https://github.com/QwenLM/qwen-code/pull/8884)) by @yiliang114
- 修复了 provider 版本同步问题，以正确区分已安装模型的哈希值与模板刷新状态。([#8889](https://github.com/QwenLM/qwen-code/pull/8889)) by @yiliang114
### 性能
- 通过缓存已持久化的会话目录来优化 CLI 性能，用于组织和过滤守护进程列表，并在元数据变更时自动失效。([#8892](https://github.com/QwenLM/qwen-code/pull/8892)) by @doudouOUC
### 文档
- 更新文档，修正了内联图片测试的会话恢复和屏幕阅读器验证步骤。([#8746](https://github.com/QwenLM/qwen-code/pull/8746)) by @zjunothing
### 内部变更
- 新增回归测试，以确保在重试和通知回合中，/remember 命令的上下文刷新标记能正确持久化。([#8809](https://github.com/QwenLM/qwen-code/pull/8809)) by @ZijianZhang989
- 重构了 qwen review build-test 的内部工具链代码，未更改任何 CLI 参数、输出或 npm 行为。([#8776](https://github.com/QwenLM/qwen-code/pull/8776)) by @wenshao
- 无面向用户的变更；此更新提升了在 LLM 超时时退出计划模式的测试稳定性。([#8881](https://github.com/QwenLM/qwen-code/pull/8881)) by @yiliang114
- 修复了 Web Shell 测试框架中的无限渲染循环，该问题此前会导致 CI 作业因堆内存耗尽而失败。([#8934](https://github.com/QwenLM/qwen-code/pull/8934)) by @ytahdn
**完整更新日志**: https://github.com/QwenLM/qwen-code/compare/v0.21.9...v0.21.10
<!-- qwen-release-notes:v1 -->
## Highlights
- Added ACP support for configuring reasoning effort levels from Default to Max via session configuration. ([#8526](https://github.com/QwenLM/qwen-code/pull/8526))
- Clicking uploaded or pasted images in the Web Shell now opens a preview in the artifact panel with download support. ([#8930](https://github.com/QwenLM/qwen-code/pull/8930))
- Fixed AutoFix timeouts by enabling streamed progress output to detect active work during long-running headless tasks. ([#8895](https://github.com/QwenLM/qwen-code/pull/8895))
- Fixed Desktop Local Control to support enterprise LAN addresses by allowing non-private IPv4 routes on verified interfaces. ([#8866](https://github.com/QwenLM/qwen-code/pull/8866))
- Improved Web Shell responsiveness by fixing animation stutter, scrolling stability, and line-change statistics for large files. ([#8915](https://github.com/QwenLM/qwen-code/pull/8915), [#8914](https://github.com/QwenLM/qwen-code/pull/8914), [#8924](https://github.com/QwenLM/qwen-code/pull/8924))
- Fixed Windows bug where the desktop app failed to spawn the Node runtime due to unhandled verbatim path prefixes. ([#8936](https://github.com/QwenLM/qwen-code/pull/8936))
## Breaking Changes
No known breaking changes.
## Complete Change List
### Features
- OpenAI API log cleanup now applies to non-interactive sessions, headless invocations, and daemon modes to reduce disk usage. ([#8893](https://github.com/QwenLM/qwen-code/pull/8893)) by @doudouOUC
- Web Shell session lists now share cached data across components to improve loading performance and reduce network requests. ([#8891](https://github.com/QwenLM/qwen-code/pull/8891)) by @doudouOUC
- Added ACP support for configuring reasoning effort levels, allowing clients to set options from Default to Max via session configuration. ([#8526](https://github.com/QwenLM/qwen-code/pull/8526)) by @zjunothing
- Clicking uploaded or pasted images in the Web Shell now opens a preview in the right-hand artifact panel with download support. ([#8930](https://github.com/QwenLM/qwen-code/pull/8930)) by @ytahdn
### Bug Fixes
- Fixed AutoFix timeouts by enabling streamed progress output so the system can detect active work during long-running headless tasks. ([#8895](https://github.com/QwenLM/qwen-code/pull/8895)) by @qqqys
- Changed @ completion to use bare Left and Right arrow keys for switching category tabs instead of Ctrl+arrow or Ctrl+Tab. ([#8576](https://github.com/QwenLM/qwen-code/pull/8576)) by @LaZzyMan
- Corrected the round-cap marker lifecycle to prevent contradictory stop disclosures and ensure markers clear properly upon convergence. ([#8850](https://github.com/QwenLM/qwen-code/pull/8850)) by @wenshao
- Improved Web Shell subagent panel responsiveness by fixing animation stutter, elapsed time resets, and output scrolling behavior. ([#8915](https://github.com/QwenLM/qwen-code/pull/8915)) by @ytahdn
- Fixed a bug where session load timeouts prevented retrying the same session by clearing the stale target identity upon expiration. ([#8883](https://github.com/QwenLM/qwen-code/pull/8883)) by @yiliang114
- Restored Live Host release mirroring to Aliyun OSS by correcting workflow permissions and credential passing. ([#8917](https://github.com/QwenLM/qwen-code/pull/8917)) by @LaZzyMan
- Context usage indicators no longer appear twice in the UI when the status line already displays context information. ([#8749](https://github.com/QwenLM/qwen-code/pull/8749)) by @yiliang114
- Hyperlinks in chat output now correctly stop at full-width CJK punctuation to prevent broken URLs in Chinese text. ([#8755](https://github.com/QwenLM/qwen-code/pull/8755)) by @yiliang114
- Web Shell now reliably reconciles mid-turn messages with daemon state to prevent duplicate submissions after refreshes. ([#8798](https://github.com/QwenLM/qwen-code/pull/8798)) by @ytahdn
- Fixed banner duplication and screen flickering issues when resizing the terminal window or waking the application. ([#8831](https://github.com/QwenLM/qwen-code/pull/8831)) by @chiga0
- Scrolling through long Web Shell transcripts is now stable and preserves visible messages during history pagination. ([#8914](https://github.com/QwenLM/qwen-code/pull/8914)) by @ytahdn
- Made WebUI cross-session switching transactional to prevent data loss or state corruption when resuming sessions fails or times out. ([#8882](https://github.com/QwenLM/qwen-code/pull/8882)) by @doudouOUC
- Fixed Desktop Local Control to support enterprise LAN addresses by allowing non-private IPv4 routes on verified physical interfaces. ([#8866](https://github.com/QwenLM/qwen-code/pull/8866)) by @yiliang114
- Fixed Web Shell to correctly display line-change statistics for large edited files by replacing the previous comparison limit with a linear-time algorithm. ([#8924](https://github.com/QwenLM/qwen-code/pull/8924)) by @ytahdn
- Fixed an issue where deferred MCP tools were missing in resumed sessions by re-revealing them when tool declarations are refreshed. ([#8475](https://github.com/QwenLM/qwen-code/pull/8475)) by @zjunothing
- Fixed CLI VP mode to fully expand static history items when pressing Ctrl+S, ensuring long tables and output are no longer truncated. ([#8704](https://github.com/QwenLM/qwen-code/pull/8704)) by @zjunothing
- Automatic session titles now exclude hook context to prevent long generic blocks from displacing the user's actual request. ([#8781](https://github.com/QwenLM/qwen-code/pull/8781)) by @zjunothing
- Fixed a Windows bug where the desktop app failed to spawn the Node runtime due to unhandled verbatim path prefixes. ([#8936](https://github.com/QwenLM/qwen-code/pull/8936)) by @yiliang114
- Added a structured error code for closing sessions to ensure reliable retry handling in the Web UI and Java clients. ([#8884](https://github.com/QwenLM/qwen-code/pull/8884)) by @yiliang114
- Fixed provider version synchronization to correctly distinguish between installed model hashes and template refresh states. ([#8889](https://github.com/QwenLM/qwen-code/pull/8889)) by @yiliang114
### Performance
- Improved CLI performance by caching persisted session catalogs for organized and filtered daemon lists, with automatic invalidation on metadata changes. ([#8892](https://github.com/QwenLM/qwen-code/pull/8892)) by @doudouOUC
### Documentation
- Updated documentation to correct session resume and screen-reader verification steps for inline image testing. ([#8746](https://github.com/QwenLM/qwen-code/pull/8746)) by @zjunothing
### Internal Changes
- Added regression tests to ensure the /remember command context refresh marker persists correctly during retry and notification turns. ([#8809](https://github.com/QwenLM/qwen-code/pull/8809)) by @ZijianZhang989
- Refactored internal toolchain code for qwen review build-test without changing any CLI arguments, outputs, or npm behaviors. ([#8776](https://github.com/QwenLM/qwen-code/pull/8776)) by @wenshao
- No user-facing changes; this update improves test stability for plan mode exits during LLM timeouts. ([#8881](https://github.com/QwenLM/qwen-code/pull/8881)) by @yiliang114
- Fixed an infinite render loop in the Web Shell test harness that previously caused CI jobs to fail due to heap exhaustion. ([#8934](https://github.com/QwenLM/qwen-code/pull/8934)) by @ytahdn
**Full Changelog**: https://github.com/QwenLM/qwen-code/compare/v0.21.9...v0.21.10