fix: auto-publish log showing unpublished, merge changelog, version to 1.0.0-beta.2

feat: streamline dashboard, fix docs, optimize UI width
- Remove AI角色 quick action, keep only 回复日志/插件设置/刷新数据 - Fix LogsView filter/batch bar width to match content area (mx-4 mt-2 -> m-4 mb-0) - Fix docs: Halo version 2.23->2.25, GitHub link nxxy335->sunny-335, Cravatar link, variable name {{conversation}}->{{conversation_history}} - Add missing settings docs (maxConversationRounds, rateLimitPerMinute, enabledPresets) - Update README: remove version display, update changelog - Update version to 1.0.0-beta.2-r8k4n2
2026-06-17 23:13:45 +08:00 · 2026-06-17 22:03:35 +08:00 · 2026-06-17 21:10:06 +08:00 · 2026-06-17 19:40:10 +08:00 · 2026-06-17 13:26:18 +08:00
25 changed files with 1035 additions and 899 deletions
@@ -15,18 +15,17 @@
 - **文章/页面级开关** — 在文章编辑器中直接控制是否启用 AI 回复，文章默认开启，页面默认关闭
 - **评论者黑名单** — 支持按名称、邮箱和正则表达式屏蔽指定评论者，可从评论列表选择
 - **手动触发** — 在评论管理页面对历史评论手动触发 AI 回复
- **安全审核** — AI 生成的内容经过安全审核，不合规内容自动拒绝
- **Prompt 模板** — 支持自定义 Prompt 模板，提供多种模板变量（文章标题、发布日期、评论数等）
+- **两阶段安全审核** — 安全检查 + 质量评分（1-5 分映射到 0-100 分），不合规内容自动拒绝
+- **Prompt 模板** — 支持自定义 Prompt 模板，提供多种模板变量（文章标题、发布日期、评论数、对话历史等）
 - **Prompt 预设** — 内置友好型、专业型、幽默型、简洁型预设风格，可多选组合
- **仪表盘统计** — 显示回复数、情感分布、每日回复趋势等图表，支持时间范围切换
 - **插件健康检查** — 实时检测 AI Foundation 连接状态和模型可用性
 - **日志筛选** — 按状态、情感筛选，关键词搜索
 - **数据清理** — 自动清理超过指定天数的旧记录
- **AI Foundation 集成** — 必须安装 Halo AI Foundation 插件，使用其提供的 AI 模型能力
+- **AI Foundation 集成** — 通过 Halo 官方推荐的 `ExtensionGetter` 获取 AI 服务，需安装 AI Foundation 插件

 ## 前置要求

- Halo 2.23+
+- Halo 2.25+
 - AI Foundation 插件（必须）

 ## 安装
@@ -37,6 +36,10 @@
 4. 选择下载的 `.jar` 文件上传
 5. 安装完成后启用插件

+::: tip
+安装本插件后，应用市场会推荐安装 AI Foundation 插件（本插件的必要依赖）。
+:::
+
 ## 从源码构建

 ```bash
@@ -66,6 +69,64 @@ pnpm dev

 完整文档请访问 [AI回评文档站](https://nxxy335.top/comment-ai-autopilot)

+## 更新日志
+
+### v1.0.0-beta.2
+
+**改进**
+
+- 通过 Halo 官方推荐的 `ExtensionGetter` 获取 AI Foundation 的 `AiModelService`，替换原先的跨 ClassLoader 反射调用方式
+- 在 `plugin.yaml` 中声明可选插件依赖 `ai-foundation?: "*"`，建立正确的插件依赖关系
+- 新增 `store.halo.run/recommended-apps` 注解，安装后可在应用市场推荐安装 AI Foundation 插件
+- 情感分析和内容审核改用 AI Foundation 结构化输出（`OutputSpec.choice`），分类更可靠
+- AI 调用改用 `GenerateTextRequest` 并设置 `maxRetries=2`，由 SDK 自动重试瞬时错误
+- AI 对话续接时自动获取之前的回复历史并注入到 Prompt 中，AI 能更好地理解对话上下文
+- 优化 AI 自审核评分机制：改为两阶段评估（安全检查 + 质量评分 1-5 分），评分映射到 0-100 分，替代原先的二值评分
+- 精简仪表盘：移除情感分布、近7日回复趋势、平均审核评分三个卡片，快捷操作精简为回复日志、插件设置、刷新数据
+- 重做设置页面：基本设置、AI角色设置、模型设置、Prompt设置、数据清理各为独立页面，通过标签栏切换
+- 优化设置页面布局：按钮统一排版并添加图标，侧边栏新增"未保存"状态指示器
+- 优化日志页面：评分增加等级标签（优秀/良好/一般/较差），筛选下拉框修复文本与箭头重叠
+- 优化 AI Foundation 状态提示宽度，与内容区宽度一致
+- 优化插件文档：修复版本要求、变量名、链接等错误
+
+**Bug 修复**
+
+- 修复 RateLimitService 清理线程在插件停止时未关闭导致线程泄漏
+- 修复内容审核提示词要求"重新生成"但代码未使用重新生成内容的问题
+- 修复设置页面按钮图标和文字未在同一行显示的问题
+- 修复设置页面标签栏无法点击切换的问题
+- 修复设置页面配置区域布局错误的问题
+
+### v1.0.0-beta.1
+
+**新功能**
+
+- 草稿模式：关闭"自动发布"后，AI 回复将保存为草稿，需站长审核后才发布
+- 多 AI 角色支持：支持配置多个 AI 虚拟角色，每个角色有独立的提示词、头像和模型
+- 提示词预设：内置多种回复风格预设（专业型、幽默型、简洁型等），可自由组合
+- 对话历史查看：支持查看 AI 回复的完整对话上下文
+
+**Bug 修复**
+
+- 修复草稿模式下审批失败（"AI回复已存在，无法重复发布"）的问题
+- 修复批量审批时同样的去重检查冲突问题
+- 修复 AI Foundation 不可用的问题（`PluginManager` 无法通过 Spring 依赖注入获取）
+- 修复 `DefaultSpringPlugin` 包级私有类反射访问权限问题
+- 修复 CI 构建失败（`gradlew` 缺少执行权限）
+
+**改进**
+
+- 审批逻辑优化：先查找已有 Reply 扩展再决定创建或更新
+- 移除不必要的 `AiFoundationConfiguration` 配置类
+- 前端 UI 优化：移除编辑功能、简化角色排序逻辑、清理无用代码
+
+### v0.0.3
+
+- 增强插件可靠性与可用性
+- 多 AI 角色支持
+- 草稿模式初步实现
+- 文档全面更新
+
 ## 许可证

-[GPL-3.0](./LICENSE) © 暖心向阳335
+[GPL-3.0](https://github.com/sunny-335/plugin-comment-ai-autopilot/blob/main/LICENSE) © 暖心向阳335
@@ -5,7 +5,7 @@ plugins {
 }

 group 'top.nxxy335.commentaiautopilot'
-version '1.0.0-beta.1'
+version '1.0.0-beta.2'

 repositories {
    mavenCentral()
@@ -0,0 +1,69 @@
+# 更新日志
+
+## v1.0.0-beta.2
+
+> 2026-06-17
+
+### 改进
+
+- **改用 ExtensionGetter 集成 AI Foundation**：通过 Halo 官方推荐的 `ExtensionGetter.getEnabledExtension(AiModelService.class)` 获取 AI 服务，替换原先的跨 ClassLoader 反射调用方式（[Issue #1](https://github.com/sunny-335/plugin-comment-ai-autopilot/issues/1)）
+- **声明插件依赖**：在 `plugin.yaml` 中声明可选插件依赖 `ai-foundation?: "*"`，建立正确的插件依赖关系，插件在未安装 AI Foundation 时仍可正常加载
+- **应用市场推荐**：新增 `store.halo.run/recommended-apps` 注解，安装本插件后可在应用市场推荐安装 AI Foundation 插件
+- **使用结构化输出**：情感分析和内容审核改用 AI Foundation 的 `OutputSpec.choice` 结构化输出，替换原先的字符串匹配解析，分类更可靠
+- **使用 GenerateTextRequest**：AI 调用改用 `GenerateTextRequest` 并设置 `maxRetries=2`，由 SDK 自动重试瞬时错误
+- **多轮对话上下文**：AI 对话续接时自动获取之前的回复历史并注入到 Prompt 中，AI 能更好地理解对话上下文
+- **优化 AI 自审核评分机制**：审核改为两阶段评估（安全检查 + 质量评分 1-5 分），评分映射到 0-100 分（0/30/50/70/85/100），替代原先的二值评分（0/100），评分更有区分度
+- **精简仪表盘**：移除情感分布、近7日回复趋势、平均审核评分三个卡片，快捷操作精简为回复日志、插件设置、刷新数据
+- **重做设置页面**：基本设置、AI角色设置、模型设置、Prompt设置、数据清理各为独立页面，通过标签栏切换
+- **优化设置页面布局**：按钮统一排版并添加图标，侧边栏保存卡片高亮显示，新增"未保存"状态指示器
+- **优化日志页面**：评分增加等级标签（优秀/良好/一般/较差），筛选下拉框修复文本与箭头重叠
+- **优化 AI Foundation 状态提示宽度**：状态提示与内容区宽度一致
+- **优化插件文档**：修复版本要求（2.23→2.25）、变量名（`{{conversation}}`→`{{conversation_history}}`）、GitHub 链接（`nxxy335`→`sunny-335`）、Cravatar 链接（`cravatar.cn`→`cn.cravatar.com`）等错误，补充缺失的配置项文档
+
+### Bug 修复
+
+- **修复自动发布后日志显示未发布**：`generateAndPublish` 中 `Mono<Void>` 的 empty 信号触发 `switchIfEmpty` 导致 `publishReply` 被调用两次，第二次覆盖 `published=false`
+- **修复 RateLimitService 线程泄漏**：清理线程未在插件停止时关闭，实现 `DisposableBean` 正确释放资源
+- **修复 ReviewService 提示词不匹配**：审核提示词要求"重新生成"但代码未使用重新生成的内容，移除误导性指令
+- **修复设置页面按钮图标文字对齐**：通过 `:deep(.btn-content)` 设置 inline-flex 布局，图标和文字并排显示
+- **修复设置页面标签栏无法点击切换**：替换不工作的 VTabbar 组件为自定义按钮实现
+- **修复设置页面配置区域布局错误**：将标签栏移出 grid 容器，避免挤占配置区域宽度
+- **修复日志页面筛选下拉框文本与箭头重叠**：将 `px-3` 改为 `pl-3 pr-8` 为下拉箭头预留空间
+
+---
+
+## v1.0.0-beta.1
+
+> 2026-06-05
+
+### 新功能
+
+- **自动回复**：监听新评论，自动调用 AI 生成回复
+- **多语言适配**：根据评论语言自动用对应语言回复
+- **情感分析**：分析评论情感倾向，根据情感调整回复语气
+- **草稿模式**：关闭"自动发布"后，AI 回复将保存为草稿，需站长审核后才发布
+- **多 AI 角色支持**：支持配置多个 AI 虚拟角色，每个角色有独立的提示词、头像和模型
+- **提示词预设**：内置多种回复风格预设（专业型、幽默型、简洁型等），可自由组合
+- **文章/页面级开关**：在文章编辑器中直接控制是否启用 AI 回复
+- **评论者黑名单**：屏蔽指定评论者，不触发 AI 回复
+- **安全审核**：AI 生成的内容经过安全审核，不合规内容自动拒绝
+- **Prompt 模板**：支持自定义 Prompt 模板，提供多种模板变量
+- **仪表盘统计**：显示回复数等统计信息
+- **日志筛选搜索**：按状态、情感筛选，关键词搜索
+- **数据清理**：自动清理超过指定天数的旧记录
+- **AI Foundation 集成**：使用 AI Foundation 插件提供的 AI 模型能力
+- **对话历史查看**：支持查看 AI 回复的完整对话上下文
+
+### Bug 修复
+
+- 修复草稿模式下审批失败（"AI回复已存在，无法重复发布"）的问题
+- 修复批量审批时同样的去重检查冲突问题
+- 修复 AI Foundation 不可用的问题（`PluginManager` 无法通过 Spring 依赖注入获取）
+- 修复 `DefaultSpringPlugin` 包级私有类反射访问权限问题
+- 修复 CI 构建失败（`gradlew` 缺少执行权限）
+
+### 改进
+
+- 审批逻辑优化：先查找已有 Reply 扩展再决定创建或更新
+- 移除不必要的 `AiFoundationConfiguration` 配置类
+- 前端 UI 优化：移除编辑功能、简化角色排序逻辑、清理无用代码
@@ -31,6 +31,8 @@
 |--------|------|--------|
 | 自动回复 | 是否启用自动回复功能 | 开启 |
 | 自动发布 | AI回复是否自动发布，关闭则存为草稿 | 开启 |
+| 最大对话轮次 | 同一评论线程中AI最多自动回复的轮次 | 8 |
+| 速率限制 | 每分钟最大AI回复数量 | 10 |
 | 最大重试次数 | AI生成失败时的最大重试次数 | 3 |

 ## 重试机制
@@ -12,7 +12,7 @@

 1. 确认在插件设置中填写了AI角色邮箱
 2. 邮箱需要在 [Gravatar](https://gravatar.com) 上注册并设置头像
-3. 插件使用 [Cravatar](https://cravatar.cn) 作为Gravatar镜像服务
+3. 插件使用 [Cravatar](https://cn.cravatar.com) 作为Gravatar镜像服务

 ## 评论没有触发AI回复？

@@ -32,9 +32,17 @@

 1. 进入插件设置页面
 2. 在 **基本设置** 中找到 **评论者黑名单**
-3. 输入评论者的显示名称或邮箱，多个用逗号分隔
+3. 输入评论者的显示名称、邮箱或正则表达式，多个用逗号分隔
 4. 保存设置

+### 支持的格式
+
+| 格式 | 示例 | 说明 |
+|------|------|------|
+| 名称 | `张三` | 匹配评论者的显示名称 |
+| 邮箱 | `spam@example.com` | 匹配评论者邮箱（不区分大小写） |
+| 正则表达式 | `regex:^spam.*` | 以 `regex:` 开头，按正则匹配 |
+
 ### 从评论列表选择

 1. 在黑名单输入框旁点击 **添加评论者** 按钮
@@ -45,7 +53,7 @@
 ### 示例

 ```
-张三,spam@example.com,李四
+张三, spam@example.com, 李四, regex:^spam.*
 ```

-黑名单中的评论者发布评论时，插件会同时匹配显示名称和邮箱地址（不区分大小写），匹配成功则跳过AI回复。
+黑名单中的评论者发布评论时，插件会同时匹配显示名称和邮箱地址（不区分大小写），正则表达式则按模式匹配，匹配成功则跳过AI回复。
@@ -2,24 +2,28 @@

 ## 前置要求

- Halo 2.23+
- AI Foundation 插件（必须） — 本插件依赖 AI Foundation 提供的AI模型能力，请先安装并配置 AI Foundation
+- Halo 2.25+
+- AI Foundation 插件（必须） — 本插件通过 `ExtensionGetter` 调用 AI Foundation 提供的 `AiModelService` 扩展点，请先安装并配置 AI Foundation

 ## 安装

 ### 方式一：从 Release 下载

-1. 前往 [GitHub Releases](https://github.com/nxxy335/plugin-comment-ai-autopilot/releases) 下载最新的 `.jar` 文件
+1. 前往 [GitHub Releases](https://github.com/sunny-335/plugin-comment-ai-autopilot/releases) 下载最新的 `.jar` 文件
 2. 登录 Halo 管理后台
 3. 进入 **插件** → **已安装** → 点击右上角 **安装** 按钮
 4. 选择下载的 `.jar` 文件上传
 5. 安装完成后启用插件

+::: tip 推荐安装
+安装本插件后，应用市场会推荐安装 AI Foundation 插件（本插件的必要依赖）。
+:::
+
 ### 方式二：从源码构建

 ```bash
 # 克隆仓库
-git clone https://github.com/nxxy335/plugin-comment-ai-autopilot.git
+git clone https://github.com/sunny-335/plugin-comment-ai-autopilot.git
 cd plugin-comment-ai-autopilot

 # 构建
@@ -5,20 +5,22 @@ AI回评（Comment AI Autopilot）是一个 Halo 博客系统的插件，能够
 ## 核心功能

 - **自动回复** — 监听新评论，自动调用AI生成回复，支持多轮对话上下文
+- **多 AI 角色** — 支持创建多个 AI 角色，每个角色有独立的昵称、人格提示词和 Gravatar 头像，可为不同文章指定不同角色
 - **多语言适配** — 根据评论语言自动用对应语言回复
 - **情感分析** — 分析评论情感倾向（正面/中性/负面），根据情感调整回复语气
 - **草稿模式** — AI回复先存为草稿，管理员审核后再发布
 - **失败重试** — AI生成失败时自动重试，指数退避策略
 - **批量操作** — 草稿模式下支持批量通过/拒绝/删除
 - **文章/页面级开关** — 在文章编辑器中直接控制是否启用AI回复，文章默认开启，页面默认关闭
- **评论者黑名单** — 屏蔽指定评论者，不触发AI回复
+- **评论者黑名单** — 屏蔽指定评论者，不触发AI回复，支持名称、邮箱和正则表达式
 - **手动触发** — 在评论管理页面对历史评论手动触发AI回复
- **AI角色** — 自定义AI回复者的昵称、人格提示词和Gravatar头像
- **安全审核** — AI生成的内容经过安全审核，不合规内容自动拒绝
- **仪表盘统计** — 显示回复数、情感分布、每日回复趋势等图表
+- **安全审核** — AI生成的内容经过两阶段安全审核（安全检查 + 质量评分），不合规内容自动拒绝
+- **Prompt 预设** — 内置友好型、专业型、幽默型、简洁型预设风格，可多选组合
+- **对话轮次限制** — 同一评论线程中限制 AI 最多回复轮次，防止无限对话
+- **速率限制** — 每分钟最大 AI 回复数量，防止批量评论消耗过多额度
 - **日志筛选搜索** — 按状态、情感筛选，关键词搜索
 - **数据清理** — 自动清理超过指定天数的旧记录
- **AI Foundation 集成** — 必须安装 Halo AI Foundation 插件，使用其提供的AI模型能力
+- **AI Foundation 集成** — 通过 Halo 官方推荐的 `ExtensionGetter` 获取 AI 服务，需安装 AI Foundation 插件

 ## 工作流程

@@ -33,17 +35,17 @@ AI回评（Comment AI Autopilot）是一个 Halo 博客系统的插件，能够
 3. **情感分析** — 调用AI分析评论情感倾向
 4. **构建Prompt** — 结合AI角色人格、情感提示、文章内容、评论上下文构建Prompt
 5. **AI生成** — 调用AI模型生成回复内容
-6. **安全审核** — 对生成内容进行安全审核
+6. **安全审核** — 对生成内容进行两阶段审核（安全检查 + 质量评分 1-5 分映射到 0-100）
 7. **发布/草稿** — 根据设置自动发布或存为草稿等待审核
 8. **重试** — 如果AI生成失败，系统会自动重试（最多 maxRetryCount 次），每次重试间隔递增

 ## 前置要求

- Halo 2.23+
- AI Foundation 插件（必须） — 本插件依赖 AI Foundation 提供的AI模型能力
+- Halo 2.25+
+- AI Foundation 插件（必须） — 本插件通过 `ExtensionGetter` 调用 AI Foundation 提供的 `AiModelService` 扩展点

 ## 技术栈

 - **后端**：Java + Spring WebFlux + Reactive
 - **前端**：Vue 3 + @halo-dev/components
- **AI**：支持 AI Foundation 插件集成
+- **AI**：通过 AI Foundation 插件集成
@@ -31,7 +31,7 @@ AI回复者的显示名称，默认为「小回」。修改后新回复将使用
 填写邮箱后，AI回复者的头像将通过 Gravatar 服务自动生成：

 1. 插件根据邮箱生成 SHA-256 哈希
-2. 构造 Gravatar URL：`https://cn.cravatar.com/avatar/{hash}`
+2. 构造 Gravatar URL：`https://cn.cravatar.com/avatar/{hash}`（使用 [Cravatar](https://cn.cravatar.com) 镜像服务）
 3. 头像URL存储在评论的 `owner.annotations["avatar"]` 中

 ::: warning
@@ -9,6 +9,8 @@ Prompt模板控制AI生成回复时的完整提示词结构。

 {{safety_prompt}}

+【语言要求】请用评论所使用的语言回复。如果评论是英文，请用英文回复；如果是中文，请用中文回复；如果是日文，请用日文回复；以此类推。
+
 请回复以下评论。注意：
 - 回复长度应与评论长度匹配，简短问候简短回复
 - 不要复述或总结文章内容
@@ -21,6 +23,7 @@ Prompt模板控制AI生成回复时的完整提示词结构。
 文章（仅供理解上下文，不要复述）：
 {{article}}

+{{conversation_history}}
 评论：
 {{comment}}
 ```
@@ -29,23 +32,26 @@ Prompt模板控制AI生成回复时的完整提示词结构。

 | 变量 | 说明 | 注入时机 |
 |------|------|---------|
-| `{{persona_prompt}}` | AI角色人格提示词 | 始终注入 |
+| `{{persona_prompt}}` | AI角色人格提示词（含已启用的预设） | 始终注入 |
 | `{{safety_prompt}}` | 安全规范提示词 | 始终注入 |
-| `{{sentiment_prompt}}` | 情感语气提示词 | 情感分析后自动注入，不在模板中显式使用 |
 | `{{post_title}}` | 文章标题 | 始终注入 |
 | `{{post_date}}` | 文章发布日期（如 2024-01-15） | 始终注入 |
 | `{{comment_count}}` | 该文章的评论数 | 始终注入 |
 | `{{article}}` | 文章/页面内容（含标题） | 始终注入 |
+| `{{conversation_history}}` | 对话历史上下文 | 多轮对话时注入 |
 | `{{comment}}` | 评论内容（含评论者名称） | 始终注入 |
-| `{{conversation}}` | 对话上下文 | 多轮对话时注入 |
+
+::: warning 变量名注意
+对话上下文变量是 `{{conversation_history}}`（不是 `{{conversation}}`）。如果模板中使用了错误的变量名，该变量不会被替换。
+:::

 ## 情感提示

-情感提示由插件根据情感分析结果自动注入到Prompt中，不需要在模板中手动添加：
+情感提示由插件根据情感分析结果自动追加到 Prompt 末尾，不需要在模板中手动添加：

- **正面** → "评论者情绪积极友好，请用热情友好的语气回复，表达感谢和共鸣。"
- **负面** → "评论者情绪偏消极或不满，请用理性温和的语气回复，避免激化矛盾，适当表示理解。"
- **中性** → 不注入额外提示
+- **正面** → 追加"评论者情绪正面积极，请用热情友好的语气回复，可以表达感谢和共鸣。"
+- **负面** → 追加"评论者情绪偏负面，请用理性温和的语气回复，避免激化矛盾，展现理解和包容。"
+- **中性** → 不追加额外提示

 ## 安全提示

@@ -56,6 +62,17 @@ Prompt模板控制AI生成回复时的完整提示词结构。
 - 不生成虚假信息
 - 回复内容与评论相关

+## 预设风格
+
+在 Prompt 设置页面可以多选启用预设风格，启用后预设提示词会自动合并到 `{{persona_prompt}}` 之后：
+
+| 预设 | 说明 |
+|------|------|
+| 友好型 | 热情友好，多用感叹号和表情符号，口语化表达 |
+| 专业型 | 专业严谨，正式语言风格，有逻辑性 |
+| 幽默型 | 适当加入幽默元素，轻松诙谐但不过度 |
+| 简洁型 | 非常简洁，一两句话即可，不展开讨论 |
+
 ## 自定义建议

 自定义Prompt模板时，建议：
@@ -63,20 +80,7 @@ Prompt模板控制AI生成回复时的完整提示词结构。
 1. 保留 `{{persona_prompt}}` 和 `{{safety_prompt}}` 变量
 2. 保留 `{{article}}` 和 `{{comment}}` 变量
 3. 利用 `{{post_title}}`、`{{post_date}}`、`{{comment_count}}` 提供更丰富的上下文
-4. 在变量之间添加清晰的分隔和指令
-5. 避免让AI复述文章内容
-6. 控制回复长度和风格
-
-## 变量使用示例
-
-### 根据评论数调整回复风格
-
-```
-{{comment_count}}条评论说明这篇文章{{#if comment_count > 10}}很受欢迎{{/if}}。
-```
-
-### 利用发布日期
-
-```
-这篇文章发布于{{post_date}}，回复时请考虑时效性。
-```
+4. 保留 `{{conversation_history}}` 以支持多轮对话上下文
+5. 在变量之间添加清晰的分隔和指令
+6. 避免让AI复述文章内容
+7. 控制回复长度和风格
@@ -13,9 +13,9 @@
 ## 工作原理

 1. 评论通过过滤检查后，调用AI对评论内容进行情感分析
-2. AI返回情感分类结果（POSITIVE / NEUTRAL / NEGATIVE）
+2. AI 使用结构化输出（`OutputSpec.choice`）返回情感分类结果（POSITIVE / NEUTRAL / NEGATIVE）
 3. 如果情感分析失败（如AI不可用），默认降级为 NEUTRAL
-4. 情感结果传入 PromptBuilder，在生成Prompt时注入对应的语气提示
+4. 情感结果传入 PromptBuilder，在生成Prompt时追加对应的语气提示
 5. 情感结果同时记录在 `AiCommentReply` 的 `sentiment` 字段中

 ## 日志展示
@@ -1,30 +1,50 @@
 # 插件设置

-插件设置页面位于 **AI回评** → **插件设置**，包含以下配置组：
+插件设置页面位于 **AI回评** → **插件设置**，通过标签栏切换以下五个配置页面：
+
+- 基本设置
+- AI角色设置
+- 模型设置
+- Prompt设置
+- 数据清理

 ## 基本设置

 | 配置项 | 说明 | 默认值 |
 |--------|------|--------|
 | 自动回复 | 是否启用自动回复功能 | 开启 |
-| 自动发布 | AI回复是否自动发布 | 开启 |
+| 自动发布 | AI回复是否自动发布，关闭则存为草稿 | 开启 |
+| 最大对话轮次 | 同一评论线程中AI最多自动回复的轮次 | 8 |
+| 速率限制 | 每分钟最大AI回复数量，防止批量评论消耗过多额度 | 10 |
 | 最大重试次数 | AI生成失败时的最大重试次数 | 3 |
-| 评论者黑名单 | 不触发AI回复的评论者显示名称或邮箱，逗号分隔 | 空 |
+| 评论者黑名单 | 不触发AI回复的评论者，支持名称、邮箱和正则表达式（`regex:` 开头），逗号分隔 | 空 |
+
+::: tip 评论者黑名单
+黑名单支持三种格式：
+- **名称**：如 `张三`
+- **邮箱**：如 `spam@example.com`（不区分大小写）
+- **正则表达式**：以 `regex:` 开头，如 `regex:^spam.*`
+
+点击"添加评论者"按钮可从已有评论列表中选择评论者自动添加到黑名单。
+:::

 ## AI角色设置

+AI角色定义了回复评论的虚拟身份。支持创建多个角色，每个角色有独立的昵称、人格提示词和 Gravatar 头像，可指定一个为默认角色。
+
 | 配置项 | 说明 | 默认值 |
 |--------|------|--------|
-| AI角色昵称 | AI回复者的显示名称 | 小回 |
-| AI角色人格提示词 | 定义AI角色的人格和回复风格 | 见下方 |
-| AI角色邮箱 | 用于Gravatar头像服务展示头像 | 空 |
+| 角色昵称 | AI回复者的显示名称 | 小回 |
+| 人格提示词 | 定义AI角色的人格和回复风格 | 见下方 |
+| 邮箱 | 用于 Gravatar 头像服务展示头像 | 空 |
+| 设为默认 | 将该角色设为默认角色 | 第一个角色默认 |

 默认人格提示词：

 > 你是「小回」，一个友善的评论者。你的回复简洁自然，像朋友聊天一样。简短的评论就简短回复，有深度的讨论才展开回应。不要长篇大论，不要复述文章内容。

 ::: tip Gravatar头像
-填写邮箱后，AI回复者的头像将通过 [Gravatar](https://gravatar.com) 服务自动生成。如果不填写邮箱，将使用默认头像。
+填写邮箱后，AI回复者的头像将通过 [Gravatar](https://gravatar.com) 服务自动生成，使用 [Cravatar](https://cn.cravatar.com) 镜像。如果不填写邮箱，将使用默认头像。
 :::

 ## 模型设置
@@ -42,17 +62,38 @@
 | 配置项 | 说明 | 默认值 |
 |--------|------|--------|
 | 自定义Prompt模板 | AI生成回复时使用的Prompt模板 | 见下方 |
+| 启用预设 | 选择要启用的Prompt预设风格（可多选） | 空 |
+
+### 预设风格
+
+| 预设 | 说明 |
+|------|------|
+| 友好型 | 热情友好，多用感叹号和表情符号，口语化表达 |
+| 专业型 | 专业严谨，正式语言风格，有逻辑性 |
+| 幽默型 | 适当加入幽默元素，轻松诙谐但不过度 |
+| 简洁型 | 非常简洁，一两句话即可，不展开讨论 |
+
+预设提示词会自动合并到角色人格提示词之后。

 ### 模板变量

-| 变量 | 说明 |
-|------|------|
-| `{{persona_prompt}}` | AI角色人格提示词 |
-| `{{safety_prompt}}` | 安全规范提示词 |
-| `{{sentiment_prompt}}` | 情感语气提示词（自动注入） |
-| `{{article}}` | 文章内容 |
-| `{{comment}}` | 评论内容 |
-| `{{conversation}}` | 对话上下文（多轮对话时） |
+| 变量 | 说明 | 注入时机 |
+|------|------|---------|
+| `{{persona_prompt}}` | AI角色人格提示词（含已启用的预设） | 始终注入 |
+| `{{safety_prompt}}` | 安全规范提示词 | 始终注入 |
+| `{{post_title}}` | 文章标题 | 始终注入 |
+| `{{post_date}}` | 文章发布日期（如 2024-01-15） | 始终注入 |
+| `{{comment_count}}` | 该文章的评论数 | 始终注入 |
+| `{{article}}` | 文章/页面内容（含标题） | 始终注入 |
+| `{{conversation_history}}` | 对话历史上下文 | 多轮对话时注入 |
+| `{{comment}}` | 评论内容（含评论者名称） | 始终注入 |
+
+::: tip 情感提示
+情感提示由插件根据情感分析结果自动追加到 Prompt 末尾，不需要在模板中手动添加：
+- **正面** → 追加热情友好的语气提示
+- **负面** → 追加理性温和的语气提示
+- **中性** → 不追加额外提示
+:::

 ## 数据清理

@@ -64,3 +105,7 @@
 ::: tip
 你也可以在数据清理页面点击"立即清理"按钮手动触发清理操作。
 :::
+
+::: warning
+清理操作仅删除 `AiCommentReply` 记录（插件内部的日志记录），不会删除已发布的 Halo Reply 评论。
+:::
@@ -216,100 +216,25 @@ public class CommentAiAutopilotEndpoint implements CustomEndpoint {
    }

    private Mono<ServerResponse> getStats(ServerRequest request) {
-        String range = request.queryParam("range").orElse("7");
-
        return client.listAll(AiCommentReply.class, ListOptions.builder().build(), Sort.unsorted())
            .collectList()
-            .map(allReplies -> {
-                // 根据 range 计算截止时间
-                ZoneId zoneId = ZoneId.systemDefault();
-                LocalDate today = LocalDate.now(zoneId);
-                Instant cutoffInstant;
-                int trendDays;
-
-                if ("all".equals(range)) {
-                    cutoffInstant = null; // 不做时间过滤
-                    trendDays = 30; // "all" 时趋势也展示最近30天
-                } else {
-                    int days = Integer.parseInt(range);
-                    cutoffInstant = today.minusDays(days).atStartOfDay(zoneId).toInstant();
-                    trendDays = days;
-                }
-
-                // 根据 range 过滤记录
-                List<AiCommentReply> replies;
-                if (cutoffInstant != null) {
-                    replies = allReplies.stream()
-                        .filter(r -> {
-                            Instant ts = r.getMetadata().getCreationTimestamp();
-                            return ts != null && !ts.isBefore(cutoffInstant);
-                        })
-                        .toList();
-                } else {
-                    replies = allReplies;
-                }
-
+            .map(replies -> {
                long total = replies.size();
                long passCount = replies.stream()
                    .filter(r -> "PASS".equals(r.getSpec().getStatus())).count();
                long failCount = replies.stream()
                    .filter(r -> "FAIL".equals(r.getSpec().getStatus())).count();
-                double avgScore = replies.stream()
-                    .filter(r -> r.getSpec().getScore() != null && r.getSpec().getScore() > 0)
-                    .mapToInt(r -> r.getSpec().getScore())
-                    .average().orElse(0.0);

                long reviewingCount = replies.stream()
                    .filter(r -> "PASS".equals(r.getSpec().getStatus())
                        && !Boolean.TRUE.equals(r.getSpec().getPublished()))
                    .count();

-                Map<String, Long> sentimentDistribution = new HashMap<>();
-                sentimentDistribution.put("POSITIVE", 0L);
-                sentimentDistribution.put("NEUTRAL", 0L);
-                sentimentDistribution.put("NEGATIVE", 0L);
-                sentimentDistribution.put("UNKNOWN", 0L);
-                for (var r : replies) {
-                    String sentiment = r.getSpec().getSentiment();
-                    if (sentiment == null || sentiment.isBlank()) {
-                        sentimentDistribution.merge("UNKNOWN", 1L, Long::sum);
-                    } else {
-                        sentimentDistribution.merge(sentiment, 1L, Long::sum);
-                    }
-                }
-
-                // 计算 dailyTrend
-                DateTimeFormatter formatter = DateTimeFormatter.ofPattern("yyyy-MM-dd");
-                Map<LocalDate, Long> dailyMap = new HashMap<>();
-                for (int i = 0; i < trendDays; i++) {
-                    dailyMap.put(today.minusDays(i), 0L);
-                }
-                for (var r : replies) {
-                    Instant timestamp = r.getMetadata().getCreationTimestamp();
-                    if (timestamp != null) {
-                        try {
-                            LocalDate date = timestamp.atZone(zoneId).toLocalDate();
-                            if (dailyMap.containsKey(date)) {
-                                dailyMap.merge(date, 1L, Long::sum);
-                            }
-                        } catch (Exception ignored) {
-                        }
-                    }
-                }
-                List<DailyCount> dailyTrend = new ArrayList<>();
-                for (int i = 0; i < trendDays; i++) {
-                    LocalDate date = today.minusDays(i);
-                    dailyTrend.add(new DailyCount(date.format(formatter), dailyMap.get(date)));
-                }
-
-                return new StatsResponse(total, passCount, failCount, avgScore,
-                    reviewingCount, sentimentDistribution, dailyTrend);
+                return new StatsResponse(total, passCount, failCount, reviewingCount);
            })
            .onErrorResume(e -> {
                log.warn("Failed to fetch stats: {}", e.getMessage());
-                return Mono.just(new StatsResponse(0, 0, 0, 0.0, 0L,
-                    Map.of("POSITIVE", 0L, "NEUTRAL", 0L, "NEGATIVE", 0L, "UNKNOWN", 0L),
-                    List.of()));
+                return Mono.just(new StatsResponse(0, 0, 0, 0));
            })
            .flatMap(stats -> ServerResponse.ok().bodyValue(stats));
    }
@@ -346,16 +271,11 @@ public class CommentAiAutopilotEndpoint implements CustomEndpoint {
            )));
    }

-    public record DailyCount(String date, long count) {}
-
    public record StatsResponse(
        long total,
        long passCount,
        long failCount,
-        double avgScore,
-        long reviewingCount,
-        Map<String, Long> sentimentDistribution,
-        List<DailyCount> dailyTrend
+        long reviewingCount
    ) {}

    public record PersonaResponse(
@@ -1,85 +1,58 @@
 package top.nxxy335.commentaiautopilot.service;

 import lombok.extern.slf4j.Slf4j;
-import org.springframework.context.ApplicationContext;
 import org.springframework.stereotype.Component;
 import reactor.core.publisher.Mono;
-import run.halo.app.core.extension.Plugin;
-import run.halo.app.extension.ReactiveExtensionClient;
+import run.halo.aifoundation.AiModelService;
+import run.halo.aifoundation.chat.GenerateTextRequest;
+import run.halo.aifoundation.chat.GenerateTextResult;
+import run.halo.aifoundation.schema.OutputSpec;
+import run.halo.app.plugin.extensionpoint.ExtensionGetter;

-import java.lang.reflect.Method;
-import java.util.Map;
+import java.util.List;

 /**
- * AI Foundation client that uses runtime class loading and reflection
- * to call the AI Foundation plugin's AiModelService.
+ * AI Foundation client that uses Halo's {@link ExtensionGetter} to obtain the
+ * {@link AiModelService} extension provided by the ai-foundation plugin.
 * <p>
- * This approach avoids classloader identity issues by loading AiModelService
- * from ai-foundation's own classloader, so that Spring's getBeansOfType()
- * can correctly match the implementation bean.
+ * This is the recommended way to integrate with AI Foundation, see
+ * <a href="https://github.com/halo-dev/plugin-ai-foundation/blob/main/dev/dev.md">dev guide</a>.
 * <p>
- * No @ConditionalOnClass or pluginDependencies needed.
- * Always registered as a bean; availability is checked at runtime.
+ * Requires the following declaration in plugin.yaml:
+ * <pre>
+ * spec:
+ *   pluginDependencies:
+ *     ai-foundation?: "*"
+ * </pre>
+ * The dependency is optional, so the plugin still loads when AI Foundation is
+ * not installed; availability is checked at runtime and all calls return empty
+ * in that case.
 */
@Slf4j
@Component
 public class AiFoundationClient {

-    private static final String AI_FOUNDATION_PLUGIN_NAME = "ai-foundation";
-    private static final String AI_MODEL_SERVICE_CLASS = "run.halo.aifoundation.AiModelService";
+    private final ExtensionGetter extensionGetter;

-    private final ReactiveExtensionClient client;
-    private final ApplicationContext applicationContext;
-
-    public AiFoundationClient(ReactiveExtensionClient client, ApplicationContext applicationContext) {
-        this.client = client;
-        this.applicationContext = applicationContext;
+    public AiFoundationClient(ExtensionGetter extensionGetter) {
+        this.extensionGetter = extensionGetter;
    }

    /**
     * Call AI Foundation to generate a chat response using the specified model.
+     * Uses {@link GenerateTextRequest} with {@code maxRetries=2} so that
+     * transient model errors are retried by the SDK.
     *
     * @param prompt    the prompt text
     * @param modelName the AiModel metadata.name, null or blank to use default model
     * @return the generated text, or empty if AI Foundation is unavailable
     */
    public Mono<String> chat(String prompt, String modelName) {
-        return isAiFoundationEnabled()
-            .flatMap(enabled -> {
-                if (!enabled) {
-                    log.warn("AI Foundation plugin is not installed or not enabled, skipping AI reply");
-                    return Mono.empty();
-                }
-                return doChat(prompt, modelName);
-            });
-    }
-
-    /**
-     * Check if AI Foundation is available: plugin installed, enabled, and AiModelService bean found.
-     */
-    public Mono<Boolean> isAvailable() {
-        return isAiFoundationEnabled()
-            .flatMap(enabled -> {
-                if (!enabled) return Mono.just(false);
-                return findAiModelService().hasElement();
-            });
-    }
-
-    private Mono<Boolean> isAiFoundationEnabled() {
-        return client.fetch(Plugin.class, AI_FOUNDATION_PLUGIN_NAME)
-            .map(plugin -> plugin.getSpec().getEnabled())
-            .defaultIfEmpty(false)
-            .onErrorResume(e -> {
-                log.debug("Failed to check AI Foundation plugin status: {}", e.getMessage());
-                return Mono.just(false);
-            });
-    }
-
-    private Mono<String> doChat(String prompt, String modelName) {
-        return findAiModelService()
-            .flatMap(service -> invokeLanguageModel(service, modelName)
-                .flatMap(model -> invokeGenerateText(model, prompt))
-            )
+        return aiModelService()
+            .flatMap(service -> service.languageModel(modelName != null ? modelName : "")
+                .flatMap(model -> model.generateText(
+                    GenerateTextRequest.builder().prompt(prompt).maxRetries(2).build()))
+                .map(GenerateTextResult::getText))
            .doOnError(e -> log.error("AI Foundation call failed: {}", e.getMessage()))
            .onErrorResume(e -> {
                log.warn("AI Foundation not available: {}", e.getMessage());
@@ -88,132 +61,68 @@ public class AiFoundationClient {
    }

    /**
-     * Get PluginManager via the pluginWrapper bean registered in our plugin context.
-     * Halo's DefaultPluginApplicationContextFactory registers pluginWrapper as a singleton:
-     *   beanFactory.registerSingleton("pluginWrapper", pluginWrapper);
-     * Then PluginWrapper.getPluginManager() gives us the PluginManager instance.
+     * Call AI Foundation to classify text into one of the given choices using
+     * structured output ({@link OutputSpec#choice(List)}).
+     * <p>
+     * This is the recommended way to do classification per the dev guide,
+     * as it is more reliable than prompt parsing.
+     *
+     * @param systemPrompt system prompt describing the task
+     * @param userPrompt   the user input to classify
+     * @param choices      the allowed classification values
+     * @param modelName    the AiModel metadata.name, null or blank to use default model
+     * @return the selected choice string, or empty if AI Foundation is unavailable
     */
-    private Object findPluginManager() {
-        try {
-            Object pluginWrapper = applicationContext.getBean("pluginWrapper");
-            Method getPluginManagerMethod = pluginWrapper.getClass().getMethod("getPluginManager");
-            getPluginManagerMethod.setAccessible(true);
-            Object pm = getPluginManagerMethod.invoke(pluginWrapper);
-            if (pm != null) {
-                log.info("Found PluginManager via pluginWrapper bean: {}", pm.getClass().getName());
-            }
-            return pm;
-        } catch (NoSuchMethodException e) {
-            log.warn("pluginWrapper does not have getPluginManager() method: {}", e.getMessage());
-        } catch (Exception e) {
-            log.warn("Failed to get PluginManager via pluginWrapper: {}", e.getMessage());
-        }
-        log.warn("PluginManager not found");
-        return null;
+    public Mono<String> classify(String systemPrompt, String userPrompt,
+                                  List<String> choices, String modelName) {
+        return aiModelService()
+            .flatMap(service -> service.languageModel(modelName != null ? modelName : "")
+                .flatMap(model -> model.generateText(
+                    GenerateTextRequest.builder()
+                        .system(systemPrompt)
+                        .prompt(userPrompt)
+                        .output(OutputSpec.choice(choices))
+                        .maxRetries(2)
+                        .build()))
+                .map(result -> {
+                    Object output = result.getOutput();
+                    return output != null ? String.valueOf(output).trim() : "";
+                }))
+            .doOnError(e -> log.error("AI Foundation classify failed: {}", e.getMessage()))
+            .onErrorResume(e -> {
+                log.warn("AI Foundation not available: {}", e.getMessage());
+                return Mono.empty();
+            });
    }

    /**
-     * Find the AiModelService bean from ai-foundation's PluginApplicationContext.
-     * Uses PluginManager.getPlugin() to get the plugin wrapper, then reflection
-     * to get the plugin's ApplicationContext.
+     * Check if AI Foundation is available: plugin installed and an
+     * AiModelService extension is enabled.
     */
-    private Mono<Object> findAiModelService() {
-        return Mono.fromCallable(() -> {
-            Object pm = findPluginManager();
-            if (pm == null) return null;
-
-            // Call pm.getPlugin("ai-foundation") via reflection
-            Method getPluginMethod = pm.getClass().getMethod("getPlugin", String.class);
-            getPluginMethod.setAccessible(true);
-            Object pluginWrapper = getPluginMethod.invoke(pm, AI_FOUNDATION_PLUGIN_NAME);
-            if (pluginWrapper == null) {
-                log.debug("ai-foundation plugin not found in PluginManager");
-                return null;
-            }
-
-            // Call pluginWrapper.getPlugin() to get the plugin instance
-            Method getPluginInstanceMethod = pluginWrapper.getClass().getMethod("getPlugin");
-            getPluginInstanceMethod.setAccessible(true);
-            Object pluginInstance = getPluginInstanceMethod.invoke(pluginWrapper);
-            if (pluginInstance == null) {
-                log.debug("ai-foundation plugin instance is null");
-                return null;
-            }
-
-            // Get the plugin's ApplicationContext via reflection on SpringPlugin
-            // DefaultSpringPlugin is package-private, so we need setAccessible
-            Method getCtxMethod = pluginInstance.getClass().getMethod("getApplicationContext");
-            getCtxMethod.setAccessible(true);
-            ApplicationContext pluginAppContext = (ApplicationContext) getCtxMethod.invoke(pluginInstance);
-
-            // Get the plugin classloader
-            Method getClassLoaderMethod = pluginWrapper.getClass().getMethod("getPluginClassLoader");
-            getClassLoaderMethod.setAccessible(true);
-            ClassLoader pluginClassLoader = (ClassLoader) getClassLoaderMethod.invoke(pluginWrapper);
-
-            // Load AiModelService from ai-foundation's classloader
-            Class<?> aiModelServiceClass = pluginClassLoader.loadClass(AI_MODEL_SERVICE_CLASS);
-
-            // Find the AiModelService bean in ai-foundation's ApplicationContext
-            Map<String, ?> beans = pluginAppContext.getBeansOfType(aiModelServiceClass);
-            if (beans.isEmpty()) {
-                log.debug("AiModelService bean not found in ai-foundation's ApplicationContext");
-                return null;
-            }
-
-            log.info("Found AiModelService bean in ai-foundation's ApplicationContext");
-            Object result = beans.values().iterator().next();
-            return (Object) result;
-        }).doOnError(e -> log.error("Failed to find AiModelService: {}", e.getMessage()));
+    public Mono<Boolean> isAvailable() {
+        return aiModelService().hasElement()
+            .onErrorResume(e -> {
+                log.debug("AI Foundation not available: {}", e.getMessage());
+                return Mono.just(false);
+            });
    }

    /**
-     * Call service.languageModel(modelName) or service.languageModel() via reflection.
-     * Returns Mono&lt;LanguageModel&gt; from ai-foundation's classloader.
+     * Obtain the enabled AiModelService extension via ExtensionGetter.
+     * <p>
+     * Wrapped in {@link Mono#defer} with a {@link NoClassDefFoundError} guard so
+     * that the plugin still works when the optional ai-foundation dependency is
+     * not installed (the AiModelService API class is then absent from the
+     * classloader).
     */
-    private Mono<Object> invokeLanguageModel(Object service, String modelName) {
-        return Mono.fromCallable(() -> {
-            Method method;
-            if (modelName != null && !modelName.isBlank()) {
-                method = service.getClass().getMethod("languageModel", String.class);
-                method.setAccessible(true);
-                return method.invoke(service, modelName);
-            } else {
-                method = service.getClass().getMethod("languageModel");
-                method.setAccessible(true);
-                return method.invoke(service);
+    private Mono<AiModelService> aiModelService() {
+        return Mono.defer(() -> {
+            try {
+                return extensionGetter.getEnabledExtension(AiModelService.class);
+            } catch (NoClassDefFoundError e) {
+                log.debug("AI Foundation API not on classpath: {}", e.getMessage());
+                return Mono.empty();
            }
-        }).flatMap(result -> {
-            if (result instanceof Mono<?> mono) return mono;
-            return Mono.justOrEmpty(result);
        });
    }
-
-    /**
-     * Call model.generateText(prompt) via reflection, then extract text from result.
-     * Returns the generated text string.
-     */
-    private Mono<String> invokeGenerateText(Object model, String prompt) {
-        return Mono.fromCallable(() -> {
-            Method method = model.getClass().getMethod("generateText", String.class);
-            method.setAccessible(true);
-            return method.invoke(model, prompt);
-        }).flatMap(result -> {
-            if (result instanceof Mono<?> mono) {
-                return mono.map(this::extractText);
-            }
-            return Mono.justOrEmpty(extractText(result));
-        });
-    }
-
-    private String extractText(Object result) {
-        if (result == null) return null;
-        try {
-            Method getText = result.getClass().getMethod("getText");
-            getText.setAccessible(true);
-            return (String) getText.invoke(result);
-        } catch (Exception e) {
-            throw new RuntimeException("Failed to call getText() on GenerateTextResult: " + e.getMessage(), e);
-        }
-    }
 }
@@ -231,12 +231,13 @@ public class AiReplyOrchestrator {
                        }
                        return publishReply(context, aiReply, replyRecord, reviewResult.score(), personaName);
                    })
-                    .switchIfEmpty(
-                        publishReply(context, aiReply, replyRecord, 100, personaName)
-                    )
                    .onErrorResume(e -> {
-                        log.warn("[Orchestrator] Review error, auto-passing: {}", e.getMessage());
-                        return publishReply(context, aiReply, replyRecord, 100, personaName);
+                        // review() already handles errors internally (returns PASS),
+                        // so this only fires for errors from publishReply/updateRecord.
+                        // Do NOT re-call publishReply to avoid double-publish / overwriting published=false.
+                        log.error("[Orchestrator] Error during review/publish for {}: {}",
+                            context.commentId(), e.getMessage(), e);
+                        return Mono.empty();
                    });
            });
    }
@@ -42,6 +42,55 @@ public class ContextExtractor {
            });
    }

+    /**
+     * Fetch previous replies in the comment thread to provide conversation history.
+     * Only includes replies created before the triggering reply.
+     */
+    private Mono<String> fetchConversationHistory(String commentName, String triggerReplyName) {
+        if (triggerReplyName == null || triggerReplyName.isBlank()) {
+            return Mono.just("");
+        }
+        return client.fetch(Reply.class, triggerReplyName)
+            .flatMap(triggerReply -> {
+                var triggerTime = triggerReply.getMetadata().getCreationTimestamp();
+                return client.list(Reply.class,
+                        reply -> {
+                            if (!commentName.equals(reply.getSpec().getCommentName())) {
+                                return false;
+                            }
+                            if (triggerReplyName.equals(reply.getMetadata().getName())) {
+                                return false;
+                            }
+                            // Only include replies created before the trigger reply
+                            var replyTime = reply.getMetadata().getCreationTimestamp();
+                            return replyTime != null && triggerTime != null
+                                && !replyTime.isAfter(triggerTime);
+                        },
+                        null)
+                    .collectList()
+                    .map(replies -> {
+                        if (replies.isEmpty()) return "";
+                        // Sort by creation time
+                        replies.sort(java.util.Comparator.comparing(
+                            r -> r.getMetadata().getCreationTimestamp()));
+                        var sb = new StringBuilder();
+                        for (var r : replies) {
+                            var owner = r.getSpec().getOwner();
+                            String name = (owner != null && owner.getDisplayName() != null)
+                                ? owner.getDisplayName() : "匿名用户";
+                            boolean isAi = owner != null && owner.getAnnotations() != null
+                                && "true".equals(owner.getAnnotations().get("comment-ai-autopilot.nxxy335.top/is-ai"));
+                            String role = isAi ? "AI" : "用户";
+                            String content = extractReplyContent(r);
+                            sb.append(role).append("(").append(name).append("): ")
+                              .append(content).append("\n");
+                        }
+                        return sb.toString();
+                    });
+            })
+            .defaultIfEmpty("");
+    }
+
    private Mono<CommentContext> buildContext(Comment comment, boolean isAiConversation) {
        var commentContent = extractCommentContent(comment);
        var commentOwner = extractCommentOwner(comment);
@@ -63,7 +112,8 @@ public class ContextExtractor {
                            null,
                            isAiConversation,
                            formatPostDate(post),
-                            commentCount
+                            commentCount,
+                            ""
                        ))
                    )
                )
@@ -78,7 +128,8 @@ public class ContextExtractor {
                    null,
                    isAiConversation,
                    "",
-                    0
+                    0,
+                    ""
                ));
        }

@@ -93,7 +144,8 @@ public class ContextExtractor {
            null,
            isAiConversation,
            "",
-            0
+            0,
+            ""
        ));
    }

@@ -101,55 +153,68 @@ public class ContextExtractor {
        var replyContent = extractReplyContent(reply);
        var replyOwner = extractReplyOwner(reply);
        var subjectRef = comment.getSpec().getSubjectRef();
+        var commentName = comment.getMetadata().getName();
+        var replyName = reply.getMetadata().getName();
+
+        // Fetch conversation history for AI conversations
+        Mono<String> historyMono = isAiConversation
+            ? fetchConversationHistory(commentName, replyName)
+            : Mono.just("");

        if (subjectRef != null && "Post".equals(subjectRef.getKind())) {
            String postName = subjectRef.getName();
            return client.fetch(Post.class, postName)
                .flatMap(post -> getPostContent(postName)
-                    .flatMap(content -> getCommentCount(comment.getMetadata().getName())
-                        .map(commentCount -> new CommentContext(
-                            comment.getMetadata().getName(),
-                            postName,
-                            post.getSpec().getSlug(),
-                            replyContent,
-                            replyOwner,
-                            post.getSpec().getTitle(),
-                            content,
-                            reply.getMetadata().getName(),
-                            isAiConversation,
-                            formatPostDate(post),
-                            commentCount
-                        ))
+                    .flatMap(content -> getCommentCount(commentName)
+                        .flatMap(commentCount -> historyMono
+                            .map(history -> new CommentContext(
+                                commentName,
+                                postName,
+                                post.getSpec().getSlug(),
+                                replyContent,
+                                replyOwner,
+                                post.getSpec().getTitle(),
+                                content,
+                                replyName,
+                                isAiConversation,
+                                formatPostDate(post),
+                                commentCount,
+                                history
+                            ))
+                        )
                    )
                )
                .defaultIfEmpty(new CommentContext(
-                    comment.getMetadata().getName(),
+                    commentName,
                    postName,
                    "",
                    replyContent,
                    replyOwner,
                    "",
                    "",
-                    reply.getMetadata().getName(),
+                    replyName,
                    isAiConversation,
                    "",
-                    0
+                    0,
+                    ""
                ));
        }

-        return Mono.just(new CommentContext(
-            comment.getMetadata().getName(),
-            "",
-            "",
-            replyContent,
-            replyOwner,
-            "",
-            "",
-            reply.getMetadata().getName(),
-            isAiConversation,
-            "",
-            0
-        ));
+        return historyMono
+            .map(history -> new CommentContext(
+                commentName,
+                "",
+                "",
+                replyContent,
+                replyOwner,
+                "",
+                "",
+                replyName,
+                isAiConversation,
+                "",
+                0,
+                history
+            ));
    }

    private String extractCommentContent(Comment comment) {
@@ -245,6 +310,7 @@ public class ContextExtractor {
        String replyTo,
        boolean isAiConversation,
        String postDate,
-        int commentCount
+        int commentCount,
+        String conversationHistory
    ) {}
 }
@@ -75,6 +75,7 @@ public class PromptBuilder {
            文章（仅供理解上下文，不要复述）：
            {{article}}

+            {{conversation_history}}
            评论：
            {{comment}}
            """;
@@ -103,6 +104,7 @@ public class PromptBuilder {
                    .replace("{{post_date}}", context.postDate() != null ? context.postDate() : "")
                    .replace("{{comment_count}}", String.valueOf(context.commentCount()))
                    .replace("{{article}}", context.postTitle() + "\n" + context.postContent())
+                    .replace("{{conversation_history}}", formatConversationHistory(context))
                    .replace("{{comment}}", context.commentOwner() + ": " + context.commentContent());

                return prompt;
@@ -133,6 +135,7 @@ public class PromptBuilder {
                    .replace("{{post_date}}", context.postDate() != null ? context.postDate() : "")
                    .replace("{{comment_count}}", String.valueOf(context.commentCount()))
                    .replace("{{article}}", context.postTitle() + "\n" + context.postContent())
+                    .replace("{{conversation_history}}", formatConversationHistory(context))
                    .replace("{{comment}}", context.commentOwner() + ": " + context.commentContent());

                if (sentiment == null || "NEUTRAL".equals(sentiment)) {
@@ -147,6 +150,18 @@ public class PromptBuilder {
            });
    }

+    /**
+     * Format conversation history for inclusion in the prompt.
+     * Returns empty string if no history is available.
+     */
+    private String formatConversationHistory(ContextExtractor.CommentContext context) {
+        String history = context.conversationHistory();
+        if (history == null || history.isBlank()) {
+            return "";
+        }
+        return "对话历史（供理解上下文）：\n" + history + "\n";
+    }
+
    private Mono<String> getPromptTemplate() {
        return client.fetch(ConfigMap.class, CONFIG_MAP_NAME)
            .mapNotNull(cm -> {
@@ -1,6 +1,7 @@
 package top.nxxy335.commentaiautopilot.service;

 import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.DisposableBean;
 import org.springframework.stereotype.Component;

 import java.util.concurrent.ConcurrentHashMap;
@@ -8,15 +9,18 @@ import java.util.concurrent.atomic.AtomicInteger;

@Slf4j
@Component
-public class RateLimitService {
+public class RateLimitService implements DisposableBean {
    private final ConcurrentHashMap<Long, AtomicInteger> windowMap = new ConcurrentHashMap<>();
+    private final Thread cleanupThread;
+    private volatile boolean running = true;

    public RateLimitService() {
        // 每5分钟清理过期窗口，防止内存泄漏
-        Thread cleanupThread = new Thread(() -> {
-            while (!Thread.currentThread().isInterrupted()) {
+        cleanupThread = new Thread(() -> {
+            while (running && !Thread.currentThread().isInterrupted()) {
                try {
                    Thread.sleep(5 * 60 * 1000);
+                    if (!running) break;
                    cleanup();
                } catch (InterruptedException e) {
                    Thread.currentThread().interrupt();
@@ -55,4 +59,13 @@ public class RateLimitService {
            log.debug("[RateLimit] Cleaned up {} expired windows", removed);
        }
    }
+
+    @Override
+    public void destroy() {
+        running = false;
+        if (cleanupThread != null) {
+            cleanupThread.interrupt();
+        }
+        log.info("[RateLimit] Cleanup thread stopped");
+    }
 }
@@ -4,6 +4,8 @@ import lombok.extern.slf4j.Slf4j;
 import org.springframework.stereotype.Component;
 import reactor.core.publisher.Mono;

+import java.util.List;
+
@Component
@Slf4j
 public class ReviewService {
@@ -14,58 +16,135 @@ public class ReviewService {
        this.aiFoundationClient = aiFoundationClient;
    }

-    private static final String REVIEW_PROMPT_TEMPLATE = """
-            请你以内容安全审核员的身份，严格评估你刚刚生成的这段回复：
+    /**
+     * Safety classification choices.
+     * Using OutputSpec.choice for reliable structured output.
+     */
+    private static final String SAFE = "安全";
+    private static final String UNSAFE = "不安全";
+    private static final List<String> SAFETY_CHOICES = List.of(SAFE, UNSAFE);

-            文章：
-            %s
+    /**
+     * Quality rating choices for content quality assessment.
+     * Used to produce a meaningful score (1-5) instead of binary 0/100.
+     */
+    private static final String RATING_EXCELLENT = "5";
+    private static final String RATING_GOOD = "4";
+    private static final String RATING_AVERAGE = "3";
+    private static final String RATING_POOR = "2";
+    private static final String RATING_BAD = "1";
+    private static final List<String> QUALITY_CHOICES = List.of(
+        RATING_EXCELLENT, RATING_GOOD, RATING_AVERAGE, RATING_POOR, RATING_BAD
+    );

-            原始评论：
-            %s
+    private static final String SAFETY_SYSTEM_PROMPT = """
+            你是内容安全审核员。请严格评估AI生成的回复内容是否安全合规。
+            检查以下方面：
+            1. 是否包含辱骂、仇恨、偏见或煽动性言论
+            2. 是否包含不适宜公开传播的内容
+            3. 是否涉及暴力、歧视、人身攻击或违反法律法规的内容
+            只返回"安全"或"不安全"。""";

-            生成的回复：
-            %s
-
-            请检查以下内容：
-            1. 是否包含任何辱骂、仇恨、偏见或煽动性言论？
-            2. 是否包含不适宜公开传播的内容？
-            3. 是否涉及暴力、歧视、人身攻击或违反法律法规的内容？
-
-            请直接回答"安全"或"不安全"。如果"不安全"，请重新生成一段符合规范的温和回复。
-            """;
+    private static final String QUALITY_SYSTEM_PROMPT = """
+            你是AI回复质量评估员。请对AI生成的回复进行质量评分（1-5分）。
+            评分标准：
+            5分：回复切题、自然流畅、语气恰当、有帮助
+            4分：回复较好，基本切题，语气自然
+            3分：回复一般，基本切题但略显生硬或冗长
+            2分：回复较差，偏离主题或语气不当
+            1分：回复很差，完全不相关或不当
+            只返回数字 1-5。""";

+    /**
+     * Review AI reply content using two-stage assessment:
+     * 1. Safety check via OutputSpec.choice (安全/不安全)
+     * 2. Quality rating via OutputSpec.choice (1-5) — only if safe
+     *
+     * <p>Score mapping (0-100):
+     * <ul>
+     *   <li>Unsafe → 0 (FAIL)</li>
+     *   <li>Rating 5 → 100 (PASS)</li>
+     *   <li>Rating 4 → 85 (PASS)</li>
+     *   <li>Rating 3 → 70 (PASS)</li>
+     *   <li>Rating 2 → 50 (PASS, borderline)</li>
+     *   <li>Rating 1 → 30 (PASS, but low quality)</li>
+     * </ul>
+     */
    public Mono<ReviewResult> review(String articleContent, String commentContent, String aiReply,
                                      String modelName) {
-        String reviewPrompt = String.format(REVIEW_PROMPT_TEMPLATE,
-            truncate(articleContent, 2000),
+        String userPrompt = String.format("""
+                原始评论：
+                %s
+
+                生成的回复：
+                %s
+
+                请判断以上回复是否安全合规。""",
            truncate(commentContent, 500),
            truncate(aiReply, 500));

-        return aiFoundationClient.chat(reviewPrompt, modelName)
-            .map(this::parseSafetyResult)
+        // Stage 1: Safety check
+        return aiFoundationClient.classify(SAFETY_SYSTEM_PROMPT, userPrompt, SAFETY_CHOICES, modelName)
+            .flatMap(safetyResult -> {
+                if (UNSAFE.equals(safetyResult)) {
+                    log.warn("[Review] Content is UNSAFE");
+                    return Mono.just(new ReviewResult(0, "FAIL", "内容安全审核不通过"));
+                }
+                if (!SAFE.equals(safetyResult)) {
+                    log.warn("[Review] Unexpected safety result: {}, treating as unsafe", safetyResult);
+                    return Mono.just(new ReviewResult(0, "FAIL", "内容安全审核结果异常"));
+                }
+                // Stage 2: Quality rating (only for safe content)
+                return rateQuality(commentContent, aiReply, modelName);
+            })
            .defaultIfEmpty(new ReviewResult(100, "PASS", "审核无响应，自动通过"))
            .onErrorResume(e -> {
-                log.warn("Review failed, auto-passing: {}", e.getMessage());
+                log.warn("[Review] Review failed, auto-passing: {}", e.getMessage());
                return Mono.just(new ReviewResult(100, "PASS", "审核服务异常，自动通过"));
            });
    }

-    private ReviewResult parseSafetyResult(String response) {
-        if (response == null || response.isBlank()) {
-            return new ReviewResult(100, "PASS", "审核无响应，自动通过");
-        }
-        String trimmed = response.trim().toLowerCase();
-        if (trimmed.contains("不安全") || trimmed.contains("unsafe")) {
-            log.warn("AI Review: content is UNSAFE, response: {}", response);
-            return new ReviewResult(0, "FAIL", "内容安全审核不通过");
-        }
-        if (trimmed.contains("安全") || trimmed.contains("safe")) {
-            log.info("AI Review: content is SAFE");
-            return new ReviewResult(100, "PASS", "内容安全审核通过");
-        }
-        // If unclear response, default to pass
-        log.warn("AI Review: unclear response, auto-passing: {}", response);
-        return new ReviewResult(100, "PASS", "审核结果不明确，自动通过");
+    /**
+     * Rate the quality of a safe AI reply (1-5) and map to a 0-100 score.
+     */
+    private Mono<ReviewResult> rateQuality(String commentContent, String aiReply, String modelName) {
+        String qualityPrompt = String.format("""
+                评论：
+                %s
+
+                回复：
+                %s
+
+                请对以上回复进行质量评分（1-5分）。""",
+            truncate(commentContent, 500),
+            truncate(aiReply, 500));
+
+        return aiFoundationClient.classify(QUALITY_SYSTEM_PROMPT, qualityPrompt, QUALITY_CHOICES, modelName)
+            .map(rating -> {
+                int score = mapRatingToScore(rating);
+                String reason = "安全通过，质量评分: " + rating + "/5";
+                log.info("[Review] Content is SAFE, quality rating: {}/5, score: {}", rating, score);
+                return new ReviewResult(score, "PASS", reason);
+            })
+            .defaultIfEmpty(new ReviewResult(85, "PASS", "安全通过，质量评分默认 4/5"))
+            .onErrorResume(e -> {
+                log.warn("[Review] Quality rating failed, defaulting to 85: {}", e.getMessage());
+                return Mono.just(new ReviewResult(85, "PASS", "安全通过，质量评分异常"));
+            });
+    }
+
+    /**
+     * Map a 1-5 quality rating to a 0-100 score.
+     */
+    private int mapRatingToScore(String rating) {
+        return switch (rating) {
+            case RATING_EXCELLENT -> 100;
+            case RATING_GOOD -> 85;
+            case RATING_AVERAGE -> 70;
+            case RATING_POOR -> 50;
+            case RATING_BAD -> 30;
+            default -> 70; // default to average
+        };
    }

    private String truncate(String text, int maxLength) {
@@ -4,6 +4,8 @@ import lombok.extern.slf4j.Slf4j;
 import org.springframework.stereotype.Component;
 import reactor.core.publisher.Mono;

+import java.util.List;
+
@Component
@Slf4j
 public class SentimentService {
@@ -20,13 +22,27 @@ public class SentimentService {
        public static final String NEGATIVE = "NEGATIVE";
    }

-    public Mono<SentimentResult> analyzeSentiment(String commentContent, String modelName) {
-        String prompt = buildSentimentPrompt(commentContent);
+    private static final List<String> CHOICES = List.of(
+        SentimentResult.POSITIVE, SentimentResult.NEUTRAL, SentimentResult.NEGATIVE
+    );

-        return aiFoundationClient.chat(prompt, modelName)
-            .map(response -> {
-                String sentiment = parseSentiment(response);
-                return new SentimentResult(sentiment, 1.0);
+    /**
+     * Analyze sentiment using AI Foundation structured output
+     * ({@code OutputSpec.choice}) for reliable classification.
+     */
+    public Mono<SentimentResult> analyzeSentiment(String commentContent, String modelName) {
+        String systemPrompt = "你是一个情感分析助手。请分析评论的情感倾向，只返回 POSITIVE、NEUTRAL 或 NEGATIVE 之一。";
+        String userPrompt = "分析以下评论的情感倾向：\n\n" + commentContent;
+
+        return aiFoundationClient.classify(systemPrompt, userPrompt, CHOICES, modelName)
+            .map(sentiment -> {
+                String upper = sentiment.toUpperCase();
+                // Validate against known choices; default to NEUTRAL if unexpected
+                if (!CHOICES.contains(upper)) {
+                    log.warn("[Sentiment] Unexpected classification result: {}, defaulting to NEUTRAL", sentiment);
+                    return new SentimentResult(SentimentResult.NEUTRAL, 0.0);
+                }
+                return new SentimentResult(upper, 1.0);
            })
            .onErrorResume(e -> {
                log.warn("[Sentiment] Failed to analyze sentiment, defaulting to NEUTRAL: {}", e.getMessage());
@@ -34,16 +50,4 @@ public class SentimentService {
            })
            .defaultIfEmpty(new SentimentResult(SentimentResult.NEUTRAL, 0.0));
    }
-
-    private String buildSentimentPrompt(String commentContent) {
-        return "请分析以下评论的情感倾向。只回复一个词：POSITIVE（正面）、NEUTRAL（中性）或 NEGATIVE（负面）。\n\n评论内容：\n" + commentContent;
-    }
-
-    private String parseSentiment(String response) {
-        if (response == null || response.isBlank()) return SentimentResult.NEUTRAL;
-        String upper = response.trim().toUpperCase();
-        if (upper.contains("POSITIVE")) return SentimentResult.POSITIVE;
-        if (upper.contains("NEGATIVE")) return SentimentResult.NEGATIVE;
-        return SentimentResult.NEUTRAL;
-    }
 }
@@ -54,7 +54,7 @@ spec:
        - $formkit: textarea
          name: customPromptTemplate
          label: 自定义Prompt模板
-          value: "{{persona_prompt}}\n\n{{safety_prompt}}\n\n【语言要求】请用评论所使用的语言回复。如果评论是英文，请用英文回复；如果是中文，请用中文回复；如果是日文，请用日文回复；以此类推。\n\n请回复以下评论。注意：\n- 回复长度应与评论长度匹配，简短问候简短回复\n- 不要复述或总结文章内容\n- 自然对话，不要写小作文\n- 只有评论涉及具体内容时才针对性回应\n\n文章（仅供理解上下文，不要复述）：\n{{article}}\n\n评论：\n{{comment}}"
+          value: "{{persona_prompt}}\n\n{{safety_prompt}}\n\n【语言要求】请用评论所使用的语言回复。如果评论是英文，请用英文回复；如果是中文，请用中文回复；如果是日文，请用日文回复；以此类推。\n\n请回复以下评论。注意：\n- 回复长度应与评论长度匹配，简短问候简短回复\n- 不要复述或总结文章内容\n- 自然对话，不要写小作文\n- 只有评论涉及具体内容时才针对性回应\n\n文章（仅供理解上下文，不要复述）：\n{{article}}\n\n{{conversation_history}}\n评论：\n{{comment}}"
        - $formkit: select
          name: enabledPresets
          label: 启用预设
@@ -5,9 +5,17 @@ kind: Plugin
 metadata:
  # The name defines how the plugin is invoked, A unique name
  name: comment-ai-autopilot
+  annotations:
+    # Recommend installing AI Foundation from the app store after installing this plugin
+    # https://www.halo.run/store/apps/app-acslk9nu
+    "store.halo.run/recommended-apps": '["app-acslk9nu"]'
 spec:
  enabled: true
  requires: ">=2.25.0"
+  pluginDependencies:
+    # Optional dependency: plugin still loads without AI Foundation,
+    # but AI features require it to be installed and enabled.
+    ai-foundation?: "*"
  author:
    name: 暖心向阳335
    website: https://nxxy335.top
@@ -22,4 +30,4 @@ spec:
      url: "https://github.com/sunny-335/plugin-comment-ai-autopilot/blob/main/LICENSE"
  settingName: "comment-ai-autopilot-settings"
  configMapName: "comment-ai-autopilot-configmap"
-  version: "1.0.0-beta.1"
+  version: "1.0.0-beta.2"
@@ -108,152 +108,11 @@
        </VCard>
      </div>

-      <!-- Middle: Sentiment + Trend -->
-      <div class="grid grid-cols-1 gap-4 mt-4 lg:grid-cols-2">
-        <!-- Sentiment Distribution -->
-        <VCard :body-class="['!p-5']">
-          <h3 class="text-sm font-medium text-gray-500 mb-4">情感分布</h3>
-          <div class="space-y-3">
-            <div class="flex items-center gap-3">
-              <div class="w-2 h-2 rounded-full bg-green-500 shrink-0"></div>
-              <div class="flex-1 min-w-0">
-                <div class="flex items-center justify-between text-sm">
-                  <span class="text-gray-700">正面</span>
-                  <span class="font-medium text-green-600">{{ stats?.sentimentDistribution?.POSITIVE || 0 }}</span>
-                </div>
-                <div class="mt-1 h-1.5 bg-gray-100 rounded-full overflow-hidden">
-                  <div
-                    class="h-full bg-green-400 rounded-full transition-all duration-500"
-                    :style="{ width: getSentimentPercent('POSITIVE') + '%' }"
-                  ></div>
-                </div>
-              </div>
-            </div>
-            <div class="flex items-center gap-3">
-              <div class="w-2 h-2 rounded-full bg-gray-400 shrink-0"></div>
-              <div class="flex-1 min-w-0">
-                <div class="flex items-center justify-between text-sm">
-                  <span class="text-gray-700">中性</span>
-                  <span class="font-medium text-gray-600">{{ stats?.sentimentDistribution?.NEUTRAL || 0 }}</span>
-                </div>
-                <div class="mt-1 h-1.5 bg-gray-100 rounded-full overflow-hidden">
-                  <div
-                    class="h-full bg-gray-400 rounded-full transition-all duration-500"
-                    :style="{ width: getSentimentPercent('NEUTRAL') + '%' }"
-                  ></div>
-                </div>
-              </div>
-            </div>
-            <div class="flex items-center gap-3">
-              <div class="w-2 h-2 rounded-full bg-red-500 shrink-0"></div>
-              <div class="flex-1 min-w-0">
-                <div class="flex items-center justify-between text-sm">
-                  <span class="text-gray-700">负面</span>
-                  <span class="font-medium text-red-500">{{ stats?.sentimentDistribution?.NEGATIVE || 0 }}</span>
-                </div>
-                <div class="mt-1 h-1.5 bg-gray-100 rounded-full overflow-hidden">
-                  <div
-                    class="h-full bg-red-400 rounded-full transition-all duration-500"
-                    :style="{ width: getSentimentPercent('NEGATIVE') + '%' }"
-                  ></div>
-                </div>
-              </div>
-            </div>
-            <div class="flex items-center gap-3">
-              <div class="w-2 h-2 rounded-full bg-gray-300 shrink-0"></div>
-              <div class="flex-1 min-w-0">
-                <div class="flex items-center justify-between text-sm">
-                  <span class="text-gray-700">未知</span>
-                  <span class="font-medium text-gray-400">{{ stats?.sentimentDistribution?.UNKNOWN || 0 }}</span>
-                </div>
-                <div class="mt-1 h-1.5 bg-gray-100 rounded-full overflow-hidden">
-                  <div
-                    class="h-full bg-gray-300 rounded-full transition-all duration-500"
-                    :style="{ width: getSentimentPercent('UNKNOWN') + '%' }"
-                  ></div>
-                </div>
-              </div>
-            </div>
-          </div>
-        </VCard>
-
-        <!-- Daily Trend -->
-        <VCard :body-class="['!p-5']">
-          <div class="flex items-center justify-between mb-4">
-            <h3 class="text-sm font-medium text-gray-500">近7日回复趋势</h3>
-            <div class="inline-flex rounded-md border border-gray-200 overflow-hidden">
-              <button
-                class="px-2.5 py-1 text-xs transition-colors"
-                :class="range === '7' ? 'bg-blue-500 text-white' : 'bg-white text-gray-600 hover:bg-gray-50'"
-                @click="range = '7'"
-              >
-                7天
-              </button>
-              <button
-                class="px-2.5 py-1 text-xs border-l border-gray-200 transition-colors"
-                :class="range === '30' ? 'bg-blue-500 text-white' : 'bg-white text-gray-600 hover:bg-gray-50'"
-                @click="range = '30'"
-              >
-                30天
-              </button>
-              <button
-                class="px-2.5 py-1 text-xs border-l border-gray-200 transition-colors"
-                :class="range === 'all' ? 'bg-blue-500 text-white' : 'bg-white text-gray-600 hover:bg-gray-50'"
-                @click="range = 'all'"
-              >
-                全部
-              </button>
-            </div>
-          </div>
-          <div v-if="stats?.dailyTrend?.length" class="flex items-end gap-3" style="height: 160px">
-            <div
-              v-for="day in stats.dailyTrend"
-              :key="day.date"
-              class="flex-1 flex flex-col items-center justify-end h-full"
-            >
-              <div class="text-xs text-gray-500 mb-1 font-medium">{{ day.count }}</div>
-              <div
-                class="w-full rounded-t-md transition-all duration-500"
-                :class="day.count > 0 ? 'bg-gradient-to-t from-blue-500 to-blue-400' : 'bg-gray-100'"
-                :style="{ height: getTrendBarHeight(day.count) + 'px' }"
-              ></div>
-              <div class="text-[10px] text-gray-400 mt-2 whitespace-nowrap">{{ formatTrendDate(day.date) }}</div>
-            </div>
-          </div>
-          <div v-else class="flex items-center justify-center text-sm text-gray-400" style="height: 160px">
-            暂无数据
-          </div>
-        </VCard>
-      </div>
-
-      <!-- Bottom: Score + Quick Actions -->
-      <div class="grid grid-cols-1 gap-4 mt-4 sm:grid-cols-2">
-        <!-- Avg Score -->
-        <VCard :body-class="['!p-5']">
-          <h3 class="text-sm font-medium text-gray-500 mb-3">平均审核评分</h3>
-          <div class="flex items-center gap-4">
-            <div class="text-4xl font-bold" :class="scoreColor">{{ stats?.avgScore?.toFixed(1) || '0.0' }}</div>
-            <div class="flex-1">
-              <div class="h-3 bg-gray-100 rounded-full overflow-hidden">
-                <div
-                  class="h-full rounded-full transition-all duration-500"
-                  :class="scoreBarColor"
-                  :style="{ width: (stats?.avgScore || 0) * 10 + '%' }"
-                ></div>
-              </div>
-              <div class="flex justify-between text-[10px] text-gray-300 mt-1">
-                <span>0</span>
-                <span>5</span>
-                <span>10</span>
-              </div>
-            </div>
-          </div>
-        </VCard>
-
-        <!-- Quick Actions -->
+      <!-- Quick Actions -->
+      <div class="mt-4">
        <VCard :body-class="['!p-5']">
          <h3 class="text-sm font-medium text-gray-500 mb-3">快捷操作</h3>
-          <div class="grid grid-cols-2 gap-2">
+          <div class="grid grid-cols-1 gap-2 sm:grid-cols-3">
            <button
              class="flex items-center gap-2 px-3 py-2.5 rounded-md bg-gray-50 hover:bg-gray-100 transition-colors text-sm text-gray-700"
              @click="$router.push({ name: 'CommentAiAutopilotLogs' })"
@@ -273,15 +132,6 @@
              </svg>
              插件设置
            </button>
-            <button
-              class="flex items-center gap-2 px-3 py-2.5 rounded-md bg-gray-50 hover:bg-gray-100 transition-colors text-sm text-gray-700"
-              @click="$router.push({ name: 'CommentAiAutopilotSettings' })"
-            >
-              <svg class="w-4 h-4 text-gray-400" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-                <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M16 7a4 4 0 11-8 0 4 4 0 018 0zM12 14a7 7 0 00-7 7h14a7 7 0 00-7-7z" />
-              </svg>
-              AI角色
-            </button>
            <button
              class="flex items-center gap-2 px-3 py-2.5 rounded-md bg-gray-50 hover:bg-gray-100 transition-colors text-sm text-gray-700"
              @click="refreshData"
@@ -299,24 +149,16 @@
 </template>

 <script setup lang="ts">
-import { ref, computed, onMounted, watch } from "vue"
+import { ref, computed, onMounted } from "vue"
 import { axiosInstance } from "@halo-dev/api-client"
 import { VPageHeader, VButton, VCard, Toast } from "@halo-dev/components"
 import { IconPlug } from "@halo-dev/components"

-interface DailyCount {
-  date: string
-  count: number
-}
-
 interface StatsResponse {
  total: number
  passCount: number
  failCount: number
  reviewingCount: number
-  avgScore: number
-  sentimentDistribution: Record<string, number>
-  dailyTrend: DailyCount[]
 }

 interface PersonaResponse {
@@ -331,7 +173,6 @@ interface HealthResponse {

 const stats = ref<StatsResponse | null>(null)
 const persona = ref<PersonaResponse | null>(null)
-const range = ref("7")
 const health = ref<HealthResponse | null>(null)
 const healthVisible = ref(true)

@@ -340,24 +181,10 @@ const passRate = computed(() => {
  return Math.round((stats.value.passCount / stats.value.total) * 100)
 })

-const scoreColor = computed(() => {
-  const score = stats.value?.avgScore || 0
-  if (score >= 7) return "text-green-600"
-  if (score >= 4) return "text-amber-500"
-  return "text-red-500"
-})
-
-const scoreBarColor = computed(() => {
-  const score = stats.value?.avgScore || 0
-  if (score >= 7) return "bg-gradient-to-r from-green-400 to-green-500"
-  if (score >= 4) return "bg-gradient-to-r from-amber-400 to-amber-500"
-  return "bg-gradient-to-r from-red-400 to-red-500"
-})
-
 const fetchStats = async () => {
  try {
    const { data } = await axiosInstance.get(
-      `/apis/console.api.comment-ai-autopilot.nxxy335.top/v1alpha1/stats?range=${range.value}`,
+      `/apis/console.api.comment-ai-autopilot.nxxy335.top/v1alpha1/stats?range=7`,
    )
    stats.value = data
  } catch (e) {
@@ -421,31 +248,6 @@ const openSettings = () => {
  window.location.href = "/console/comment-ai-autopilot/settings"
 }

-const getSentimentPercent = (sentiment: string): number => {
-  const dist = stats.value?.sentimentDistribution
-  if (!dist) return 0
-  const total = Object.values(dist).reduce((a, b) => a + b, 0)
-  if (total === 0) return 0
-  return Math.round(((dist[sentiment] || 0) / total) * 100)
-}
-
-const getTrendBarHeight = (count: number): number => {
-  const trend = stats.value?.dailyTrend
-  if (!trend || trend.length === 0) return 0
-  const max = Math.max(...trend.map(d => d.count), 1)
-  return Math.max((count / max) * 100, count > 0 ? 8 : 4)
-}
-
-const formatTrendDate = (dateStr: string): string => {
-  if (!dateStr) return ''
-  const parts = dateStr.split('-')
-  return parts.length >= 3 ? `${parts[1]}/${parts[2]}` : dateStr
-}
-
-watch(range, () => {
-  fetchStats()
-})
-
 onMounted(() => {
  fetchStats()
  fetchPersona()
@@ -10,7 +10,7 @@
    </VPageHeader>

    <!-- Batch Operation Toolbar -->
-    <div v-if="selectedNames.size > 0" class="mx-4 mt-2 flex items-center gap-3 bg-blue-50 border border-blue-200 rounded-lg px-4 py-2.5">
+    <div v-if="selectedNames.size > 0" class="m-4 mb-0 flex items-center gap-3 bg-blue-50 border border-blue-200 rounded-lg px-4 py-2.5">
      <span class="text-sm text-blue-700">已选择 {{ selectedNames.size }} 项</span>
      <button
        class="text-xs px-3 py-1 rounded bg-green-600 text-white hover:bg-green-700 transition-colors"
@@ -39,10 +39,10 @@
    </div>

    <!-- Filter Bar -->
-    <div class="mx-4 mt-2 flex items-center gap-3">
+    <div class="m-4 mb-0 flex items-center gap-3">
      <select
        v-model="filterStatus"
-        class="rounded-md border border-gray-300 px-3 py-1.5 text-sm focus:border-blue-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+        class="rounded-md border border-gray-300 pl-3 pr-8 py-1.5 text-sm focus:border-blue-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
      >
        <option value="">全部状态</option>
        <option value="PASS">通过</option>
@@ -52,7 +52,7 @@
      </select>
      <select
        v-model="filterSentiment"
-        class="rounded-md border border-gray-300 px-3 py-1.5 text-sm focus:border-blue-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+        class="rounded-md border border-gray-300 pl-3 pr-8 py-1.5 text-sm focus:border-blue-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
      >
        <option value="">全部情感</option>
        <option value="POSITIVE">正面</option>
@@ -152,7 +152,8 @@
          <div class="px-4 py-2.5 bg-gray-50 border-t border-gray-100 flex items-center justify-between">
            <div class="flex items-center gap-4 text-xs text-gray-400">
              <span>
-                评分 <span :class="getScoreClass(reply.spec.score)" class="font-medium text-gray-600">{{ reply.spec.score }}</span>
+                评分 <span :class="getScoreClass(reply.spec.score)" class="font-medium">{{ reply.spec.score }}</span>
+                <span class="text-gray-300 ml-0.5">{{ getScoreLabel(reply.spec.score) }}</span>
              </span>
              <span v-if="reply.spec.postSlug" class="flex items-center gap-1">
                文章
@@ -508,6 +509,14 @@ const getScoreClass = (score: number) => {
  return "text-red-600"
 }

+const getScoreLabel = (score: number) => {
+  if (score >= 85) return "优秀"
+  if (score >= 70) return "良好"
+  if (score >= 50) return "一般"
+  if (score > 0) return "较差"
+  return ""
+}
+
 const getStatusClass = (status: string) => {
  switch (status) {
    case "PASS":
@@ -5,291 +5,324 @@
        <IconPlug class="mr-2 self-center" />
      </template>
      <template #actions>
-        <VButton type="secondary" size="sm" @click="exportConfig">导出配置</VButton>
-        <label class="inline-flex items-center px-3 py-1.5 text-sm font-medium rounded-md border border-gray-300 bg-white text-gray-700 hover:bg-gray-50 cursor-pointer transition-colors">
-          导入配置
-          <input type="file" accept=".json" class="hidden" @change="handleImportFile" />
-        </label>
-        <VButton @click="$router.push({ name: 'CommentAiAutopilot' })">返回概览</VButton>
+        <VSpace spacing="sm">
+          <VButton size="sm" @click="exportConfig">
+            <template #icon>
+              <svg style="width:14px;height:14px" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M4 16v1a3 3 0 003 3h10a3 3 0 003-3v-1m-4-4l-4 4m0 0l-4-4m4 4V4"/></svg>
+            </template>
+            导出
+          </VButton>
+          <label class="inline-flex items-center px-3 py-1.5 text-sm font-medium rounded-md border border-gray-300 bg-white text-gray-700 hover:bg-gray-50 cursor-pointer transition-colors">
+            <svg style="width:14px;height:14px;margin-right:4px" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M4 16v1a3 3 0 003 3h10a3 3 0 003-3v-1m-4-8l-4-4m0 0L8 8m4-4v12"/></svg>
+            导入
+            <input type="file" accept=".json" class="hidden" @change="handleImportFile" />
+          </label>
+          <VButton size="sm" @click="$router.push({ name: 'CommentAiAutopilot' })">返回概览</VButton>
+        </VSpace>
      </template>
    </VPageHeader>

    <div class="m-4">
      <VLoading v-if="loading" />

-      <div v-if="!loading" class="settings-container">
-        <!-- Left: Settings Sections -->
-        <div class="settings-sections space-y-5">
-          <!-- ========== 基本设置 ========== -->
-          <section class="settings-section">
-            <div class="section-header section-header--blue">
-              <div class="section-header__icon">
-                <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M10.325 4.317c.426-1.756 2.924-1.756 3.35 0a1.724 1.724 0 002.573 1.066c1.543-.94 3.31.826 2.37 2.37a1.724 1.724 0 001.066 2.573c1.756.426 1.756 2.924 0 3.35a1.724 1.724 0 00-1.066 2.573c.94 1.543-.826 3.31-2.37 2.37a1.724 1.724 0 00-2.573 1.066c-.426 1.756-2.924 1.756-3.35 0a1.724 1.724 0 00-2.573-1.066c-1.543.94-3.31-.826-2.37-2.37a1.724 1.724 0 00-1.066-2.573c-1.756-.426-1.756-2.924 0-3.35a1.724 1.724 0 001.066-2.573c-.94-1.543.826-3.31 2.37-2.37.996.608 2.296.07 2.572-1.065z"/><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M15 12a3 3 0 11-6 0 3 3 0 016 0z"/></svg>
-              </div>
-              <div class="section-header__text">
-                <h3>基本设置</h3>
-                <p>控制AI回复的基本行为</p>
-              </div>
-            </div>
-            <div class="section-body">
-              <!-- Switch Row: 自动回复 -->
-              <div class="form-row">
-                <div class="form-row__label">
-                  <span class="form-label">自动回复</span>
-                  <span class="form-hint">启用后，AI将自动回复新评论</span>
-                </div>
-                <label class="toggle">
-                  <input type="checkbox" v-model="settings.basic.autoReply" />
-                  <span class="toggle__track"><span class="toggle__thumb"></span></span>
-                </label>
-              </div>
-              <!-- Switch Row: 自动发布 -->
-              <div class="form-row">
-                <div class="form-row__label">
-                  <span class="form-label">自动发布</span>
-                  <span class="form-hint">关闭后AI回复将存为草稿，需手动审核发布</span>
-                </div>
-                <label class="toggle">
-                  <input type="checkbox" v-model="settings.basic.autoPublish" />
-                  <span class="toggle__track"><span class="toggle__thumb"></span></span>
-                </label>
-              </div>
-              <!-- Slider: 最大对话轮次 -->
-              <div class="form-field">
-                <div class="form-field__header">
-                  <span class="form-label">最大对话轮次</span>
-                  <span class="form-badge">{{ settings.basic.maxConversationRounds }}</span>
-                </div>
-                <span class="form-hint">同一评论线程中AI最多自动回复的轮次</span>
-                <div class="slider">
-                  <input type="range" v-model.number="settings.basic.maxConversationRounds" min="1" max="100" class="slider__input" />
-                  <div class="slider__marks">
-                    <span>1</span><span>50</span><span>100</span>
-                  </div>
-                </div>
-              </div>
-              <!-- Input: 速率限制 -->
-              <div class="form-field">
-                <label class="form-label">速率限制</label>
-                <span class="form-hint">每分钟最大AI回复数量</span>
-                <input type="number" v-model.number="settings.basic.rateLimitPerMinute" min="1" max="100" class="form-input" placeholder="10" />
-              </div>
-              <!-- Slider: 最大重试次数 -->
-              <div class="form-field">
-                <div class="form-field__header">
-                  <span class="form-label">最大重试次数</span>
-                  <span class="form-badge">{{ settings.basic.maxRetryCount }}</span>
-                </div>
-                <span class="form-hint">AI生成失败时的最大重试次数，采用指数退避策略</span>
-                <div class="slider">
-                  <input type="range" v-model.number="settings.basic.maxRetryCount" min="1" max="10" class="slider__input" />
-                  <div class="slider__marks">
-                    <span>1</span><span>5</span><span>10</span>
-                  </div>
-                </div>
-              </div>
-              <!-- Textarea: 评论者黑名单 -->
-              <div class="form-field">
-                <div class="form-field__header">
-                  <span class="form-label">评论者黑名单</span>
-                  <button class="btn-link" @click="openCommenterDialog">
-                    <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M18 9v3m0 0v3m0-3h3m-3 0h-3m-2-5a4 4 0 11-8 0 4 4 0 018 0zM3 20a6 6 0 0112 0v1H3v-1z"/></svg>
-                    添加评论者
-                  </button>
-                </div>
-                <span class="form-hint">支持名称、邮箱和正则表达式。正则以 regex: 开头，如 regex:^spam.*</span>
-                <textarea v-model="settings.basic.blockedCommenters" rows="2" class="form-textarea" placeholder="例如：张三, spam@example.com, 李四"></textarea>
-              </div>
-            </div>
-          </section>
+      <div v-if="!loading">
+        <!-- Tab Navigation (full width, above the grid) -->
+        <div class="settings-tabs">
+          <button
+            v-for="tab in tabItems"
+            :key="tab.value"
+            class="settings-tab"
+            :class="{ 'settings-tab--active': activeTab === tab.value }"
+            @click="activeTab = tab.value"
+          >
+            {{ tab.label }}
+          </button>
+        </div>

-          <!-- ========== AI角色设置 ========== -->
-          <section class="settings-section">
-            <div class="section-header section-header--purple">
-              <div class="section-header__icon">
-                <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M16 7a4 4 0 11-8 0 4 4 0 018 0zM12 14a7 7 0 00-7 7h14a7 7 0 00-7-7z"/></svg>
-              </div>
-              <div class="section-header__text">
-                <h3>AI角色设置</h3>
-                <p>定义AI虚拟评论者的身份和风格</p>
-              </div>
-            </div>
-            <div class="section-body">
-              <!-- Persona List -->
-              <VLoading v-if="personasLoading" />
-              <div v-else-if="personas.length === 0" class="persona-empty">
-                <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M17 20h5v-2a3 3 0 00-5.356-1.857M17 20H7m10 0v-2c0-.656-.126-1.283-.356-1.857M7 20H2v-2a3 3 0 015.356-1.857M7 20v-2c0-.656.126-1.283.356-1.857m0 0a5.002 5.002 0 019.288 0M15 7a3 3 0 11-6 0 3 3 0 016 0z"/></svg>
-                <span>请添加至少一个AI角色</span>
-              </div>
-              <div v-else class="persona-list">
-                <div v-for="p in personas" :key="p.metadata.name" class="persona-card">
-                  <div class="persona-card__avatar">
-                    <img v-if="getPersonaAvatar(p)" :src="getPersonaAvatar(p)" alt="头像" />
-                    <span v-else class="persona-card__avatar-fallback">{{ (p.spec.displayName || '?').charAt(0) }}</span>
-                  </div>
-                  <div class="persona-card__info">
-                    <div class="persona-card__name">
-                      {{ p.spec.displayName || '未命名' }}
-                      <span v-if="p.spec.isDefault" class="persona-card__badge">默认</span>
-                    </div>
-                    <div class="persona-card__prompt">{{ p.spec.prompt || '暂无提示词' }}</div>
-                  </div>
-                  <div class="persona-card__actions">
-                    <button class="btn-icon" title="编辑" @click="openPersonaDialog(p)">
-                      <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M11 5H6a2 2 0 00-2 2v11a2 2 0 002 2h11a2 2 0 002-2v-5m-1.414-9.414a2 2 0 112.828 2.828L11.828 15H9v-2.828l8.586-8.586z"/></svg>
-                    </button>
-                    <button v-if="!p.spec?.isDefault" class="btn-icon btn-icon--danger" title="删除" @click="deletePersona(p)">
-                      <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M19 7l-.867 12.142A2 2 0 0116.138 21H7.862a2 2 0 01-1.995-1.858L5 7m5 4v6m4-6v6m1-10V4a1 1 0 00-1-1h-4a1 1 0 00-1 1v3M4 7h16"/></svg>
-                    </button>
-                    <button v-if="!p.spec?.isDefault" class="btn-icon" title="设为默认" @click="setDefaultPersona(p)">
-                      <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M11.049 2.927c.3-.921 1.603-.921 1.902 0l1.519 4.674a1 1 0 00.95.69h4.915c.969 0 1.371 1.24.588 1.81l-3.976 2.888a1 1 0 00-.363 1.118l1.518 4.674c.3.922-.755 1.688-1.538 1.118l-3.976-2.888a1 1 0 00-1.176 0l-3.976 2.888c-.783.57-1.838-.197-1.538-1.118l1.518-4.674a1 1 0 00-.363-1.118l-3.976-2.888c-.784-.57-.38-1.81.588-1.81h4.914a1 1 0 00.951-.69l1.519-4.674z"/></svg>
-                    </button>
-                  </div>
+        <!-- Settings Grid: sections (left) + sidebar (right) -->
+        <div class="settings-container">
+          <!-- Left: Settings Sections -->
+          <div class="settings-sections space-y-5">
+            <!-- ========== 基本设置 ========== -->
+            <section v-if="activeTab === 'basic'" class="settings-section">
+              <div class="section-header section-header--blue">
+                <div class="section-header__icon">
+                  <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M10.325 4.317c.426-1.756 2.924-1.756 3.35 0a1.724 1.724 0 002.573 1.066c1.543-.94 3.31.826 2.37 2.37a1.724 1.724 0 001.066 2.573c1.756.426 1.756 2.924 0 3.35a1.724 1.724 0 00-1.066 2.573c.94 1.543-.826 3.31-2.37 2.37a1.724 1.724 0 00-2.573 1.066c-.426 1.756-2.924 1.756-3.35 0a1.724 1.724 0 00-2.573-1.066c-1.543.94-3.31-.826-2.37-2.37a1.724 1.724 0 00-1.066-2.573c-1.756-.426-1.756-2.924 0-3.35a1.724 1.724 0 001.066-2.573c-.94-1.543.826-3.31 2.37-2.37.996.608 2.296.07 2.572-1.065z"/><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M15 12a3 3 0 11-6 0 3 3 0 016 0z"/></svg>
+                </div>
+                <div class="section-header__text">
+                  <h3>基本设置</h3>
+                  <p>控制AI回复的基本行为</p>
                </div>
              </div>
-              <VButton type="secondary" @click="openPersonaDialog(null)">
-                <svg style="width:16px;height:16px;margin-right:4px" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M12 4v16m8-8H4"/></svg>
-                添加角色
-              </VButton>
-            </div>
-          </section>
-
-          <!-- ========== 模型设置 ========== -->
-          <section class="settings-section">
-            <div class="section-header section-header--green">
-              <div class="section-header__icon">
-                <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9.75 17L9 20l-1 1h8l-1-1-.75-3M3 13h18M5 17h14a2 2 0 002-2V5a2 2 0 00-2-2H5a2 2 0 00-2 2v10a2 2 0 002 2z"/></svg>
-              </div>
-              <div class="section-header__text">
-                <h3>模型设置</h3>
-                <p>配置AI Foundation提供的模型</p>
-              </div>
-            </div>
-            <div class="section-body">
-              <div class="form-field">
-                <label class="form-label">AI模型名称</label>
-                <span class="form-hint">留空使用AI Foundation默认模型，填写AiModel资源名称可指定模型</span>
-                <input type="text" v-model="settings.model.modelName" class="form-input" placeholder="留空使用默认模型" />
-              </div>
-            </div>
-          </section>
-
-          <!-- ========== Prompt设置 ========== -->
-          <section class="settings-section">
-            <div class="section-header section-header--amber">
-              <div class="section-header__icon">
-                <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 12h6m-6 4h6m2 5H7a2 2 0 01-2-2V5a2 2 0 012-2h5.586a1 1 0 01.707.293l5.414 5.414a1 1 0 01.293.707V19a2 2 0 01-2 2z"/></svg>
-              </div>
-              <div class="section-header__text">
-                <h3>Prompt设置</h3>
-                <p>自定义AI回复的提示词模板</p>
-              </div>
-            </div>
-            <div class="section-body">
-              <!-- Preset Selection -->
-              <div class="form-field">
-                <label class="form-label">Prompt预设</label>
-                <span class="form-hint">选择预设风格，可多选</span>
-                <div class="preset-grid">
-                  <label v-for="p in promptPresets" :key="p.key" class="preset-item" :class="{ 'preset-item--active': isPresetEnabled(p.key) }">
-                    <input type="checkbox" :checked="isPresetEnabled(p.key)" @change="togglePreset(p.key)" class="preset-checkbox" />
-                    <div class="preset-item__content">
-                      <span class="preset-item__label">{{ p.label }}</span>
-                      <span class="preset-item__desc">{{ p.desc }}</span>
-                    </div>
+              <div class="section-body">
+                <!-- Switch Row: 自动回复 -->
+                <div class="form-row">
+                  <div class="form-row__label">
+                    <span class="form-label">自动回复</span>
+                    <span class="form-hint">启用后，AI将自动回复新评论</span>
+                  </div>
+                  <label class="toggle">
+                    <input type="checkbox" v-model="settings.basic.autoReply" />
+                    <span class="toggle__track"><span class="toggle__thumb"></span></span>
                  </label>
                </div>
-              </div>
-              <div class="form-field">
-                <div class="form-field__header">
-                  <label class="form-label">自定义Prompt模板</label>
-                  <span class="form-hint--inline">留空使用默认模板</span>
+                <!-- Switch Row: 自动发布 -->
+                <div class="form-row">
+                  <div class="form-row__label">
+                    <span class="form-label">自动发布</span>
+                    <span class="form-hint">关闭后AI回复将存为草稿，需手动审核发布</span>
+                  </div>
+                  <label class="toggle">
+                    <input type="checkbox" v-model="settings.basic.autoPublish" />
+                    <span class="toggle__track"><span class="toggle__thumb"></span></span>
+                  </label>
                </div>
-                <div class="var-tags">
-                  <span v-for="v in promptVariables" :key="v.name" class="var-tag">
-                    <code>{{ v.name }}</code> {{ v.desc }}
-                  </span>
-                </div>
-                <textarea v-model="settings.prompt.customPromptTemplate" rows="10" class="form-textarea form-textarea--mono" placeholder="自定义Prompt模板"></textarea>
-              </div>
-            </div>
-          </section>
-
-          <!-- ========== 数据清理 ========== -->
-          <section class="settings-section">
-            <div class="section-header section-header--red">
-              <div class="section-header__icon">
-                <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M19 7l-.867 12.142A2 2 0 0116.138 21H7.862a2 2 0 01-1.995-1.858L5 7m5 4v6m4-6v6m1-10V4a1 1 0 00-1-1h-4a1 1 0 00-1 1v3M4 7h16"/></svg>
-              </div>
-              <div class="section-header__text">
-                <h3>数据清理</h3>
-                <p>自动清理过期的AI回复记录</p>
-              </div>
-            </div>
-            <div class="section-body">
-              <!-- Switch: 启用自动清理 -->
-              <div class="form-row">
-                <div class="form-row__label">
-                  <span class="form-label">启用自动清理</span>
-                  <span class="form-hint">每天自动清理超过保留天数的记录</span>
-                </div>
-                <label class="toggle">
-                  <input type="checkbox" v-model="settings.cleanup.cleanupEnabled" />
-                  <span class="toggle__track"><span class="toggle__thumb"></span></span>
-                </label>
-              </div>
-              <!-- Slider: 保留天数 -->
-              <div class="form-field">
-                <div class="form-field__header">
-                  <span class="form-label">保留天数</span>
-                  <span class="form-badge">{{ settings.cleanup.retentionDays }} 天</span>
-                </div>
-                <span class="form-hint">超过此天数的AI回复记录将被自动清理</span>
-                <div class="slider">
-                  <input type="range" v-model.number="settings.cleanup.retentionDays" min="1" max="365" class="slider__input" />
-                  <div class="slider__marks">
-                    <span>1天</span><span>180天</span><span>365天</span>
+                <!-- Slider: 最大对话轮次 -->
+                <div class="form-field">
+                  <div class="form-field__header">
+                    <span class="form-label">最大对话轮次</span>
+                    <span class="form-badge">{{ settings.basic.maxConversationRounds }}</span>
+                  </div>
+                  <span class="form-hint">同一评论线程中AI最多自动回复的轮次</span>
+                  <div class="slider">
+                    <input type="range" v-model.number="settings.basic.maxConversationRounds" min="1" max="100" class="slider__input" />
+                    <div class="slider__marks">
+                      <span>1</span><span>50</span><span>100</span>
+                    </div>
                  </div>
                </div>
-              </div>
-              <!-- Manual Cleanup -->
-              <div class="form-row form-row--bordered">
-                <div class="form-row__label">
-                  <span class="form-label">手动清理</span>
-                  <span class="form-hint">立即执行一次清理操作</span>
+                <!-- Input: 速率限制 -->
+                <div class="form-field">
+                  <label class="form-label">速率限制</label>
+                  <span class="form-hint">每分钟最大AI回复数量</span>
+                  <input type="number" v-model.number="settings.basic.rateLimitPerMinute" min="1" max="100" class="form-input" placeholder="10" />
                </div>
-                <div class="form-row__action">
-                  <VButton size="sm" type="secondary" @click="performCleanup" :disabled="cleanupLoading">
-                    {{ cleanupLoading ? '清理中...' : '立即清理' }}
+                <!-- Slider: 最大重试次数 -->
+                <div class="form-field">
+                  <div class="form-field__header">
+                    <span class="form-label">最大重试次数</span>
+                    <span class="form-badge">{{ settings.basic.maxRetryCount }}</span>
+                  </div>
+                  <span class="form-hint">AI生成失败时的最大重试次数，采用指数退避策略</span>
+                  <div class="slider">
+                    <input type="range" v-model.number="settings.basic.maxRetryCount" min="1" max="10" class="slider__input" />
+                    <div class="slider__marks">
+                      <span>1</span><span>5</span><span>10</span>
+                    </div>
+                  </div>
+                </div>
+                <!-- Textarea: 评论者黑名单 -->
+                <div class="form-field">
+                  <div class="form-field__header">
+                    <span class="form-label">评论者黑名单</span>
+                    <button class="btn-link" @click="openCommenterDialog">
+                      <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M18 9v3m0 0v3m0-3h3m-3 0h-3m-2-5a4 4 0 11-8 0 4 4 0 018 0zM3 20a6 6 0 0112 0v1H3v-1z"/></svg>
+                      添加评论者
+                    </button>
+                  </div>
+                  <span class="form-hint">支持名称、邮箱和正则表达式。正则以 regex: 开头，如 regex:^spam.*</span>
+                  <textarea v-model="settings.basic.blockedCommenters" rows="2" class="form-textarea" placeholder="例如：张三, spam@example.com, 李四"></textarea>
+                </div>
+              </div>
+            </section>
+
+            <!-- ========== AI角色设置 ========== -->
+            <section v-if="activeTab === 'persona'" class="settings-section">
+              <div class="section-header section-header--purple">
+                <div class="section-header__icon">
+                  <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M16 7a4 4 0 11-8 0 4 4 0 018 0zM12 14a7 7 0 00-7 7h14a7 7 0 00-7-7z"/></svg>
+                </div>
+                <div class="section-header__text">
+                  <h3>AI角色设置</h3>
+                  <p>定义AI虚拟评论者的身份和风格</p>
+                </div>
+              </div>
+              <div class="section-body">
+                <!-- Persona List -->
+                <VLoading v-if="personasLoading" />
+                <div v-else-if="personas.length === 0" class="persona-empty">
+                  <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M17 20h5v-2a3 3 0 00-5.356-1.857M17 20H7m10 0v-2c0-.656-.126-1.283-.356-1.857M7 20H2v-2a3 3 0 015.356-1.857M7 20v-2c0-.656.126-1.283.356-1.857m0 0a5.002 5.002 0 019.288 0M15 7a3 3 0 11-6 0 3 3 0 016 0z"/></svg>
+                  <span>请添加至少一个AI角色</span>
+                </div>
+                <div v-else class="persona-list">
+                  <div v-for="p in personas" :key="p.metadata.name" class="persona-card">
+                    <div class="persona-card__avatar">
+                      <img v-if="getPersonaAvatar(p)" :src="getPersonaAvatar(p)" alt="头像" />
+                      <span v-else class="persona-card__avatar-fallback">{{ (p.spec.displayName || '?').charAt(0) }}</span>
+                    </div>
+                    <div class="persona-card__info">
+                      <div class="persona-card__name">
+                        {{ p.spec.displayName || '未命名' }}
+                        <span v-if="p.spec.isDefault" class="persona-card__badge">默认</span>
+                      </div>
+                      <div class="persona-card__prompt">{{ p.spec.prompt || '暂无提示词' }}</div>
+                    </div>
+                    <div class="persona-card__actions">
+                      <button class="btn-icon" title="编辑" @click="openPersonaDialog(p)">
+                        <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M11 5H6a2 2 0 00-2 2v11a2 2 0 002 2h11a2 2 0 002-2v-5m-1.414-9.414a2 2 0 112.828 2.828L11.828 15H9v-2.828l8.586-8.586z"/></svg>
+                      </button>
+                      <button v-if="!p.spec?.isDefault" class="btn-icon btn-icon--danger" title="删除" @click="deletePersona(p)">
+                        <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M19 7l-.867 12.142A2 2 0 0116.138 21H7.862a2 2 0 01-1.995-1.858L5 7m5 4v6m4-6v6m1-10V4a1 1 0 00-1-1h-4a1 1 0 00-1 1v3M4 7h16"/></svg>
+                      </button>
+                      <button v-if="!p.spec?.isDefault" class="btn-icon" title="设为默认" @click="setDefaultPersona(p)">
+                        <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M11.049 2.927c.3-.921 1.603-.921 1.902 0l1.519 4.674a1 1 0 00.95.69h4.915c.969 0 1.371 1.24.588 1.81l-3.976 2.888a1 1 0 00-.363 1.118l1.518 4.674c.3.922-.755 1.688-1.538 1.118l-3.976-2.888a1 1 0 00-1.176 0l-3.976 2.888c-.783.57-1.838-.197-1.538-1.118l1.518-4.674a1 1 0 00-.363-1.118l-3.976-2.888c-.784-.57-.38-1.81.588-1.81h4.914a1 1 0 00.951-.69l1.519-4.674z"/></svg>
+                      </button>
+                    </div>
+                  </div>
+                </div>
+                <VButton type="secondary" @click="openPersonaDialog(null)">
+                  <svg style="width:16px;height:16px" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M12 4v16m8-8H4"/></svg>
+                  添加角色
+                </VButton>
+              </div>
+            </section>
+
+            <!-- ========== 模型设置 ========== -->
+            <section v-if="activeTab === 'model'" class="settings-section">
+              <div class="section-header section-header--green">
+                <div class="section-header__icon">
+                  <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9.75 17L9 20l-1 1h8l-1-1-.75-3M3 13h18M5 17h14a2 2 0 002-2V5a2 2 0 00-2-2H5a2 2 0 00-2 2v10a2 2 0 002 2z"/></svg>
+                </div>
+                <div class="section-header__text">
+                  <h3>模型设置</h3>
+                  <p>配置AI Foundation提供的模型</p>
+                </div>
+              </div>
+              <div class="section-body">
+                <div class="form-field">
+                  <label class="form-label">AI模型名称</label>
+                  <span class="form-hint">留空使用AI Foundation默认模型，填写AiModel资源名称可指定模型</span>
+                  <input type="text" v-model="settings.model.modelName" class="form-input" placeholder="留空使用默认模型" />
+                </div>
+              </div>
+            </section>
+
+            <!-- ========== Prompt设置 ========== -->
+            <section v-if="activeTab === 'prompt'" class="settings-section">
+              <div class="section-header section-header--amber">
+                <div class="section-header__icon">
+                  <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 12h6m-6 4h6m2 5H7a2 2 0 01-2-2V5a2 2 0 012-2h5.586a1 1 0 01.707.293l5.414 5.414a1 1 0 01.293.707V19a2 2 0 01-2 2z"/></svg>
+                </div>
+                <div class="section-header__text">
+                  <h3>Prompt设置</h3>
+                  <p>自定义AI回复的提示词模板</p>
+                </div>
+              </div>
+              <div class="section-body">
+                <!-- Preset Selection -->
+                <div class="form-field">
+                  <label class="form-label">Prompt预设</label>
+                  <span class="form-hint">选择预设风格，可多选</span>
+                  <div class="preset-grid">
+                    <label v-for="p in promptPresets" :key="p.key" class="preset-item" :class="{ 'preset-item--active': isPresetEnabled(p.key) }">
+                      <input type="checkbox" :checked="isPresetEnabled(p.key)" @change="togglePreset(p.key)" class="preset-checkbox" />
+                      <div class="preset-item__content">
+                        <span class="preset-item__label">{{ p.label }}</span>
+                        <span class="preset-item__desc">{{ p.desc }}</span>
+                      </div>
+                    </label>
+                  </div>
+                </div>
+                <div class="form-field">
+                  <div class="form-field__header">
+                    <label class="form-label">自定义Prompt模板</label>
+                    <span class="form-hint--inline">留空使用默认模板</span>
+                  </div>
+                  <div class="var-tags">
+                    <span v-for="v in promptVariables" :key="v.name" class="var-tag">
+                      <code>{{ v.name }}</code> {{ v.desc }}
+                    </span>
+                  </div>
+                  <textarea v-model="settings.prompt.customPromptTemplate" rows="10" class="form-textarea form-textarea--mono" placeholder="自定义Prompt模板"></textarea>
+                </div>
+              </div>
+            </section>
+
+            <!-- ========== 数据清理 ========== -->
+            <section v-if="activeTab === 'cleanup'" class="settings-section">
+              <div class="section-header section-header--red">
+                <div class="section-header__icon">
+                  <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M19 7l-.867 12.142A2 2 0 0116.138 21H7.862a2 2 0 01-1.995-1.858L5 7m5 4v6m4-6v6m1-10V4a1 1 0 00-1-1h-4a1 1 0 00-1 1v3M4 7h16"/></svg>
+                </div>
+                <div class="section-header__text">
+                  <h3>数据清理</h3>
+                  <p>自动清理过期的AI回复记录</p>
+                </div>
+              </div>
+              <div class="section-body">
+                <!-- Switch: 启用自动清理 -->
+                <div class="form-row">
+                  <div class="form-row__label">
+                    <span class="form-label">启用自动清理</span>
+                    <span class="form-hint">每天自动清理超过保留天数的记录</span>
+                  </div>
+                  <label class="toggle">
+                    <input type="checkbox" v-model="settings.cleanup.cleanupEnabled" />
+                    <span class="toggle__track"><span class="toggle__thumb"></span></span>
+                  </label>
+                </div>
+                <!-- Slider: 保留天数 -->
+                <div class="form-field">
+                  <div class="form-field__header">
+                    <span class="form-label">保留天数</span>
+                    <span class="form-badge">{{ settings.cleanup.retentionDays }} 天</span>
+                  </div>
+                  <span class="form-hint">超过此天数的AI回复记录将被自动清理</span>
+                  <div class="slider">
+                    <input type="range" v-model.number="settings.cleanup.retentionDays" min="1" max="365" class="slider__input" />
+                    <div class="slider__marks">
+                      <span>1天</span><span>180天</span><span>365天</span>
+                    </div>
+                  </div>
+                </div>
+                <!-- Manual Cleanup -->
+                <div class="form-row form-row--bordered">
+                  <div class="form-row__label">
+                    <span class="form-label">手动清理</span>
+                    <span class="form-hint">立即执行一次清理操作</span>
+                  </div>
+                  <div class="form-row__action">
+                    <VButton size="sm" type="secondary" @click="performCleanup" :disabled="cleanupLoading">
+                      {{ cleanupLoading ? '清理中...' : '立即清理' }}
+                    </VButton>
+                  </div>
+                </div>
+                <div v-if="cleanupResult !== null" class="cleanup-result">
+                  <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 12l2 2 4-4m6 2a9 9 0 11-18 0 9 9 0 0118 0z"/></svg>
+                  清理完成，共删除 {{ cleanupResult }} 条记录
+                </div>
+              </div>
+            </section>
+          </div>
+
+          <!-- Right: Sticky Save Bar -->
+          <div class="settings-sidebar">
+            <div class="settings-sidebar__inner">
+              <div class="sidebar-card sidebar-card--primary">
+                <div class="sidebar-card__header">
+                  <h4 class="sidebar-card__title">操作</h4>
+                  <span v-if="hasUnsavedChanges" class="sidebar-card__badge">未保存</span>
+                </div>
+                <div class="sidebar-card__actions">
+                  <VButton block type="primary" @click="saveSettings" :disabled="saving">
+                    <template v-if="!saving">
+                      <svg style="width:14px;height:14px" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M5 13l4 4L19 7"/></svg>
+                    </template>
+                    {{ saving ? '保存中...' : '保存设置' }}
+                  </VButton>
+                  <VButton block @click="fetchSettings" :disabled="saving">
+                    <svg style="width:14px;height:14px" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M4 4v5h.582m15.356 2A8.001 8.001 0 004.582 9m0 0H9m11 11v-5h-.581m0 0a8.003 8.003 0 01-15.357-2m15.357 2H15"/></svg>
+                    重置为当前值
                  </VButton>
                </div>
              </div>
-              <div v-if="cleanupResult !== null" class="cleanup-result">
-                <svg fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 12l2 2 4-4m6 2a9 9 0 11-18 0 9 9 0 0118 0z"/></svg>
-                清理完成，共删除 {{ cleanupResult }} 条记录
-              </div>
-            </div>
-          </section>
-        </div>
-
-        <!-- Right: Sticky Save Bar -->
-        <div class="settings-sidebar">
-          <div class="settings-sidebar__inner">
-            <div class="sidebar-card">
-              <h4 class="sidebar-card__title">操作</h4>
-              <div class="sidebar-card__actions">
-                <VButton block type="primary" @click="saveSettings" :disabled="saving">
-                  {{ saving ? '保存中...' : '保存设置' }}
-                </VButton>
-                <VButton block @click="fetchSettings">重置为当前值</VButton>
-              </div>
-            </div>
-            <div class="sidebar-card">
-              <h4 class="sidebar-card__title">可用模板变量</h4>
-              <div class="sidebar-card__vars">
-                <div v-for="v in promptVariables" :key="v.name" class="sidebar-var">
-                  <code>{{ v.name }}</code>
-                  <span>{{ v.desc }}</span>
+              <div v-if="activeTab === 'prompt'" class="sidebar-card">
+                <h4 class="sidebar-card__title">可用模板变量</h4>
+                <div class="sidebar-card__vars">
+                  <div v-for="v in promptVariables" :key="v.name" class="sidebar-var">
+                    <code>{{ v.name }}</code>
+                    <span>{{ v.desc }}</span>
+                  </div>
                </div>
              </div>
            </div>
@@ -414,8 +447,16 @@
 <script setup lang="ts">
 import { ref, reactive, computed, onMounted, watch } from "vue"
 import { axiosInstance, coreApiClient } from "@halo-dev/api-client"
-import { VPageHeader, VButton, VLoading, Toast, VModal, VSpace } from "@halo-dev/components"
-import { IconPlug } from "@halo-dev/components"
+import { VPageHeader, VButton, VLoading, Toast, VModal, VSpace, IconPlug } from "@halo-dev/components"
+
+const activeTab = ref("basic")
+const tabItems = [
+  { label: "基本设置", value: "basic" },
+  { label: "AI角色", value: "persona" },
+  { label: "模型设置", value: "model" },
+  { label: "Prompt", value: "prompt" },
+  { label: "数据清理", value: "cleanup" },
+]

 const promptVariables = [
  { name: '{{persona_prompt}}', desc: '人格提示词' },
@@ -489,6 +530,12 @@ const settings = reactive({
  cleanup: { cleanupEnabled: true, retentionDays: 30 },
 })

+// Track unsaved changes: snapshot of settings after last fetch/save
+const lastSavedSnapshot = ref("")
+const hasUnsavedChanges = computed(() => {
+  return JSON.stringify(settings) !== lastSavedSnapshot.value
+})
+
 const configMapName = "comment-ai-autopilot-configmap"
 const apiBase = "/apis/console.api.comment-ai-autopilot.nxxy335.top/v1alpha1"

@@ -817,7 +864,11 @@ const fetchSettings = async () => {
      if (Object.keys(cleanup).length) { settings.cleanup.cleanupEnabled = cleanup.cleanupEnabled !== false; settings.cleanup.retentionDays = (cleanup.retentionDays as number) || 30 }
    }
  } catch (e) { console.error("Failed to fetch settings", e) }
-  finally { loading.value = false }
+  finally {
+    loading.value = false
+    // Update snapshot after fetch to reset unsaved indicator
+    lastSavedSnapshot.value = JSON.stringify(settings)
+  }
 }

 const saveSettings = async () => {
@@ -834,6 +885,8 @@ const saveSettings = async () => {
    }
    await coreApiClient.configMap.updateConfigMap({ name: configMapName, configMap: updated })
    Toast.success("设置已保存")
+    // Update snapshot after save to reset unsaved indicator
+    lastSavedSnapshot.value = JSON.stringify(settings)
  } catch (e) { console.error("Failed to save settings", e); Toast.error("保存设置失败") }
  finally { saving.value = false }
 }
@@ -847,6 +900,43 @@ onMounted(async () => {

 <style scoped>
 /* ===== Layout ===== */
+.settings-tabs {
+  display: flex;
+  gap: 4px;
+  margin-bottom: 20px;
+  padding: 4px;
+  background: #fff;
+  border: 1px solid #e5e7eb;
+  border-radius: 12px;
+  overflow-x: auto;
+}
+.settings-tab {
+  flex: 1;
+  padding: 10px 16px;
+  font-size: 14px;
+  font-weight: 500;
+  color: #6b7280;
+  background: transparent;
+  border: none;
+  border-radius: 8px;
+  cursor: pointer;
+  transition: all 0.15s ease;
+  white-space: nowrap;
+  text-align: center;
+}
+.settings-tab:hover {
+  background: #f3f4f6;
+  color: #374151;
+}
+.settings-tab--active {
+  background: #3b82f6;
+  color: #fff;
+}
+.settings-tab--active:hover {
+  background: #2563eb;
+  color: #fff;
+}
+
 .settings-container {
  display: grid;
  grid-template-columns: 1fr 280px;
@@ -1221,14 +1311,39 @@ onMounted(async () => {
  background: #fff; border: 1px solid #e5e7eb; border-radius: 12px;
  padding: 20px;
 }
+.sidebar-card--primary {
+  border-color: #dbeafe;
+  background: linear-gradient(135deg, #eff6ff 0%, #fff 100%);
+}
+.sidebar-card__header {
+  display: flex; align-items: center; justify-content: space-between;
+  margin-bottom: 14px; padding-bottom: 10px;
+  border-bottom: 1px solid #f3f4f6;
+}
+.sidebar-card__badge {
+  display: inline-flex; align-items: center;
+  padding: 2px 8px; font-size: 11px; font-weight: 500;
+  background: #fef3c7; color: #d97706; border-radius: 4px;
+}
 .sidebar-card__title {
  font-size: 14px; font-weight: 600; color: #1f2937;
  margin: 0 0 14px; padding-bottom: 10px;
  border-bottom: 1px solid #f3f4f6;
 }
+.sidebar-card--primary .sidebar-card__title {
+  margin-bottom: 0; padding-bottom: 0; border-bottom: none;
+}
 .sidebar-card__actions {
  display: flex; flex-direction: column; gap: 8px;
 }
+/* Align icon + text inside VButton (btn-content wraps the default slot) */
+.sidebar-card__actions :deep(.btn-content),
+.section-body :deep(.btn-content) {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 4px;
+}
 .sidebar-card__vars {
  display: flex; flex-direction: column; gap: 8px;
 }
Author	SHA1	Message	Date
sunny-335	e6839b4aad	fix: auto-publish log showing unpublished, merge changelog, version to 1.0.0-beta.2	2026-06-17 23:13:45 +08:00
sunny-335	c3e003a572	feat: streamline dashboard, fix docs, optimize UI width - Remove AI角色 quick action, keep only 回复日志/插件设置/刷新数据 - Fix LogsView filter/batch bar width to match content area (mx-4 mt-2 -> m-4 mb-0) - Fix docs: Halo version 2.23->2.25, GitHub link nxxy335->sunny-335, Cravatar link, variable name {{conversation}}->{{conversation_history}} - Add missing settings docs (maxConversationRounds, rateLimitPerMinute, enabledPresets) - Update README: remove version display, update changelog - Update version to 1.0.0-beta.2-r8k4n2	2026-06-17 22:03:35 +08:00
sunny-335	be6a01a06e	fix: logs filter dropdown text overlapping with caret - Change select padding from px-3 to pl-3 pr-8 to reserve space for the native dropdown arrow - Update version to 1.0.0-beta.2-q9x3m7	2026-06-17 21:10:06 +08:00
sunny-335	767efdccbb	feat: v1.0.0-beta.2 - ExtensionGetter integration, UI revamp, bug fixes - Replace cross-ClassLoader reflection with ExtensionGetter.getEnabledExtension(AiModelService.class) - Declare optional pluginDependencies (ai-foundation?: "*") and recommended-apps annotation - Use OutputSpec.choice for structured classification (sentiment, review safety/quality) - Use GenerateTextRequest with maxRetries=2 for reliable chat generation - Add multi-turn conversation context (conversation_history placeholder) - Fix RateLimitService thread leak (implement DisposableBean) - Two-stage AI review: safety check + 1-5 quality score mapped to 0-100 - Remove redundant dashboard cards (sentiment distribution, 7-day trend, avg score) - Redesign settings page with tabbed navigation (basic/persona/model/prompt/cleanup) - Fix settings layout (move tab bar out of grid container) - Fix button icon+text alignment via :deep(.btn-content) inline-flex - Add review score grade labels (excellent/good/fair/poor) in logs	2026-06-17 19:40:10 +08:00
sunny-335	bf989ca3f2	docs: update README with changelog and fix license link, add CHANGELOG.md	2026-06-17 13:26:18 +08:00