diff --git a/.gitignore b/.gitignore
index dd3ca40..141eb19 100644
--- a/.gitignore
+++ b/.gitignore
@@ -71,9 +71,14 @@ env/
 
 # local storage
 api/storage/
+apps/api/storage/
 
 # agent memory (runtime data, user-specific — not for version control)
 api/data/memory/
+apps/api/data/memory/
+
+# local runtime caches
+apps/api/tmp/
 
 # daily rotating log files
 api/logs/
diff --git a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/3acf8dfe-1723-4896-a9e6-23a547d5737a.md b/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/3acf8dfe-1723-4896-a9e6-23a547d5737a.md
deleted file mode 100644
index 901fdac..0000000
--- a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/3acf8dfe-1723-4896-a9e6-23a547d5737a.md
+++ /dev/null
@@ -1,3 +0,0 @@
-# test
-
-this is a tiny ingestion regression check
\ No newline at end of file
diff --git a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/41b24de4-9d20-48d8-bc3c-9fe8fdeb4934.md b/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/41b24de4-9d20-48d8-bc3c-9fe8fdeb4934.md
deleted file mode 100644
index 1822a81..0000000
--- a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/41b24de4-9d20-48d8-bc3c-9fe8fdeb4934.md
+++ /dev/null
@@ -1,7 +0,0 @@
-# Codex Ingestion Smoke Test
-
-This source is used to verify the pending -> processing -> indexed pipeline.
-
-- queue routing
-- after commit dispatch
-- worker consumption
diff --git a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/72708d3a-a42e-4a96-9fe2-28081462ed31.md b/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/72708d3a-a42e-4a96-9fe2-28081462ed31.md
deleted file mode 100644
index 1822a81..0000000
--- a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/72708d3a-a42e-4a96-9fe2-28081462ed31.md
+++ /dev/null
@@ -1,7 +0,0 @@
-# Codex Ingestion Smoke Test
-
-This source is used to verify the pending -> processing -> indexed pipeline.
-
-- queue routing
-- after commit dispatch
-- worker consumption
diff --git a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/85239377-83ad-478b-b819-f36d894bb9e3.pdf b/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/85239377-83ad-478b-b819-f36d894bb9e3.pdf
deleted file mode 100644
index e1571d1..0000000
Binary files a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/85239377-83ad-478b-b819-f36d894bb9e3.pdf and /dev/null differ
diff --git a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/a9b726fd-1b6f-4444-a297-0dbdb73fdd3c.md b/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/a9b726fd-1b6f-4444-a297-0dbdb73fdd3c.md
deleted file mode 100644
index 1822a81..0000000
--- a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/a9b726fd-1b6f-4444-a297-0dbdb73fdd3c.md
+++ /dev/null
@@ -1,7 +0,0 @@
-# Codex Ingestion Smoke Test
-
-This source is used to verify the pending -> processing -> indexed pipeline.
-
-- queue routing
-- after commit dispatch
-- worker consumption
diff --git a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/a9e24316-a846-4e3c-8b20-52234859609e.md b/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/a9e24316-a846-4e3c-8b20-52234859609e.md
deleted file mode 100644
index d3b8c27..0000000
--- a/apps/api/storage/notebooks/9a2371fa-8a99-4436-b330-f35bf43c893c/a9e24316-a846-4e3c-8b20-52234859609e.md
+++ /dev/null
@@ -1,240 +0,0 @@
-# LyraNote 本科毕设全局设计说明（论文版）
-
-> 文档用途：用于本科毕业论文中的“研究背景、系统设计、技术特点、关键难点与创新点”章节初稿。
-> 
-> 版本日期：2026-03-26
-> 
-> 适用对象：论文作者、答辩评审、项目协作者
-
----
-
-## 摘要
-
-LyraNote 是一个面向个人研究与知识管理场景的 AI 驱动系统，目标是将传统“知识存储工具”升级为“可持续协作的研究伙伴”。系统围绕“私有知识优先”的设计原则，整合了 RAG 检索增强、多 Agent 编排、长期记忆、主动感知、深度研究与可视化生成界面（GenUI）等能力，形成从信息采集、理解、生成到反馈演化的闭环。
-
-与常见 AI 笔记产品相比，LyraNote 不仅提供被动问答，还重点探索 AI 的主动性与持续性：在用户未显式提问时，系统仍可基于上下文状态进行低干扰主动辅助；在长周期任务中，系统支持定时调度与自动投递；在多轮交互中，系统通过多层记忆与画像机制持续提升个性化质量。该系统具有较强工程实现价值与研究讨论价值，适合作为“智能知识系统 + Agent 工程化”方向的本科毕设课题。
-
----
-
-## 1. 研究背景与意义
-
-### 1.1 研究背景
-
-当前主流知识管理工具（笔记类应用、文档类应用）普遍存在两类不足：
-
-1. 知识层面：多来源信息可存储但难以深度理解，跨文档关联弱，检索结果与用户真实任务脱节。
-2. AI 层面：大多停留在“你问我答”的被动模式，缺乏持续状态感知、长期个性化与任务自动执行能力。
-
-与此同时，通用大模型虽然具备强生成能力，但在真实研究流程中仍面临幻觉、上下文窗口限制、长任务中断、可解释性不足、部署复杂等问题。基于此，LyraNote 以“面向研究场景的 AI 原生第二大脑”为目标，尝试把 LLM 能力转化为可持续、可控、可落地的系统能力。
-
-### 1.2 研究意义
-
-1. 应用意义：提高个人研究与学习效率，减少“搜集-整理-输出”的重复劳动。
-2. 工程意义：构建可扩展的 AI 系统架构，验证多 Agent、记忆系统、事件驱动链路在真实产品中的可行性。
-3. 学术意义：为“从被动对话到主动协作”的 AI Agent 产品化路径提供可复用方法与实验设计。
-
----
-
-## 2. 研究目标与问题定义
-
-### 2.1 总体目标
-
-构建一个支持“知识摄取—语义检索—深度研究—结构化表达—长期演化”的 AI 研究笔记系统，实现 AI 从工具属性向伙伴属性的演进。
-
-### 2.2 核心研究问题
-
-1. 如何让 AI 回答建立在用户私有知识之上，而不是泛化互联网知识？
-2. 如何在复杂任务中提升质量与稳定性，避免单 Agent 架构的能力拥挤？
-3. 如何让 AI 在不打扰用户的前提下实现主动感知与主动辅助？
-4. 如何在工程上实现可扩展、可维护、可自托管的系统架构？
-
----
-
-## 3. 系统总体架构设计
-
-### 3.1 分层总体架构
-
-LyraNote 采用“前端交互层 + 后端服务层 + AI 能力层 + 基础设施层”的分层设计：
-
-1. 前端交互层（Next.js + Tiptap + SSE）
-用于笔记编辑、对话交互、研究进度可视化、结构化内容展示。
-2. 后端服务层（FastAPI）
-负责路由编排、鉴权、业务规则与 API 暴露。
-3. AI 能力层（Agents + Skills + Memory）
-负责检索、写作、深度研究、工具调用、记忆更新与主动洞察。
-4. 基础设施层（PostgreSQL/pgvector + Redis + Celery + S3 兼容存储）
-负责结构化数据、向量索引、异步任务、对象存储与调度。
-
-### 3.2 关键数据闭环
-
-系统形成“源→知识→对话→笔记→再入库”的闭环：
-
-1. 用户导入 PDF/网页/Markdown。
-2. 后端异步完成解析、分块、向量化并入库（RAG 索引）。
-3. 对话阶段基于检索证据生成回答并附引用。
-4. 用户将结果沉淀为笔记、摘要或报告（Artifact）。
-5. 笔记可再次作为来源进入知识库（Note-as-Source），形成持续演化。
-
-### 3.3 AI 核心架构
-
-#### 3.3.1 多 Agent 协作架构（System A）
-
-采用“主 Agent（Orchestrator）+ 专家 Agent（RAG/Research/Writing/Memory/Web）”的两层决策：
-
-1. 主 Agent 负责任务级路由（把问题交给最合适的专家）。
-2. 专家 Agent 负责步骤级执行（检索、搜索、写作、记忆更新等）。
-
-该设计缓解了单 ReAct 循环能力拥挤、复杂任务步骤不足的问题。
-
-#### 3.3.2 Soul 持续思维与表达机制（System B）
-
-通过活动感知、后台思维循环和表达门控机制，让 AI 在用户静默期仍可进行低频高价值思考，并在合适时机以可忽略的方式推送洞察。
-
-#### 3.3.3 用户画像与记忆体系（System C）
-
-系统从“碎片记忆”升级到“多层记忆 + 用户画像”：
-
-1. 记忆层存储偏好、事实、场景与反思信息。
-2. 画像层按周期进行综合，形成对用户研究方向、能力水平、表达偏好的立体表征。
-3. 画像结果回注到路由决策与回答策略，实现长期个性化。
-
----
-
-## 4. 核心特点（面向论文阐述）
-
-### 4.1 私有知识优先的 RAG 管道
-
-系统在 Query 改写、多路检索、混合排序、重排与去重环节进行优化，重点解决多轮指代、召回不足和内容冗余问题。根据优化设计预期：
-
-1. 召回率提升约 25%～35%。
-2. 精确率提升约 20%。
-3. 首次响应延迟下降约 30%。
-
-### 4.2 深度研究（Deep Research）能力
-
-区别于单轮问答，深度研究支持“规划→检索→递归扩展→综合成文”的多阶段流程，支持 Quick/Deep 两种预算模式，并通过 SSE 反馈研究过程。为提升鲁棒性，研究任务与连接流解耦，支持刷新恢复与断点续传。
-
-### 4.3 GenUI 结构化表达能力
-
-通过统一 `genui` 协议将模型输出映射为可交互组件（图表、表格、时间轴、矩阵、看板等），提升研究结果可读性与可操作性，使 AI 输出从“纯文本”升级为“可视化知识单元”。
-
-### 4.4 主动感知与自动化执行
-
-系统提出“微交互渗透→上下文智能→自主行动”的三层主动架构：
-
-1. 在索引完成、编辑停顿、页面切换等时机提供低打扰提示。
-2. 通过定时任务实现持续任务自动执行（采集、生成、投递）。
-3. 将 AI 从一次性交互扩展为长期运行助手。
-
-### 4.5 可自托管与工程可扩展性
-
-1. 认证从外部 Clerk 迁移到本地单用户方案（bcrypt + JWT + 初始化向导），降低部署门槛。
-2. 存储层引入 `StorageProvider` 抽象，统一本地与 S3 兼容后端，支持横向扩展与多环境部署。
-3. 后端引入分层重构思路（Router/Service/Modules/Infrastructure），提升可维护性与测试性。
-
----
-
-## 5. 关键难点与解决思路
-
-### 难点 1：检索质量与多轮语义理解
-
-问题：用户问题口语化、上下文省略、语义跨度大，导致检索召回不稳定。  
-解决：引入对话感知 Query 改写、多 Query 并行、向量+全文混合检索、MMR 去重与 Cross-Encoder 重排。
-
-### 难点 2：深度研究任务长链路稳定性
-
-问题：若研究流程与 SSE 连接强耦合，用户刷新页面会导致任务中断。  
-解决：任务后台化（独立运行）+ 事件缓冲 + 状态查询，支持断点续传与结果恢复。
-
-### 难点 3：上下文窗口与成本控制
-
-问题：深度研究在多轮检索后易超出上下文窗口，且联网搜索有调用成本。  
-解决：学习结果摘要压缩、结果数量上限、检索阈值触发 Web Search、关键词去重与模式化预算（Quick/Deep）。
-
-### 难点 4：主动性与用户体验平衡
-
-问题：主动提示过多会造成干扰。  
-解决：采用“适时、适量、可忽略”原则，限制触发频率与展示密度，允许用户关闭或忽略。
-
-### 难点 5：系统复杂度持续上升
-
-问题：功能扩展导致路由臃肿、职责混乱、可维护性下降。  
-解决：推进 Service 分层、模块化 Agents、统一 Provider 接口、Schema 规范化，降低耦合度。
-
-### 难点 6：自托管部署与多环境兼容
-
-问题：文件存储、认证依赖、异步任务在不同部署环境中易出兼容性问题。  
-解决：认证本地化、存储抽象化、任务队列标准化（Celery + Redis）、配置集中化管理。
-
----
-
-## 6. 论文可主张的创新点
-
-1. 提出并实现“多 Agent 协作 + Soul 持续感知 + 用户画像演化”的三系统耦合架构。
-2. 将 AI 交互从“被动问答”扩展到“主动感知与任务执行”的产品形态。
-3. 在研究报告场景中实现 Deep Research 的工程化闭环（多阶段推理、可恢复任务、可复用产物）。
-4. 提出并落地面向生成式应用的统一 GenUI 协议，增强 LLM 输出可视化表达能力。
-5. 结合记忆与技能系统，实现“能力可插拔 + 个性可演化”的长期 Agent 机制。
-
----
-
-## 7. 论文实验与评估建议（可直接写入实验章节）
-
-### 7.1 建议评估维度
-
-1. 检索效果：Recall@K、Precision@K、nDCG。
-2. 回答质量：事实一致性、引用覆盖率、结构完整度。
-3. 任务能力：复杂任务完成率、平均完成时长、失败重试率。
-4. 主动交互：建议点击率、忽略率、用户主观满意度。
-5. 个性化效果：开启/关闭记忆后的质量差异、长期 `quality_score` 变化。
-6. 工程性能：首 token 延迟、平均响应时延、任务吞吐量、资源消耗。
-
-### 7.2 建议对比实验
-
-1. 单 Agent vs 多 Agent（复杂任务完成率对比）。
-2. 无记忆 vs V1 记忆 vs V2 记忆（个性化质量对比）。
-3. 被动模式 vs 主动模式（交互效率与满意度对比）。
-4. 纯文本输出 vs GenUI 输出（信息理解效率与可用性对比）。
-
----
-
-## 8. 局限性与后续工作
-
-1. 深度研究仍依赖外部搜索服务，成本与稳定性受第三方接口影响。
-2. 多 Agent 路由效果对 Prompt 与结构化输出稳定性敏感。
-3. 主动机制目前主要面向单用户场景，多用户协作机制有待拓展。
-4. 评估体系仍需更大规模用户实验来验证泛化性。
-
-后续可重点推进：
-
-1. 更细粒度的路由评估与自动纠偏机制。
-2. 更强的多模态研究能力（图像/表格/代码执行）。
-3. 研究任务模板化与可复现实验流水线。
-4. 面向团队协作的多租户与权限机制。
-
----
-
-## 9. 文档依据（可作为论文内部参考来源）
-
-1. `architecture.md`（系统总架构、AI 流水线）
-2. `lyra-soul-system.md`（多 Agent + Soul + 画像）
-3. `memory-system-v2.md`（五层记忆体系）
-4. `skills-system.md`（可插拔技能架构）
-5. `deep-research.md`（多阶段研究流程与任务解耦）
-6. `rag-optimization.md`（检索优化链路）
-7. `genui-integration.md`（GenUI 协议）
-8. `proactive-ai-system.md`（主动感知体系）
-9. `scheduled-tasks.md`（定时任务自动化）
-10. `backend-architecture-refactor.md`（后端分层重构）
-11. `storage-system.md`（存储抽象）
-12. `single-user-auth.md`（单用户自托管认证）
-
----
-
-## 附：论文写作使用建议
-
-1. 可将第 1-3 节作为“绪论 + 总体设计”。
-2. 可将第 4-5 节拆为“关键技术实现”。
-3. 可将第 6-7 节作为“创新点与实验设计”。
-4. 可将第 8 节作为“总结与展望”。
-
diff --git a/patch_home.js b/patch_home.js
deleted file mode 100644
index 96c1888..0000000
--- a/patch_home.js
+++ /dev/null
@@ -1,140 +0,0 @@
-const fs = require("fs");
-const path = require("path");
-
-const filePath = path.join(process.cwd(), "apps/desktop/src/pages/home/home-page.tsx");
-let content = fs.readFileSync(filePath, "utf-8");
-
-// replace SUGGESTIONS
-content = content.replace(
-  /const SUGGESTIONS = \[\s*"分析知识库核心主题",\s*"生成结构化研究摘要",\s*"对比不同来源观点",\s*"根据笔记制定学习计划",\s*\]/g,
-  `const SUGGESTIONS = [
-  { icon: <Sparkles size={14} className="text-[var(--color-accent)] opacity-80" />, text: "帮我分析知识库中的核心主题" },
-  { icon: <Sparkles size={14} className="text-[var(--color-accent)] opacity-80" />, text: "为我的研究生成一份结构化摘要" },
-  { icon: <Sparkles size={14} className="text-[var(--color-accent)] opacity-80" />, text: "对比不同来源中的相似观点" },
-  { icon: <Sparkles size={14} className="text-[var(--color-accent)] opacity-80" />, text: "根据笔记内容生成学习计划" },
-]`
-);
-
-// replace render
-content = content.replace(
-  /return \(\s*<div className="flex flex-col items-center justify-center h-full px-6 pb-12">[\s\S]*?\)\n\}/,
-  \`return (
-    <div className="flex flex-col items-center justify-center h-full px-6 pb-12 gap-8">
-      {/* Title Area */}
-      <motion.div
-        initial={{ opacity: 0, y: 12 }}
-        animate={{ opacity: 1, y: 0 }}
-        transition={{ ...springs.gentle, delay: 0.05 }}
-        className="flex flex-col items-center gap-4 text-center mt-8"
-      >
-        <div className="w-11 h-11 rounded-full flex items-center justify-center flex-shrink-0 overflow-hidden ring-1 ring-white/10 bg-white/5 drop-shadow-xl">
-          <img src="/bot_avatar.png" alt="AI Avatar" className="w-full h-full object-cover" />
-        </div>
-        <div>
-          <h2 className="text-[17px] font-semibold" style={{ color: "var(--color-text-primary)", letterSpacing: "-0.01em" }}>
-            有什么我可以帮你的？
-          </h2>
-          <p className="text-[12px] mt-1.5" style={{ color: "var(--color-text-tertiary)" }}>
-            基于你的知识库，我可以帮你分析、总结和探索任何内容
-          </p>
-        </div>
-      </motion.div>
-
-      {/* Suggestions Grid */}
-      <motion.div
-        initial={{ opacity: 0 }}
-        animate={{ opacity: 1 }}
-        transition={{ delay: 0.1, duration: 0.4 }}
-        className="grid grid-cols-2 gap-2 w-full max-w-[560px]"
-      >
-        {SUGGESTIONS.map((s, idx) => (
-          <motion.button
-            key={idx}
-            initial={{ opacity: 0, y: 6 }}
-            animate={{ opacity: 1, y: 0 }}
-            transition={{ ...springs.bouncy, delay: 0.15 + idx * 0.05 }}
-            whileHover={{ scale: 1.01, y: -1 }}
-            whileTap={{ scale: 0.98 }}
-            onClick={() => handleSuggestion(s.text)}
-            className="flex items-center gap-2.5 px-4 py-3.5 rounded-xl text-left text-[12.5px] transition-colors"
-            style={{
-              background: "rgba(255,255,255,0.015)",
-              border: "1px solid rgba(255,255,255,0.05)",
-              color: "var(--color-text-secondary)",
-            }}
-          >
-            <span className="shrink-0">{s.icon}</span>
-            <span className="leading-snug truncate">{s.text}</span>
-          </motion.button>
-        ))}
-      </motion.div>
-
-      {/* Input card */}
-      <motion.div
-        initial={{ opacity: 0, y: 16, scale: 0.98 }}
-        animate={{ opacity: 1, y: 0, scale: 1 }}
-        transition={{ ...springs.smooth, delay: 0.2 }}
-        className="w-full max-w-[720px] mt-4"
-      >
-        <motion.div
-          animate={{
-            boxShadow: focused
-              ? "0 0 0 1.5px rgba(124,110,247,0.5), 0 8px 32px rgba(0,0,0,0.4)"
-              : "0 0 0 1px rgba(255,255,255,0.08), 0 4px 16px rgba(0,0,0,0.25)",
-          }}
-          transition={springs.snappy}
-          className="rounded-[20px] overflow-hidden"
-          style={{ background: "var(--color-bg-elevated)", border: "none" }}
-          onClick={() => textareaRef.current?.focus()}
-        >
-          <textarea
-            ref={textareaRef}
-            value={input}
-            onChange={handleInput}
-            onKeyDown={(e) => {
-              if (e.key === "Enter" && !e.shiftKey) { e.preventDefault(); handleSubmit() }
-            }}
-            onFocus={() => setFocused(true)}
-            onBlur={() => setFocused(false)}
-            placeholder="向 AI 提问，或描述你想探索的内容…"
-            rows={1}
-            className="w-full bg-transparent outline-none resize-none text-[14px] leading-relaxed px-5 pt-4 pb-12 no-scrollbar"
-            style={{
-              color: "var(--color-text-primary)",
-              fontFamily: "var(--font-sans)",
-              minHeight: "44px",
-              caretColor: "var(--color-accent)",
-            }}
-          />
-          <div className="absolute bottom-2 left-2 right-2 flex items-center justify-between pointer-events-none">
-            <div className="relative flex items-center gap-1.5 pl-2 pb-1 pointer-events-auto">
-              <button className="flex items-center justify-center w-[30px] h-[30px] rounded-lg text-[var(--color-text-tertiary)] hover:text-[var(--color-text-primary)] hover:bg-[#ffffff12] transition-colors">
-                <Paperclip size={15} strokeWidth={1.5} className="opacity-70" />
-              </button>
-            </div>
-            <div className="flex items-center gap-2.5 pr-2 pb-1 pointer-events-auto">
-              <span className="text-[11px] text-[var(--color-text-tertiary)] opacity-60 select-none mr-1">
-                按 Enter 发送 · Shift+Enter 换行
-              </span>
-              <motion.button
-                whileTap={{ scale: 0.9 }}
-                onClick={handleSubmit}
-                disabled={!input.trim()}
-                className="flex items-center justify-center w-[30px] h-[30px] rounded-full transition-all disabled:opacity-30 hover:bg-[#ffffff12]"
-                style={{
-                  background: input.trim() ? "var(--color-accent)" : "transparent",
-                  color: input.trim() ? "white" : "var(--color-text-secondary)",
-                }}
-              >
-                <ArrowUp size={16} strokeWidth={2} />
-              </motion.button>
-            </div>
-          </div>
-        </motion.div>
-      </motion.div>
-    </div>
-  )
-}\`
-);
-
-fs.writeFileSync(filePath, content, "utf-8");
diff --git a/scripts/create_db_table_design_docx.py b/scripts/create_db_table_design_docx.py
deleted file mode 100644
index 95607a8..0000000
--- a/scripts/create_db_table_design_docx.py
+++ /dev/null
@@ -1,470 +0,0 @@
-from pathlib import Path
-
-from docx import Document
-from docx.enum.table import WD_CELL_VERTICAL_ALIGNMENT, WD_ROW_HEIGHT_RULE, WD_TABLE_ALIGNMENT
-from docx.enum.text import WD_ALIGN_PARAGRAPH
-from docx.oxml import OxmlElement
-from docx.oxml.ns import qn
-from docx.shared import Cm, Mm, Pt
-
-
-HEADERS = ["字段名称", "类型", "长度", "字段说明"]
-COLUMN_WIDTHS_CM = (4.2, 2.1, 1.6, 7.1)
-PRIMARY_OUTPUT_DOCX = Path("output/doc/LyraNote数据库表设计.docx")
-VERSIONED_OUTPUT_DOCX = Path("output/doc/LyraNote数据库表设计-三线表-最新模型.docx")
-
-TABLES = [
-    (
-        "表3-1 用户信息表（users）",
-        [
-            ("id", "UUID", "36", "用户主键"),
-            ("username", "varchar", "255", "用户名"),
-            ("password_hash", "varchar", "255", "密码哈希"),
-            ("email", "varchar", "255", "邮箱"),
-            ("name", "varchar", "255", "用户姓名/昵称"),
-            ("avatar_url", "text", "-", "用户头像地址"),
-            ("google_id", "varchar", "255", "Google 登录标识"),
-            ("github_id", "varchar", "255", "GitHub 登录标识"),
-            ("oauth_unbound", "varchar", "64", "已解绑第三方登录标记"),
-            ("created_at", "datetime", "-", "创建时间"),
-        ],
-    ),
-    (
-        "表3-2 应用配置表（app_config）",
-        [
-            ("key", "varchar", "255", "配置项主键"),
-            ("value", "text", "-", "配置项值"),
-        ],
-    ),
-    (
-        "表3-3 笔记本信息表（notebooks）",
-        [
-            ("id", "UUID", "36", "笔记本主键"),
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("title", "varchar", "500", "笔记本标题"),
-            ("description", "text", "-", "笔记本描述"),
-            ("status", "varchar", "50", "笔记本状态"),
-            ("is_global", "boolean", "-", "是否全局知识库"),
-            ("is_system", "boolean", "-", "是否系统笔记本"),
-            ("system_type", "varchar", "50", "系统笔记本类型"),
-            ("source_count", "integer", "-", "来源数量"),
-            ("is_public", "boolean", "-", "是否公开"),
-            ("published_at", "datetime", "-", "发布时间"),
-            ("cover_emoji", "varchar", "10", "封面表情"),
-            ("cover_gradient", "varchar", "50", "封面渐变样式"),
-            ("created_at", "datetime", "-", "创建时间"),
-            ("updated_at", "datetime", "-", "更新时间"),
-        ],
-    ),
-    (
-        "表3-4 笔记本摘要表（notebook_summaries）",
-        [
-            ("notebook_id", "UUID", "36", "所属笔记本 ID"),
-            ("summary_md", "text", "-", "Markdown 摘要内容"),
-            ("key_themes", "json", "-", "关键主题列表"),
-            ("last_synced_at", "datetime", "-", "最近同步时间"),
-        ],
-    ),
-    (
-        "表3-5 知识来源表（sources）",
-        [
-            ("id", "UUID", "36", "来源主键"),
-            ("notebook_id", "UUID", "36", "所属笔记本 ID"),
-            ("title", "varchar", "500", "来源标题"),
-            ("type", "varchar", "50", "来源类型"),
-            ("status", "varchar", "50", "索引状态"),
-            ("file_path", "text", "-", "本地文件路径"),
-            ("url", "text", "-", "网页地址"),
-            ("raw_text", "text", "-", "抽取后的原始文本"),
-            ("summary", "text", "-", "来源摘要"),
-            ("storage_key", "varchar", "500", "对象存储键"),
-            ("storage_backend", "varchar", "20", "存储后端类型"),
-            ("created_at", "datetime", "-", "创建时间"),
-            ("updated_at", "datetime", "-", "更新时间"),
-        ],
-    ),
-    (
-        "表3-6 知识分块表（chunks）",
-        [
-            ("id", "UUID", "36", "分块主键"),
-            ("source_id", "UUID", "36", "所属来源 ID"),
-            ("notebook_id", "UUID", "36", "所属笔记本 ID"),
-            ("content", "text", "-", "分块正文"),
-            ("chunk_index", "integer", "-", "分块顺序号"),
-            ("embedding", "vector", "-", "向量表示"),
-            ("token_count", "integer", "-", "Token 数量"),
-            ("metadata", "json", "-", "分块元数据"),
-            ("created_at", "datetime", "-", "创建时间"),
-            ("source_type", "varchar", "20", "分块来源类型"),
-            ("note_id", "UUID", "36", "关联笔记 ID"),
-        ],
-    ),
-    (
-        "表3-7 对话信息表（conversations）",
-        [
-            ("id", "UUID", "36", "对话主键"),
-            ("notebook_id", "UUID", "36", "关联笔记本 ID"),
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("title", "varchar", "500", "对话标题"),
-            ("source", "varchar", "20", "对话来源"),
-            ("created_at", "datetime", "-", "创建时间"),
-        ],
-    ),
-    (
-        "表3-8 消息信息表（messages）",
-        [
-            ("id", "UUID", "36", "消息主键"),
-            ("conversation_id", "UUID", "36", "所属对话 ID"),
-            ("generation_id", "UUID", "36", "关联生成任务 ID"),
-            ("role", "varchar", "50", "消息角色"),
-            ("status", "varchar", "20", "消息状态"),
-            ("content", "text", "-", "消息正文"),
-            ("reasoning", "text", "-", "推理内容"),
-            ("citations", "json", "-", "引用信息"),
-            ("agent_steps", "json", "-", "Agent 执行步骤"),
-            ("attachments", "json", "-", "附件信息"),
-            ("speed", "json", "-", "流式速度指标"),
-            ("mind_map", "json", "-", "思维导图结果"),
-            ("diagram", "json", "-", "图形结果"),
-            ("mcp_result", "json", "-", "MCP 工具结果"),
-            ("ui_elements", "json", "-", "生成式界面元素"),
-            ("parent_message_id", "UUID", "36", "父消息 ID"),
-            ("created_at", "datetime", "-", "创建时间"),
-        ],
-    ),
-    (
-        "表3-9 对话摘要表（conversation_summaries）",
-        [
-            ("conversation_id", "UUID", "36", "所属对话 ID"),
-            ("summary_text", "text", "-", "压缩后的对话摘要"),
-            ("compressed_message_count", "integer", "-", "已压缩消息数量"),
-            ("compressed_through", "datetime", "-", "摘要覆盖截止时间"),
-            ("updated_at", "datetime", "-", "更新时间"),
-        ],
-    ),
-    (
-        "表3-10 深度研究任务表（research_tasks）",
-        [
-            ("id", "UUID", "36", "任务主键"),
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("notebook_id", "varchar", "100", "关联笔记本 ID"),
-            ("conversation_id", "UUID", "36", "关联对话 ID"),
-            ("query", "text", "-", "研究问题"),
-            ("mode", "varchar", "20", "研究模式"),
-            ("status", "varchar", "20", "任务状态"),
-            ("report", "text", "-", "研究报告"),
-            ("deliverable_json", "json", "-", "结构化交付物"),
-            ("timeline_json", "json", "-", "过程时间线"),
-            ("web_sources_json", "json", "-", "外部资料列表"),
-            ("error_message", "text", "-", "错误信息"),
-            ("created_at", "datetime", "-", "创建时间"),
-            ("completed_at", "datetime", "-", "完成时间"),
-        ],
-    ),
-    (
-        "表3-11 定时任务表（scheduled_tasks）",
-        [
-            ("id", "UUID", "36", "定时任务主键"),
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("name", "varchar", "255", "任务名称"),
-            ("description", "text", "-", "任务描述"),
-            ("task_type", "varchar", "50", "任务类型"),
-            ("schedule_cron", "varchar", "100", "调度表达式"),
-            ("timezone", "varchar", "50", "时区"),
-            ("parameters", "json", "-", "任务参数"),
-            ("delivery_config", "json", "-", "投递配置"),
-            ("enabled", "boolean", "-", "是否启用"),
-            ("last_run_at", "datetime", "-", "上次执行时间"),
-            ("next_run_at", "datetime", "-", "下次执行时间"),
-            ("run_count", "integer", "-", "执行次数"),
-            ("last_result", "text", "-", "最近执行结果"),
-            ("last_error", "text", "-", "最近错误信息"),
-            ("consecutive_failures", "integer", "-", "连续失败次数"),
-            ("created_at", "datetime", "-", "创建时间"),
-            ("updated_at", "datetime", "-", "更新时间"),
-        ],
-    ),
-    (
-        "表3-12 定时任务执行记录表（scheduled_task_runs）",
-        [
-            ("id", "UUID", "36", "执行记录主键"),
-            ("task_id", "UUID", "36", "所属定时任务 ID"),
-            ("status", "varchar", "20", "执行状态"),
-            ("started_at", "datetime", "-", "开始时间"),
-            ("finished_at", "datetime", "-", "结束时间"),
-            ("duration_ms", "integer", "-", "执行耗时（毫秒）"),
-            ("result_summary", "text", "-", "结果摘要"),
-            ("error_message", "text", "-", "错误信息"),
-            ("generated_content", "text", "-", "生成内容"),
-            ("sources_count", "integer", "-", "涉及来源数量"),
-            ("delivery_status", "json", "-", "投递状态"),
-        ],
-    ),
-    (
-        "表3-13 用户记忆表（user_memories）",
-        [
-            ("id", "UUID", "36", "用户记忆主键"),
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("key", "varchar", "100", "记忆键"),
-            ("value", "text", "-", "记忆值"),
-            ("confidence", "float", "-", "置信度"),
-            ("memory_type", "varchar", "20", "记忆类型"),
-            ("memory_kind", "varchar", "32", "记忆类别"),
-            ("access_count", "integer", "-", "访问次数"),
-            ("last_accessed_at", "datetime", "-", "最后访问时间"),
-            ("expires_at", "datetime", "-", "过期时间"),
-            ("reinforced_by", "varchar", "36", "强化来源反思 ID"),
-            ("updated_at", "datetime", "-", "更新时间"),
-            ("embedding", "vector", "-", "记忆向量"),
-            ("source", "varchar", "20", "记忆来源类型"),
-            ("evidence", "text", "-", "证据链或来源记录"),
-            ("conflict_flag", "boolean", "-", "冲突标记"),
-        ],
-    ),
-    (
-        "表3-14 用户画像表（user_portraits）",
-        [
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("portrait_json", "json", "-", "六维画像 JSON"),
-            ("avatar_url", "text", "-", "画像头像地址"),
-            ("synthesis_summary", "text", "-", "画像合成摘要"),
-            ("version", "integer", "-", "画像版本号"),
-            ("synthesized_at", "datetime", "-", "合成时间"),
-            ("created_at", "datetime", "-", "创建时间"),
-            ("updated_at", "datetime", "-", "更新时间"),
-        ],
-    ),
-    (
-        "表3-15 Agent 思想记录表（agent_thoughts）",
-        [
-            ("id", "UUID", "36", "思想记录主键"),
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("visibility", "varchar", "20", "可见性状态"),
-            ("content", "text", "-", "思想内容"),
-            ("activity_context", "json", "-", "活动上下文快照"),
-            ("notebook_id", "UUID", "36", "关联笔记本 ID"),
-            ("created_at", "datetime", "-", "创建时间"),
-        ],
-    ),
-    (
-        "表3-16 主动洞察表（proactive_insights）",
-        [
-            ("id", "UUID", "36", "洞察主键"),
-            ("user_id", "UUID", "36", "所属用户 ID"),
-            ("notebook_id", "UUID", "36", "关联笔记本 ID"),
-            ("insight_type", "varchar", "50", "洞察类型"),
-            ("title", "varchar", "500", "洞察标题"),
-            ("content", "text", "-", "洞察内容"),
-            ("metadata", "json", "-", "洞察附加元数据"),
-            ("is_read", "boolean", "-", "是否已读"),
-            ("created_at", "datetime", "-", "创建时间"),
-        ],
-    ),
-]
-
-
-def set_east_asia_font(run, font_name: str) -> None:
-    run.font.name = font_name
-    run._element.rPr.rFonts.set(qn("w:eastAsia"), font_name)
-
-
-def set_doc_fonts(doc: Document) -> None:
-    normal = doc.styles["Normal"]
-    normal.font.name = "宋体"
-    normal.font.size = Pt(12)
-    normal._element.rPr.rFonts.set(qn("w:eastAsia"), "宋体")
-
-
-def set_cell_margins(cell, top: int = 40, start: int = 60, bottom: int = 40, end: int = 60) -> None:
-    tc_pr = cell._tc.get_or_add_tcPr()
-    tc_mar = tc_pr.first_child_found_in("w:tcMar")
-    if tc_mar is None:
-        tc_mar = OxmlElement("w:tcMar")
-        tc_pr.append(tc_mar)
-
-    for edge, value in {"top": top, "start": start, "bottom": bottom, "end": end}.items():
-        element = tc_mar.find(qn(f"w:{edge}"))
-        if element is None:
-            element = OxmlElement(f"w:{edge}")
-            tc_mar.append(element)
-        element.set(qn("w:w"), str(value))
-        element.set(qn("w:type"), "dxa")
-
-
-def set_cell_border(cell, **borders) -> None:
-    tc_pr = cell._tc.get_or_add_tcPr()
-    tc_borders = tc_pr.first_child_found_in("w:tcBorders")
-    if tc_borders is None:
-        tc_borders = OxmlElement("w:tcBorders")
-        tc_pr.append(tc_borders)
-
-    for edge, data in borders.items():
-        element = tc_borders.find(qn(f"w:{edge}"))
-        if element is None:
-            element = OxmlElement(f"w:{edge}")
-            tc_borders.append(element)
-        for key, value in data.items():
-            element.set(qn(f"w:{key}"), str(value))
-
-
-def clear_cell_borders(cell) -> None:
-    nil = {"val": "nil"}
-    set_cell_border(cell, top=nil, bottom=nil, left=nil, right=nil)
-
-
-def set_table_borders(table) -> None:
-    tbl_pr = table._tbl.tblPr
-    tbl_borders = tbl_pr.first_child_found_in("w:tblBorders")
-    if tbl_borders is None:
-        tbl_borders = OxmlElement("w:tblBorders")
-        tbl_pr.append(tbl_borders)
-
-    styles = {
-        "top": {"val": "single", "sz": "12", "space": "0", "color": "000000"},
-        "bottom": {"val": "single", "sz": "12", "space": "0", "color": "000000"},
-        "left": {"val": "nil"},
-        "right": {"val": "nil"},
-        "insideH": {"val": "nil"},
-        "insideV": {"val": "nil"},
-    }
-    for edge, data in styles.items():
-        element = tbl_borders.find(qn(f"w:{edge}"))
-        if element is None:
-            element = OxmlElement(f"w:{edge}")
-            tbl_borders.append(element)
-        for key, value in data.items():
-            element.set(qn(f"w:{key}"), str(value))
-
-
-def set_row_cant_split(row) -> None:
-    tr_pr = row._tr.get_or_add_trPr()
-    if tr_pr.find(qn("w:cantSplit")) is None:
-        tr_pr.append(OxmlElement("w:cantSplit"))
-
-
-def set_cell_width(cell, width_cm: float) -> None:
-    cell.width = Cm(width_cm)
-    tc_pr = cell._tc.get_or_add_tcPr()
-    tc_w = tc_pr.first_child_found_in("w:tcW")
-    if tc_w is None:
-        tc_w = OxmlElement("w:tcW")
-        tc_pr.append(tc_w)
-    tc_w.set(qn("w:w"), str(int(Cm(width_cm).twips)))
-    tc_w.set(qn("w:type"), "dxa")
-
-
-def set_cell_text(cell, text: str, *, bold: bool = False) -> None:
-    cell.text = ""
-    cell.vertical_alignment = WD_CELL_VERTICAL_ALIGNMENT.CENTER
-    paragraph = cell.paragraphs[0]
-    paragraph.alignment = WD_ALIGN_PARAGRAPH.CENTER
-    paragraph.paragraph_format.space_before = Pt(0)
-    paragraph.paragraph_format.space_after = Pt(0)
-    paragraph.paragraph_format.line_spacing = 1.0
-    run = paragraph.add_run(text)
-    run.bold = bold
-    run.font.size = Pt(10.5)
-    set_east_asia_font(run, "宋体")
-
-
-def format_three_line_table(table) -> None:
-    table.alignment = WD_TABLE_ALIGNMENT.CENTER
-    table.autofit = False
-    set_table_borders(table)
-
-    for row in table.rows:
-        set_row_cant_split(row)
-        row.height_rule = WD_ROW_HEIGHT_RULE.AT_LEAST
-        row.height = Cm(0.8)
-        for index, cell in enumerate(row.cells):
-            set_cell_width(cell, COLUMN_WIDTHS_CM[index])
-            set_cell_margins(cell)
-            clear_cell_borders(cell)
-
-    header_border = {"val": "single", "sz": "8", "space": "0", "color": "000000"}
-    for cell in table.rows[0].cells:
-        set_cell_border(cell, bottom=header_border)
-
-
-def add_caption(doc: Document, title: str) -> None:
-    paragraph = doc.add_paragraph()
-    paragraph.alignment = WD_ALIGN_PARAGRAPH.CENTER
-    paragraph.paragraph_format.space_before = Pt(8)
-    paragraph.paragraph_format.space_after = Pt(4)
-    run = paragraph.add_run(title)
-    run.bold = True
-    run.font.size = Pt(12)
-    set_east_asia_font(run, "宋体")
-
-
-def add_intro_paragraph(doc: Document, text: str) -> None:
-    paragraph = doc.add_paragraph()
-    paragraph.paragraph_format.first_line_indent = Pt(24)
-    paragraph.paragraph_format.line_spacing = 1.5
-    paragraph.paragraph_format.space_after = Pt(4)
-    run = paragraph.add_run(text)
-    run.font.size = Pt(12)
-    set_east_asia_font(run, "宋体")
-
-
-def add_table(doc: Document, title: str, rows: list[tuple[str, str, str, str]]) -> None:
-    add_caption(doc, title)
-
-    table = doc.add_table(rows=len(rows) + 1, cols=len(HEADERS))
-    for index, header in enumerate(HEADERS):
-        set_cell_text(table.cell(0, index), header, bold=True)
-
-    for row_index, row_data in enumerate(rows, start=1):
-        for col_index, value in enumerate(row_data):
-            set_cell_text(table.cell(row_index, col_index), value)
-
-    format_three_line_table(table)
-    doc.add_paragraph("")
-
-
-def build_doc() -> Document:
-    doc = Document()
-    set_doc_fonts(doc)
-
-    section = doc.sections[0]
-    section.page_width = Mm(210)
-    section.page_height = Mm(297)
-    section.top_margin = Cm(2.54)
-    section.bottom_margin = Cm(2.54)
-    section.left_margin = Cm(3.0)
-    section.right_margin = Cm(3.0)
-
-    title = doc.add_paragraph()
-    title.alignment = WD_ALIGN_PARAGRAPH.LEFT
-    title.paragraph_format.space_after = Pt(10)
-    title_run = title.add_run("3.3.2 数据库表设计")
-    title_run.bold = True
-    title_run.font.size = Pt(14)
-    set_east_asia_font(title_run, "黑体")
-
-    add_intro_paragraph(
-        doc,
-        "根据 LyraNote 系统当前代码实现，数据库表主要围绕用户管理、知识组织、对话交互、研究任务、长期记忆和主动服务等几个方面展开设计。系统底层采用 PostgreSQL 作为核心关系型数据库，并结合 pgvector 扩展保存知识分块与长期记忆的向量表示。",
-    )
-    add_intro_paragraph(
-        doc,
-        "为体现当前版本的数据结构设计，本文选取用户、应用配置、笔记本、笔记本摘要、知识来源、知识分块、对话、消息、对话摘要、深度研究任务、定时任务、定时任务执行记录、用户记忆、用户画像、Agent 思想记录和主动洞察等 16 张核心数据表进行说明。各主要数据表设计如下所示。",
-    )
-
-    for title_text, rows in TABLES:
-        add_table(doc, title_text, rows)
-
-    return doc
-
-
-def main() -> None:
-    PRIMARY_OUTPUT_DOCX.parent.mkdir(parents=True, exist_ok=True)
-    doc = build_doc()
-    doc.save(PRIMARY_OUTPUT_DOCX)
-    doc.save(VERSIONED_OUTPUT_DOCX)
-    print(PRIMARY_OUTPUT_DOCX)
-    print(VERSIONED_OUTPUT_DOCX)
-
-
-if __name__ == "__main__":
-    main()
diff --git a/scripts/fix_thesis_er_diagrams.py b/scripts/fix_thesis_er_diagrams.py
deleted file mode 100644
index 13d1e14..0000000
--- a/scripts/fix_thesis_er_diagrams.py
+++ /dev/null
@@ -1,639 +0,0 @@
-from __future__ import annotations
-
-from dataclasses import dataclass
-from pathlib import Path
-from typing import Iterable
-
-from docx import Document
-from docx.enum.text import WD_ALIGN_PARAGRAPH
-from docx.oxml import OxmlElement
-from docx.shared import Inches
-from docx.text.paragraph import Paragraph
-from PIL import Image, ImageDraw, ImageFont
-
-
-REPO_ROOT = Path(__file__).resolve().parents[1]
-SOURCE_DOC = Path("/Users/kaihuang/Documents/毕业论文/220501020064-黄凯-毕业论文-引用修订版_4.docx")
-TMP_DIR = REPO_ROOT / "tmp" / "docs" / "thesis-er-diagram-cleanup"
-OUTPUT_DIR = REPO_ROOT / "output" / "doc"
-OUTPUT_DOC = OUTPUT_DIR / "220501020064-黄凯-毕业论文-引用修订版_4-ER图修订版.docx"
-
-
-@dataclass(frozen=True)
-class EntitySpec:
-    index: int
-    figure_no: str
-    entity_name: str
-    caption: str
-    description: str
-    attrs: tuple[str, ...]
-
-    @property
-    def image_name(self) -> str:
-        return f"{self.figure_no}-{self.entity_name}.png"
-
-
-ENTITY_SPECS: tuple[EntitySpec, ...] = (
-    EntitySpec(
-        index=1,
-        figure_no="3-6",
-        entity_name="用户",
-        caption="图3-6 用户实体 E-R 图",
-        description=(
-            "（1）用户实体 E-R 图如图3-6所示。图 3-6 展示了用户实体的 E-R 结构。"
-            "用户实体用于标识系统中的真实使用者，图中仅保留用户自身属性，"
-            "包括账号信息、认证标识、头像信息以及创建时间等内容；"
-            "与笔记本、会话、任务和长期记忆等实体之间的归属关系在后续关系约束中体现。"
-        ),
-        attrs=(
-            "用户ID(PK)",
-            "用户名",
-            "密码哈希",
-            "邮箱",
-            "姓名",
-            "头像地址",
-            "Google标识",
-            "GitHub标识",
-            "OAuth解绑标记",
-            "创建时间",
-        ),
-    ),
-    EntitySpec(
-        index=2,
-        figure_no="3-7",
-        entity_name="笔记本",
-        caption="图3-7 笔记本实体 E-R 图",
-        description=(
-            "（2）笔记本实体 E-R 图如图3-7所示。图 3-7 展示了笔记本实体的 E-R 结构。"
-            "笔记本是 LyraNote 组织私有知识的核心工作空间，"
-            "图中保留了标题、描述、状态、公开性与封面等实体自有属性，"
-            "而与用户、来源、对话和笔记等对象的关联不再作为属性列重复展示。"
-        ),
-        attrs=(
-            "笔记本ID(PK)",
-            "标题",
-            "描述",
-            "状态",
-            "是否全局知识库",
-            "是否系统笔记本",
-            "系统类型",
-            "来源数量",
-            "是否公开",
-            "发布时间",
-            "封面表情",
-            "封面渐变",
-            "创建时间",
-            "更新时间",
-        ),
-    ),
-    EntitySpec(
-        index=3,
-        figure_no="3-8",
-        entity_name="知识来源",
-        caption="图3-8 知识来源实体 E-R 图",
-        description=(
-            "（3）知识来源实体 E-R 图如图3-8所示。图 3-8 展示了知识来源实体的 E-R 结构。"
-            "知识来源记录用户导入的 PDF、网页和 Markdown 等资料，"
-            "实体自身属性主要描述来源类型、索引状态、文本内容、摘要和存储信息，"
-            "其与笔记本及知识分块之间的联系通过关系设计表达。"
-        ),
-        attrs=(
-            "来源ID(PK)",
-            "标题",
-            "来源类型",
-            "索引状态",
-            "文件路径",
-            "链接地址",
-            "原始文本",
-            "摘要",
-            "存储键",
-            "存储后端",
-            "创建时间",
-            "更新时间",
-        ),
-    ),
-    EntitySpec(
-        index=4,
-        figure_no="3-9",
-        entity_name="知识分块",
-        caption="图3-9 知识分块实体 E-R 图",
-        description=(
-            "（4）知识分块实体 E-R 图如图3-9所示。图 3-9 展示了知识分块实体的 E-R 结构。"
-            "知识分块是系统执行检索增强生成时使用的最小知识单元，"
-            "图中仅保留分块正文、顺序编号、向量嵌入、元数据和来源类型等自身属性，"
-            "不再将所属来源、笔记本或笔记的外键字段混入实体属性。"
-        ),
-        attrs=(
-            "分块ID(PK)",
-            "正文内容",
-            "分块序号",
-            "向量嵌入",
-            "Token数量",
-            "元数据",
-            "来源类型",
-            "创建时间",
-        ),
-    ),
-    EntitySpec(
-        index=5,
-        figure_no="3-10",
-        entity_name="对话",
-        caption="图3-10 对话实体 E-R 图",
-        description=(
-            "（5）对话实体 E-R 图如图3-10所示。图 3-10 展示了对话实体的 E-R 结构。"
-            "对话实体用于承载一次连续的交互会话，"
-            "其自身属性主要包括对话标题、来源场景和创建时间；"
-            "与用户、笔记本以及消息记录的关联关系通过实体联系表达，而不直接写入属性列表。"
-        ),
-        attrs=(
-            "对话ID(PK)",
-            "标题",
-            "对话来源",
-            "创建时间",
-        ),
-    ),
-    EntitySpec(
-        index=6,
-        figure_no="3-11",
-        entity_name="消息",
-        caption="图3-11 消息实体 E-R 图",
-        description=(
-            "（6）消息实体 E-R 图如图3-11所示。图 3-11 展示了消息实体的 E-R 结构。"
-            "消息实体是对话过程中的最小交互单元，"
-            "保存角色、状态、正文、引用、推理轨迹以及富媒体结果等内容；"
-            "对话归属、生成归属和父消息分支等外键信息不再作为本实体属性展示。"
-        ),
-        attrs=(
-            "消息ID(PK)",
-            "角色",
-            "状态",
-            "正文内容",
-            "推理轨迹",
-            "引用信息",
-            "Agent步骤",
-            "附件信息",
-            "速度指标",
-            "思维导图",
-            "图表数据",
-            "MCP结果",
-            "UI元素",
-            "创建时间",
-        ),
-    ),
-    EntitySpec(
-        index=7,
-        figure_no="3-12",
-        entity_name="深度研究任务",
-        caption="图3-12 深度研究任务实体 E-R 图",
-        description=(
-            "（7）深度研究任务实体 E-R 图如图3-12所示。图 3-12 展示了深度研究任务实体的 E-R 结构。"
-            "深度研究任务用于表示一次长生命周期的研究流程，"
-            "实体自身属性包括研究问题、执行模式、状态、报告正文、结构化交付物和时间线等，"
-            "与用户、会话和笔记本的关联通过关系约束体现。"
-        ),
-        attrs=(
-            "任务ID(PK)",
-            "研究问题",
-            "研究模式",
-            "状态",
-            "研究报告",
-            "交付物JSON",
-            "时间线JSON",
-            "外部资料列表",
-            "错误信息",
-            "创建时间",
-            "完成时间",
-        ),
-    ),
-    EntitySpec(
-        index=8,
-        figure_no="3-13",
-        entity_name="定时任务",
-        caption="图3-13 定时任务实体 E-R 图",
-        description=(
-            "（8）定时任务实体 E-R 图如图3-13所示。图 3-13 展示了定时任务实体的 E-R 结构。"
-            "定时任务实体描述用户配置的持续服务模板，"
-            "图中保留任务名称、调度表达式、参数配置、投递配置和运行状态等自身属性，"
-            "而用户归属关系仅通过实体联系表示，不再以外键字段形式放入属性图。"
-        ),
-        attrs=(
-            "任务ID(PK)",
-            "任务名称",
-            "描述",
-            "任务类型",
-            "调度表达式",
-            "时区",
-            "参数配置",
-            "投递配置",
-            "是否启用",
-            "最近执行时间",
-            "下次执行时间",
-            "执行次数",
-            "最近结果",
-            "最近错误",
-            "连续失败次数",
-            "创建时间",
-            "更新时间",
-        ),
-    ),
-    EntitySpec(
-        index=9,
-        figure_no="3-14",
-        entity_name="用户记忆",
-        caption="图3-14 用户记忆实体 E-R 图",
-        description=(
-            "（9）用户记忆实体 E-R 图如图3-14所示。图 3-14 展示了用户记忆实体的 E-R 结构。"
-            "用户记忆用于保存系统长期积累的偏好、事实与技能画像，"
-            "实体自身属性包括键值内容、置信度、访问统计、失效时间、向量嵌入和证据来源等，"
-            "用户归属关系在关系层处理，不再混入属性集合。"
-        ),
-        attrs=(
-            "记忆ID(PK)",
-            "记忆键",
-            "记忆值",
-            "置信度",
-            "记忆类型",
-            "记忆类别",
-            "访问次数",
-            "最近访问时间",
-            "失效时间",
-            "强化来源",
-            "更新时间",
-            "向量嵌入",
-            "来源类型",
-            "证据",
-            "冲突标记",
-        ),
-    ),
-    EntitySpec(
-        index=10,
-        figure_no="3-15",
-        entity_name="用户画像",
-        caption="图3-15 用户画像实体 E-R 图",
-        description=(
-            "（10）用户画像实体 E-R 图如图3-15所示。图 3-15 展示了用户画像实体的 E-R 结构。"
-            "用户画像实体是对长期记忆与交互行为进行阶段性聚合后的结果，"
-            "图中仅保留画像内容、头像地址、合成摘要、版本号以及时间相关属性，"
-            "一对一归属关系由实体联系表达而不作为属性列出现。"
-        ),
-        attrs=(
-            "画像JSON",
-            "头像地址",
-            "合成摘要",
-            "版本号",
-            "合成时间",
-            "创建时间",
-            "更新时间",
-        ),
-    ),
-    EntitySpec(
-        index=11,
-        figure_no="3-16",
-        entity_name="主动洞察",
-        caption="图3-16 主动洞察实体 E-R 图",
-        description=(
-            "（11）主动洞察实体 E-R 图如图3-16所示。图 3-16 展示了主动洞察实体的 E-R 结构。"
-            "主动洞察实体用于保存可以直接展示给用户的系统提示内容，"
-            "其自身属性包括洞察类型、标题、正文、元数据、已读状态与创建时间，"
-            "与用户和笔记本的关联关系则由实体之间的联系体现。"
-        ),
-        attrs=(
-            "洞察ID(PK)",
-            "洞察类型",
-            "标题",
-            "内容",
-            "元数据",
-            "已读状态",
-            "创建时间",
-        ),
-    ),
-    EntitySpec(
-        index=12,
-        figure_no="3-17",
-        entity_name="系统配置",
-        caption="图3-17 系统配置实体 E-R 图",
-        description=(
-            "（12）系统配置实体 E-R 图如图3-17所示。图 3-17 展示了系统配置实体的 E-R 结构。"
-            "系统配置实体采用键值形式保存安装向导和运行期写入的配置项，"
-            "由于该实体本身不承担业务归属关系，因此图中仅包含配置键与配置值两个自身属性。"
-        ),
-        attrs=(
-            "配置键(PK)",
-            "配置值",
-        ),
-    ),
-    EntitySpec(
-        index=13,
-        figure_no="3-18",
-        entity_name="笔记本摘要",
-        caption="图3-18 笔记本摘要实体 E-R 图",
-        description=(
-            "（13）笔记本摘要实体 E-R 图如图3-18所示。图 3-18 展示了笔记本摘要实体的 E-R 结构。"
-            "笔记本摘要实体用于保存系统自动生成的笔记本级聚合摘要，"
-            "图中仅展示摘要内容、主题词以及同步时间等实体自身属性，"
-            "其与笔记本的一对一绑定关系不再作为属性列重复绘制。"
-        ),
-        attrs=(
-            "摘要内容",
-            "核心主题",
-            "最后同步时间",
-        ),
-    ),
-    EntitySpec(
-        index=14,
-        figure_no="3-19",
-        entity_name="会话摘要",
-        caption="图3-19 会话摘要实体 E-R 图",
-        description=(
-            "（14）会话摘要实体 E-R 图如图3-19所示。图 3-19 展示了会话摘要实体的 E-R 结构。"
-            "会话摘要实体用于压缩长对话中的历史消息，"
-            "实体自身属性包含摘要正文、已压缩消息数量、压缩边界时间和更新时间等内容，"
-            "与对话实体的一对一归属关系通过关系设计表示。"
-        ),
-        attrs=(
-            "摘要正文",
-            "压缩消息数",
-            "压缩边界时间",
-            "更新时间",
-        ),
-    ),
-    EntitySpec(
-        index=15,
-        figure_no="3-20",
-        entity_name="定时任务运行记录",
-        caption="图3-20 定时任务运行记录实体 E-R 图",
-        description=(
-            "（15）定时任务运行记录实体 E-R 图如图3-20所示。图 3-20 展示了定时任务运行记录实体的 E-R 结构。"
-            "该实体用于记录定时任务每一次实际执行的过程和结果，"
-            "图中保留运行状态、开始与结束时间、执行耗时、结果摘要、错误信息以及投递状态等自身属性，"
-            "所属任务关系不再写入属性图。"
-        ),
-        attrs=(
-            "运行ID(PK)",
-            "状态",
-            "开始时间",
-            "结束时间",
-            "耗时毫秒",
-            "结果摘要",
-            "错误信息",
-            "生成内容",
-            "来源数量",
-            "投递状态",
-        ),
-    ),
-    EntitySpec(
-        index=16,
-        figure_no="3-21",
-        entity_name="后台思考",
-        caption="图3-21 后台思考实体 E-R 图",
-        description=(
-            "（16）后台思考实体 E-R 图如图3-21所示。图 3-21 展示了后台思考实体的 E-R 结构。"
-            "后台思考实体用于保存 Lyra Soul 在后台循环中形成的思考记录，"
-            "其自身属性包括可见性、思考内容、活动上下文和创建时间；"
-            "与用户和笔记本的关联关系通过实体联系表达，不再作为属性字段直接绘制。"
-        ),
-        attrs=(
-            "思考ID(PK)",
-            "可见性",
-            "思考内容",
-            "活动上下文",
-            "创建时间",
-        ),
-    ),
-)
-
-
-def ensure_dirs() -> None:
-    TMP_DIR.mkdir(parents=True, exist_ok=True)
-    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
-
-
-def pick_font(size: int, *, bold: bool = False) -> ImageFont.FreeTypeFont | ImageFont.ImageFont:
-    font_candidates = [
-        "/System/Library/Fonts/PingFang.ttc",
-        "/System/Library/Fonts/STHeiti Light.ttc",
-        "/System/Library/Fonts/STHeiti Medium.ttc",
-        "/System/Library/Fonts/Supplemental/Songti.ttc",
-        "/Library/Fonts/Arial Unicode.ttf",
-    ]
-    for candidate in font_candidates:
-        path = Path(candidate)
-        if path.exists():
-            try:
-                index = 1 if bold and "PingFang" in path.name else 0
-                return ImageFont.truetype(str(path), size=size, index=index)
-            except OSError:
-                continue
-    return ImageFont.load_default()
-
-
-def wrap_text(draw: ImageDraw.ImageDraw, text: str, font: ImageFont.ImageFont, max_width: int) -> str:
-    if draw.textbbox((0, 0), text, font=font)[2] <= max_width:
-        return text
-
-    lines: list[str] = []
-    current = ""
-    for ch in text:
-        candidate = current + ch
-        width = draw.textbbox((0, 0), candidate, font=font)[2]
-        if width <= max_width or not current:
-            current = candidate
-        else:
-            lines.append(current)
-            current = ch
-    if current:
-        lines.append(current)
-    return "\n".join(lines[:2])
-
-
-def chunk_counts(n: int) -> tuple[int, ...]:
-    if n <= 6:
-        return (n,)
-    if n <= 12:
-        first = (n + 1) // 2
-        return (first, n - first)
-    first = min(6, (n + 2) // 3)
-    second = min(6, (n - first + 1) // 2)
-    third = n - first - second
-    return tuple(count for count in (first, second, third) if count)
-
-
-def generate_diagram(spec: EntitySpec) -> Path:
-    canvas_w = 2200
-    canvas_h = 1300
-    image = Image.new("RGB", (canvas_w, canvas_h), "#ffffff")
-    draw = ImageDraw.Draw(image)
-
-    title_font = pick_font(54, bold=True)
-    box_font = pick_font(56, bold=True)
-    attr_font = pick_font(34)
-
-    box_w = 460
-    box_h = 120
-    counts = chunk_counts(len(spec.attrs))
-    row_ys = [110, 300, 490][: len(counts)]
-    box_top = 770 if len(counts) >= 3 else 710 if len(counts) == 2 else 650
-    box_left = (canvas_w - box_w) // 2
-    box_right = box_left + box_w
-    box_bottom = box_top + box_h
-
-    ellipse_w = 260
-    ellipse_h = 92
-    gap_x = 36
-
-    draw.text((90, 40), f"图 {spec.figure_no}  {spec.entity_name} 实体", fill="#333333", font=title_font)
-
-    draw.rounded_rectangle(
-        (box_left, box_top, box_right, box_bottom),
-        radius=18,
-        outline="#8a7ae6",
-        width=5,
-        fill="#f7f2ff",
-    )
-    box_text = spec.entity_name
-    text_bbox = draw.multiline_textbbox((0, 0), box_text, font=box_font, spacing=8, align="center")
-    box_text_w = text_bbox[2] - text_bbox[0]
-    box_text_h = text_bbox[3] - text_bbox[1]
-    draw.multiline_text(
-        ((canvas_w - box_text_w) / 2, box_top + (box_h - box_text_h) / 2 - 4),
-        box_text,
-        fill="#473a8b",
-        font=box_font,
-        align="center",
-        spacing=8,
-    )
-
-    attr_iter = iter(spec.attrs)
-    for row_count, row_y in zip(counts, row_ys):
-        total_w = row_count * ellipse_w + (row_count - 1) * gap_x
-        start_x = (canvas_w - total_w) // 2
-        for i in range(row_count):
-            attr = next(attr_iter)
-            left = start_x + i * (ellipse_w + gap_x)
-            top = row_y
-            right = left + ellipse_w
-            bottom = top + ellipse_h
-            center_x = (left + right) // 2
-            center_y = (top + bottom) // 2
-
-            draw.line(
-                [(center_x, bottom), (canvas_w // 2, box_top)],
-                fill="#c6b9f8",
-                width=3,
-            )
-            draw.ellipse(
-                (left, top, right, bottom),
-                outline="#9d8ff0",
-                width=4,
-                fill="#f3efff",
-            )
-
-            label = wrap_text(draw, attr, attr_font, ellipse_w - 24)
-            bbox = draw.multiline_textbbox((0, 0), label, font=attr_font, spacing=4, align="center")
-            text_w = bbox[2] - bbox[0]
-            text_h = bbox[3] - bbox[1]
-            draw.multiline_text(
-                (center_x - text_w / 2, center_y - text_h / 2 - 2),
-                label,
-                fill="#2f2f2f",
-                font=attr_font,
-                align="center",
-                spacing=4,
-            )
-
-    output = TMP_DIR / spec.image_name
-    image.save(output)
-    return output
-
-
-def insert_paragraph_after(paragraph: Paragraph, text: str = "", style: str | None = None) -> Paragraph:
-    new_p = OxmlElement("w:p")
-    paragraph._p.addnext(new_p)
-    new_para = Paragraph(new_p, paragraph._parent)
-    if style:
-        new_para.style = style
-    if text:
-        new_para.add_run(text)
-    return new_para
-
-
-def delete_paragraph(paragraph: Paragraph) -> None:
-    element = paragraph._element
-    parent = element.getparent()
-    if parent is not None:
-        parent.remove(element)
-
-
-def resolve_style(document: Document, preferred: str, fallback: str = "Normal") -> str:
-    styles = {style.name for style in document.styles}
-    if preferred in styles:
-        return preferred
-    return fallback if fallback in styles else "Normal"
-
-
-def rebuild_er_section(document: Document, image_paths: dict[str, Path]) -> None:
-    body_style = resolve_style(document, "毕业设计（论文）正文")
-    caption_style = resolve_style(document, "图标", "Normal")
-    image_style = resolve_style(document, "No Spacing", "Normal")
-
-    document.paragraphs[183].text = (
-        "经过对系统分析并结合当前代码中的数据模型定义，"
-        "本文将用户、笔记本、知识来源、知识分块、对话、消息、深度研究任务、定时任务、用户记忆、"
-        "用户画像、主动洞察、系统配置、笔记本摘要、会话摘要、定时任务运行记录和后台思考共 16 个核心实体"
-        "整理为属性型 E-R 图。为突出实体自身语义，图中只保留实体固有属性，"
-        "不再将外键字段纳入属性列表，实体之间的归属和关联关系通过后续数据库约束说明体现。"
-    )
-    document.paragraphs[183].style = body_style
-
-    document.paragraphs[189].text = (
-        "在实体-关系（E-R）模型设计中，本文主要围绕关系型数据库 PostgreSQL 的核心业务实体进行建模。"
-        "各实体的自有属性直接决定数据库表中的主要字段设计，而实体之间的关系则由主键、外键和关联约束进行表达。"
-        "因此，以下实体图仅展示主键与非外键业务属性，"
-        "不再把归属关系字段重复放入属性图中，以避免实体属性与关系属性混淆。"
-    )
-    document.paragraphs[189].style = body_style
-
-    for paragraph in list(document.paragraphs[190:222]):
-        delete_paragraph(paragraph)
-
-    anchor = document.paragraphs[189]
-    for spec in ENTITY_SPECS:
-        desc_para = insert_paragraph_after(anchor, spec.description, body_style)
-        anchor = desc_para
-
-        image_para = insert_paragraph_after(anchor, style=image_style)
-        image_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
-        image_para.add_run().add_picture(str(image_paths[spec.figure_no]), width=Inches(6.8))
-        anchor = image_para
-
-        caption_para = insert_paragraph_after(anchor, spec.caption, caption_style)
-        caption_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
-        anchor = caption_para
-
-
-def validate_output(document_path: Path) -> None:
-    document = Document(document_path)
-    expected_captions = {spec.caption for spec in ENTITY_SPECS}
-    actual_captions = {p.text.strip() for p in document.paragraphs if p.text.strip().startswith("图3-")}
-    missing = expected_captions - actual_captions
-    if missing:
-        raise RuntimeError(f"输出文档缺少图题: {sorted(missing)}")
-
-
-def main() -> None:
-    ensure_dirs()
-    if not SOURCE_DOC.exists():
-        raise FileNotFoundError(f"未找到输入文档: {SOURCE_DOC}")
-
-    image_paths = {spec.figure_no: generate_diagram(spec) for spec in ENTITY_SPECS}
-
-    document = Document(SOURCE_DOC)
-    rebuild_er_section(document, image_paths)
-    document.save(OUTPUT_DOC)
-    validate_output(OUTPUT_DOC)
-
-    print(OUTPUT_DOC)
-
-
-if __name__ == "__main__":
-    main()
diff --git a/scripts/fix_thesis_er_diagrams_mermaid.py b/scripts/fix_thesis_er_diagrams_mermaid.py
deleted file mode 100644
index 617ae79..0000000
--- a/scripts/fix_thesis_er_diagrams_mermaid.py
+++ /dev/null
@@ -1,303 +0,0 @@
-from __future__ import annotations
-
-from pathlib import Path
-
-from docx import Document
-from docx.enum.text import WD_ALIGN_PARAGRAPH
-from docx.shared import Inches
-from PIL import Image, ImageDraw, ImageFont
-
-from fix_thesis_er_diagrams import (
-    ENTITY_SPECS,
-    OUTPUT_DIR,
-    REPO_ROOT,
-    delete_paragraph,
-    ensure_dirs,
-    insert_paragraph_after,
-    resolve_style,
-)
-
-
-SOURCE_DOC = OUTPUT_DIR / "220501020064-黄凯-毕业论文-引用修订版_4-ER图修订版.docx"
-TMP_DIR = REPO_ROOT / "tmp" / "docs" / "thesis-er-diagram-mermaid"
-OUTPUT_DOC = OUTPUT_DIR / "220501020064-黄凯-毕业论文-引用修订版_4-ER图Mermaid版.docx"
-
-BG_COLOR = "#ffffff"
-BOX_FILL = "#fff4dd"
-BOX_BORDER = "#8b6f47"
-HEADER_TEXT = "#3b342c"
-ROW_ODD = "#ffffff"
-ROW_EVEN = "#f2f2f2"
-GRID = "#b9a88a"
-TEXT = "#333333"
-TYPE_TEXT = "#59534e"
-KEY_FILL = "#ece8ff"
-KEY_TEXT = "#51459b"
-
-
-def pick_font(size: int, *, bold: bool = False) -> ImageFont.FreeTypeFont | ImageFont.ImageFont:
-    font_candidates = [
-        ("/System/Library/Fonts/PingFang.ttc", 1 if bold else 0),
-        ("/System/Library/Fonts/STHeiti Medium.ttc", 0),
-        ("/System/Library/Fonts/Supplemental/Songti.ttc", 0),
-    ]
-    for candidate, index in font_candidates:
-        path = Path(candidate)
-        if path.exists():
-            try:
-                return ImageFont.truetype(str(path), size=size, index=index)
-            except OSError:
-                continue
-    return ImageFont.load_default()
-
-
-def measure(draw: ImageDraw.ImageDraw, text: str, font: ImageFont.ImageFont) -> tuple[int, int]:
-    bbox = draw.textbbox((0, 0), text, font=font)
-    return bbox[2] - bbox[0], bbox[3] - bbox[1]
-
-
-def parse_attr(attr: str) -> tuple[str, str, str]:
-    key = ""
-    name = attr
-    if attr.endswith("(PK)"):
-        name = attr[:-4]
-        key = "PK"
-    elif attr.endswith("(UK)"):
-        name = attr[:-4]
-        key = "UK"
-
-    attr_type = infer_type(name)
-    return attr_type, name, key
-
-
-def infer_type(name: str) -> str:
-    if name.endswith("ID"):
-        return "uuid"
-    if name.startswith("是否") or name in {"已读状态", "冲突标记"}:
-        return "bool"
-    if "时间" in name or "日期" in name:
-        return "datetime"
-    if "数量" in name or "次数" in name or "序号" in name or "版本号" in name or "耗时" in name:
-        return "int"
-    if "置信度" in name:
-        return "float"
-    if "嵌入" in name:
-        return "vector"
-    if (
-        "JSON" in name
-        or "元数据" in name
-        or "列表" in name
-        or "上下文" in name
-        or name in {
-            "参数配置",
-            "投递配置",
-            "核心主题",
-            "活动上下文",
-            "引用信息",
-            "Agent步骤",
-            "附件信息",
-            "速度指标",
-            "思维导图",
-            "图表数据",
-            "MCP结果",
-            "UI元素",
-            "画像JSON",
-            "交付物JSON",
-            "时间线JSON",
-            "投递状态",
-        }
-    ):
-        return "json"
-    if (
-        "内容" in name
-        or "正文" in name
-        or "摘要" in name
-        or "描述" in name
-        or "报告" in name
-        or "错误" in name
-        or "证据" in name
-        or "轨迹" in name
-        or "结果" in name
-        or "文本" in name
-    ):
-        return "text"
-    return "string"
-
-
-def make_mermaid_source(entity_name: str, attrs: tuple[str, ...]) -> str:
-    rows = []
-    for attr in attrs:
-        attr_type, name, key = parse_attr(attr)
-        if key:
-            rows.append(f"        {attr_type} {name} {key}")
-        else:
-            rows.append(f"        {attr_type} {name}")
-    return "erDiagram\n" + f"    {entity_name} {{\n" + "\n".join(rows) + "\n    }\n"
-
-
-def render_mermaid_style_entity(entity_name: str, attrs: tuple[str, ...]) -> tuple[Path, tuple[int, int]]:
-    TMP_DIR.mkdir(parents=True, exist_ok=True)
-
-    rows = [parse_attr(attr) for attr in attrs]
-    mmd_path = TMP_DIR / f"{entity_name}.mmd"
-    mmd_path.write_text(make_mermaid_source(entity_name, attrs), encoding="utf-8")
-
-    title_font = pick_font(28, bold=True)
-    row_font = pick_font(24)
-    key_font = pick_font(18, bold=True)
-
-    probe = Image.new("RGB", (10, 10), BG_COLOR)
-    draw = ImageDraw.Draw(probe)
-
-    type_width = max(measure(draw, r[0], row_font)[0] for r in rows)
-    name_width = max(measure(draw, r[1], row_font)[0] for r in rows)
-    header_width = measure(draw, entity_name, title_font)[0]
-    has_key = any(r[2] for r in rows)
-
-    cell_pad_x = 18
-    header_h = 52
-    row_h = 40
-    border = 3
-    key_w = 52 if has_key else 0
-    type_col_w = type_width + cell_pad_x * 2
-    name_col_w = max(name_width + cell_pad_x * 2, header_width + 48)
-    table_w = type_col_w + name_col_w + key_w
-    table_h = header_h + len(rows) * row_h
-    image_w = table_w + border * 2
-    image_h = table_h + border * 2
-
-    image = Image.new("RGB", (image_w, image_h), BG_COLOR)
-    draw = ImageDraw.Draw(image)
-
-    left = border
-    top = border
-    right = left + table_w
-    bottom = top + table_h
-    draw.rounded_rectangle(
-        (left, top, right, bottom),
-        radius=10,
-        fill=BOX_FILL,
-        outline=BOX_BORDER,
-        width=border,
-    )
-    draw.rectangle((left, top, right, top + header_h), fill=BOX_FILL, outline=None)
-    draw.line((left, top + header_h, right, top + header_h), fill=BOX_BORDER, width=2)
-
-    header_text_w, header_text_h = measure(draw, entity_name, title_font)
-    draw.text(
-        (left + (table_w - header_text_w) / 2, top + (header_h - header_text_h) / 2 - 2),
-        entity_name,
-        fill=HEADER_TEXT,
-        font=title_font,
-    )
-
-    for idx, (attr_type, name, key) in enumerate(rows):
-        row_top = top + header_h + idx * row_h
-        row_bottom = row_top + row_h
-        draw.rectangle(
-            (left, row_top, right, row_bottom),
-            fill=ROW_EVEN if idx % 2 else ROW_ODD,
-            outline=None,
-        )
-        if idx < len(rows) - 1:
-            draw.line((left, row_bottom, right, row_bottom), fill=GRID, width=1)
-
-        draw.text((left + cell_pad_x, row_top + 7), attr_type, fill=TYPE_TEXT, font=row_font)
-        draw.text((left + type_col_w + cell_pad_x, row_top + 7), name, fill=TEXT, font=row_font)
-
-        draw.line((left + type_col_w, row_top, left + type_col_w, row_bottom), fill=GRID, width=1)
-        if has_key:
-            draw.line((right - key_w, row_top, right - key_w, row_bottom), fill=GRID, width=1)
-            if key:
-                pill_w = 34
-                pill_h = 22
-                pill_x = right - key_w + (key_w - pill_w) / 2
-                pill_y = row_top + (row_h - pill_h) / 2
-                draw.rounded_rectangle(
-                    (pill_x, pill_y, pill_x + pill_w, pill_y + pill_h),
-                    radius=10,
-                    fill=KEY_FILL,
-                    outline=None,
-                )
-                key_text_w, key_text_h = measure(draw, key, key_font)
-                draw.text(
-                    (pill_x + (pill_w - key_text_w) / 2, pill_y + (pill_h - key_text_h) / 2 - 1),
-                    key,
-                    fill=KEY_TEXT,
-                    font=key_font,
-                )
-
-    png_path = TMP_DIR / f"{entity_name}.png"
-    image.save(png_path)
-    return png_path, (image_w, image_h)
-
-
-def rebuild_er_section(document: Document, image_meta: dict[str, tuple[Path, tuple[int, int]]]) -> None:
-    body_style = resolve_style(document, "毕业设计（论文）正文")
-    caption_style = resolve_style(document, "图标", "Normal")
-    image_style = resolve_style(document, "No Spacing", "Normal")
-
-    document.paragraphs[183].text = (
-        "经过对系统分析并结合当前代码中的数据模型定义，本文将用户、笔记本、知识来源、知识分块、对话、消息、"
-        "深度研究任务、定时任务、用户记忆、用户画像、主动洞察、系统配置、笔记本摘要、会话摘要、定时任务运行记录和后台思考"
-        "共 16 个核心实体整理为属性型 E-R 图。为突出实体自身语义，图中只保留实体固有属性，不再将外键字段纳入属性列表；"
-        "本次图形表现采用 Mermaid 风格的实体卡片形式，以便在版面上更紧凑、更统一地呈现各实体结构。"
-    )
-    document.paragraphs[183].style = body_style
-
-    document.paragraphs[189].text = (
-        "在实体-关系（E-R）模型设计中，本文主要围绕关系型数据库 PostgreSQL 的核心业务实体进行建模。"
-        "各实体的自有属性直接决定数据库表中的主要字段设计，而实体之间的关系则由主键、外键和关联约束进行表达。"
-        "因此，以下实体图仅展示主键与非外键业务属性，不再把归属关系字段重复放入属性图中；"
-        "同时采用 Mermaid 风格的表格式实体表示，以减少图片冗余留白并提升可读性。"
-    )
-    document.paragraphs[189].style = body_style
-
-    for paragraph in list(document.paragraphs[190:238]):
-        delete_paragraph(paragraph)
-
-    anchor = document.paragraphs[189]
-    for spec in ENTITY_SPECS:
-        desc_para = insert_paragraph_after(anchor, spec.description, body_style)
-        anchor = desc_para
-
-        image_para = insert_paragraph_after(anchor, style=image_style)
-        image_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
-        image_path, (px_w, _px_h) = image_meta[spec.figure_no]
-        width_inches = min(5.8, max(2.6, px_w / 230))
-        image_para.add_run().add_picture(str(image_path), width=Inches(width_inches))
-        anchor = image_para
-
-        caption_para = insert_paragraph_after(anchor, spec.caption, caption_style)
-        caption_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
-        anchor = caption_para
-
-
-def validate_output(document_path: Path) -> None:
-    document = Document(document_path)
-    expected = {spec.caption for spec in ENTITY_SPECS}
-    actual = {p.text.strip() for p in document.paragraphs if p.text.strip().startswith("图3-")}
-    missing = expected - actual
-    if missing:
-        raise RuntimeError(f"输出文档缺少图题: {sorted(missing)}")
-
-
-def main() -> None:
-    ensure_dirs()
-    TMP_DIR.mkdir(parents=True, exist_ok=True)
-    if not SOURCE_DOC.exists():
-        raise FileNotFoundError(f"未找到输入文档: {SOURCE_DOC}")
-
-    image_meta = {}
-    for spec in ENTITY_SPECS:
-        image_meta[spec.figure_no] = render_mermaid_style_entity(spec.entity_name, spec.attrs)
-
-    document = Document(SOURCE_DOC)
-    rebuild_er_section(document, image_meta)
-    document.save(OUTPUT_DOC)
-    validate_output(OUTPUT_DOC)
-    print(OUTPUT_DOC)
-
-
-if __name__ == "__main__":
-    main()
diff --git a/scripts/generate_thesis_pdfs.py b/scripts/generate_thesis_pdfs.py
deleted file mode 100644
index 9caa641..0000000
--- a/scripts/generate_thesis_pdfs.py
+++ /dev/null
@@ -1,297 +0,0 @@
-from __future__ import annotations
-
-import html
-import re
-from pathlib import Path
-
-from reportlab.lib import colors
-from reportlab.lib.enums import TA_CENTER
-from reportlab.lib.pagesizes import A4
-from reportlab.lib.styles import ParagraphStyle, getSampleStyleSheet
-from reportlab.lib.units import mm
-from reportlab.pdfbase.cidfonts import UnicodeCIDFont
-from reportlab.pdfbase.pdfmetrics import registerFont
-from reportlab.platypus import (
-    HRFlowable,
-    ListFlowable,
-    ListItem,
-    PageBreak,
-    Paragraph,
-    SimpleDocTemplate,
-    Spacer,
-)
-
-
-ROOT = Path(__file__).resolve().parent.parent
-SOURCE_DIR = ROOT / "docs" / "graduation-thesis"
-OUTPUT_DIR = ROOT / "output" / "pdf"
-
-DOCUMENTS = [
-    {
-        "source": SOURCE_DIR / "lyranote-undergrad-thesis-outline.md",
-        "output": OUTPUT_DIR / "lyranote-undergrad-thesis-outline.pdf",
-        "title": "LyraNote 本科毕业论文大纲",
-    },
-    {
-        "source": SOURCE_DIR / "lyranote-undergrad-thesis-demo.md",
-        "output": OUTPUT_DIR / "lyranote-undergrad-thesis-demo.pdf",
-        "title": "LyraNote 本科毕业论文 Demo 稿",
-    },
-]
-
-
-def register_fonts() -> None:
-    registerFont(UnicodeCIDFont("STSong-Light"))
-
-
-def build_styles():
-    styles = getSampleStyleSheet()
-    body = ParagraphStyle(
-        "BodyCN",
-        parent=styles["BodyText"],
-        fontName="STSong-Light",
-        fontSize=11,
-        leading=18,
-        textColor=colors.HexColor("#1f2937"),
-        spaceAfter=6,
-        wordWrap="CJK",
-    )
-    title = ParagraphStyle(
-        "TitleCN",
-        parent=styles["Title"],
-        fontName="STSong-Light",
-        fontSize=22,
-        leading=30,
-        alignment=TA_CENTER,
-        textColor=colors.HexColor("#111827"),
-        spaceAfter=14,
-    )
-    h1 = ParagraphStyle(
-        "Heading1CN",
-        parent=styles["Heading1"],
-        fontName="STSong-Light",
-        fontSize=18,
-        leading=26,
-        textColor=colors.HexColor("#111827"),
-        spaceBefore=10,
-        spaceAfter=10,
-        wordWrap="CJK",
-    )
-    h2 = ParagraphStyle(
-        "Heading2CN",
-        parent=styles["Heading2"],
-        fontName="STSong-Light",
-        fontSize=14,
-        leading=22,
-        textColor=colors.HexColor("#111827"),
-        spaceBefore=10,
-        spaceAfter=8,
-        wordWrap="CJK",
-    )
-    h3 = ParagraphStyle(
-        "Heading3CN",
-        parent=styles["Heading3"],
-        fontName="STSong-Light",
-        fontSize=12,
-        leading=18,
-        textColor=colors.HexColor("#111827"),
-        spaceBefore=8,
-        spaceAfter=6,
-        wordWrap="CJK",
-    )
-    meta = ParagraphStyle(
-        "MetaCN",
-        parent=body,
-        fontSize=10.5,
-        leading=17,
-        alignment=TA_CENTER,
-        textColor=colors.HexColor("#4b5563"),
-        spaceAfter=4,
-    )
-    list_style = ParagraphStyle(
-        "ListCN",
-        parent=body,
-        leftIndent=0,
-        firstLineIndent=0,
-        spaceAfter=2,
-    )
-    return {
-        "body": body,
-        "title": title,
-        "h1": h1,
-        "h2": h2,
-        "h3": h3,
-        "meta": meta,
-        "list": list_style,
-    }
-
-
-def escape_text(text: str) -> str:
-    return html.escape(text).replace("\n", "<br/>")
-
-
-def add_page_number(canvas, doc) -> None:
-    canvas.setFont("STSong-Light", 9)
-    canvas.setFillColor(colors.HexColor("#6b7280"))
-    canvas.drawCentredString(A4[0] / 2, 10 * mm, f"{doc.page}")
-
-
-def flush_paragraph(story, buffer: list[str], style) -> None:
-    if not buffer:
-        return
-    text = " ".join(line.strip() for line in buffer).strip()
-    if text:
-        story.append(Paragraph(escape_text(text), style))
-    buffer.clear()
-
-
-def flush_list(story, items: list[str], list_type: str, styles) -> None:
-    if not items:
-        return
-    bullet_type = "bullet" if list_type == "bullet" else "1"
-    list_items = [
-        ListItem(Paragraph(escape_text(item), styles["list"]))
-        for item in items
-    ]
-    story.append(
-        ListFlowable(
-            list_items,
-            bulletType=bullet_type,
-            start="1",
-            leftIndent=18,
-            bulletFontName="STSong-Light",
-            bulletFontSize=10.5,
-            spaceAfter=6,
-        )
-    )
-    items.clear()
-
-
-def markdown_to_story(text: str, doc_title: str, styles) -> list:
-    story: list = []
-    story.append(Spacer(1, 22 * mm))
-    story.append(Paragraph(escape_text(doc_title), styles["title"]))
-    story.append(Spacer(1, 4 * mm))
-
-    paragraph_buffer: list[str] = []
-    list_buffer: list[str] = []
-    list_type: str | None = None
-
-    for raw_line in text.splitlines():
-        line = raw_line.rstrip()
-        stripped = line.strip()
-
-        if stripped == "<<<PAGEBREAK>>>":
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type:
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            story.append(PageBreak())
-            continue
-
-        if stripped == "":
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type:
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            continue
-
-        if stripped == "---":
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type:
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            story.append(Spacer(1, 2 * mm))
-            story.append(HRFlowable(width="100%", thickness=0.5, color=colors.HexColor("#d1d5db")))
-            story.append(Spacer(1, 3 * mm))
-            continue
-
-        if stripped.startswith("# "):
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type:
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            story.append(Paragraph(escape_text(stripped[2:].strip()), styles["h1"]))
-            continue
-
-        if stripped.startswith("## "):
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type:
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            story.append(Paragraph(escape_text(stripped[3:].strip()), styles["h2"]))
-            continue
-
-        if stripped.startswith("### "):
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type:
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            story.append(Paragraph(escape_text(stripped[4:].strip()), styles["h3"]))
-            continue
-
-        bullet_match = re.match(r"^[-*]\s+(.*)$", stripped)
-        if bullet_match:
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type and list_type != "bullet":
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            list_type = "bullet"
-            list_buffer.append(bullet_match.group(1).strip())
-            continue
-
-        ordered_match = re.match(r"^\d+\.\s+(.*)$", stripped)
-        if ordered_match:
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type and list_type != "ordered":
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            list_type = "ordered"
-            list_buffer.append(ordered_match.group(1).strip())
-            continue
-
-        if not story or len(story) <= 4:
-            flush_paragraph(story, paragraph_buffer, styles["body"])
-            if list_type:
-                flush_list(story, list_buffer, list_type, styles)
-                list_type = None
-            story.append(Paragraph(escape_text(stripped), styles["meta"]))
-            continue
-
-        paragraph_buffer.append(stripped)
-
-    flush_paragraph(story, paragraph_buffer, styles["body"])
-    if list_type:
-        flush_list(story, list_buffer, list_type, styles)
-
-    return story
-
-
-def build_pdf(source_path: Path, output_path: Path, doc_title: str, styles) -> None:
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    text = source_path.read_text(encoding="utf-8")
-    story = markdown_to_story(text, doc_title, styles)
-
-    doc = SimpleDocTemplate(
-        str(output_path),
-        pagesize=A4,
-        leftMargin=20 * mm,
-        rightMargin=20 * mm,
-        topMargin=18 * mm,
-        bottomMargin=18 * mm,
-        title=doc_title,
-        author="Codex for LyraNote",
-    )
-    doc.build(story, onFirstPage=add_page_number, onLaterPages=add_page_number)
-
-
-def main() -> None:
-    register_fonts()
-    styles = build_styles()
-    for item in DOCUMENTS:
-        build_pdf(item["source"], item["output"], item["title"], styles)
-        print(f"generated: {item['output'].relative_to(ROOT)}")
-
-
-if __name__ == "__main__":
-    main()