docs: replace clipped architecture charts with clear SVG diagrams

README.en.md

@@ -73,17 +73,10 @@ Start the Dashboard with `node <brain-path>/.loci/dashboard/server.js`, then ope
 
 ## One entry, one manual, shared data
 
-```mermaid
-flowchart TD
-  A[LOCI.md: short entry] --> B[Native instructions in each agent]
-  B --> C[Startup: short preferences]
-  H[Optional hook / same fallback reader] --> C
-  B --> D[First memory use: read LOCI-RULES.md once]
-  D --> E[Read relevant indexes and records]
-  E --> F[Existing scripts / Dashboard API]
-  F --> G[Local Markdown, JSON and attachments]
-  UI[Dashboard] <--> G
-```
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="docs/assets/architecture-en-dark.svg" />
+  <img src="docs/assets/architecture-en-light.svg" alt="Loci architecture: install an entry, load preferences at startup, read rules at first memory use, and share local data" width="960" />
+</picture>
 
 - **`LOCI.md`** provides the brain path, startup preference reader and triggers to read the manual.
 - **`LOCI-RULES.md`** contains the complete operating rules. Read it at first Loci use, reuse while valid, reload after changes or lost context.

README.md

@@ -71,17 +71,10 @@ node loci/scripts/loci-install.js --connect auto --lang zh
 
 ## 架构：一个入口，一本手册，一套数据
 
-```mermaid
-flowchart TD
-  A[LOCI.md：短入口] --> B[安装到各 Agent 原生指令文件]
-  B --> C[启动：短偏好与大脑位置]
-  H[可选 Hook / 同一个后备读取器] --> C
-  B --> D[首次涉及记忆：完整读取 LOCI-RULES.md]
-  D --> E[按需查询索引和相关记录]
-  E --> F[现有脚本 / Dashboard API]
-  F --> G[本地 Markdown、JSON、附件]
-  UI[Dashboard] <--> G
-```
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="docs/assets/architecture-zh-dark.svg" />
+  <img src="docs/assets/architecture-zh-light.svg" alt="Loci 架构：安装短入口，启动获取偏好，首次使用读手册，Agent 与 Dashboard 共用本地数据" width="960" />
+</picture>
 
 - **`LOCI.md` 是短入口**：告诉 Agent 大脑在哪、如何获取偏好、何时读取完整手册。
 - **`LOCI-RULES.md` 是完整手册**：任务、碎片、人物、项目等规则集中分块维护；首次涉及 Loci 操作时读一次，未变且仍在上下文则复用。

docs/architecture.md

@@ -4,23 +4,12 @@
 
 Loci 的核心是让不同 Agent 通过自己的原生指令入口，使用同一份本地记忆。规则告诉模型何时读取和调用工具，脚本/API 完成实际读写。Markdown 指令能够引导模型，不能从机制上强制它执行；安装验证与真实客户端验收是两件事。
 
-```mermaid
-flowchart TD
-    Source[LOCI.md：唯一短入口源文件] --> Install[同一个 Node 安装器：替换大脑路径并合并标记区]
-    Install --> Native[各 Agent 原生指令入口]
-    Native --> Prefs[启动：短偏好与大脑位置]
-    Hook[可选 SessionStart Hook] --> Context[scripts/loci-context.js]
-    Fallback[入口指令的后备调用] --> Context
-    Context --> Prefs
-    Native --> Trigger{首次涉及 Loci 记忆操作？}
-    Trigger -->|是| Rules[完整读取 LOCI-RULES.md 一次]
-    Trigger -->|否| Chat[继续普通请求]
-    Rules --> Index[按需查索引、读取最小相关数据]
-    Index --> Tools[现有脚本或 Dashboard API]
-    Tools --> Data[本地 Markdown、JSON 和附件]
-    Dashboard[Dashboard 界面] <--> Data
-    Data --> Project[项目正文留在对应仓库，大脑只保存索引]
-```
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="assets/architecture-zh-dark.svg" />
+  <img src="assets/architecture-zh-light.svg" alt="Loci 架构：安装短入口，启动获取偏好，首次使用读手册，Agent 与 Dashboard 共用本地数据" width="960" />
+</picture>
+
+图分为三个阶段：安装入口、启动获取短偏好、首次使用记忆。Dashboard 与 Agent 最终读写同一套本地数据。图片随 GitHub 的深浅主题切换；文字与流程源文件在 [`scripts/render-architecture.js`](../scripts/render-architecture.js)，运行 `node scripts/render-architecture.js` 可重新生成中英文 SVG。
 
 ## 哪些文件需要维护