dsh-memory-nebula 1.1.1 → 1.1.2

package/README.en.md

@@ -49,7 +49,7 @@ Export the whole graph (text included) as one self-contained HTML for offline re
 ## Quick start
 
 From the plugin market (recommended): add the index
-`https://cdn.jsdelivr.net/npm/dsh-memory-nebula@1.1.1/market/index.json` under
+`https://unpkg.com/dsh-memory-nebula@1.1.2/market/index.json` under
 Plugins → Discover, search for “Memory Nebula”, and install. Run the doctor in step 2.5 afterwards.
 
 Or install from this checkout:
@@ -57,7 +57,7 @@ Or install from this checkout:
 ```powershell
 # 1) link the plugin (either way)
 New-Item -ItemType Junction -Path "<profile>\node_modules\dsh-memory-nebula" -Target "<plugin>"
-npm install <path>\dsh-memory-nebula-1.1.1.tgz --prefix <profile>
+npm install <path>\dsh-memory-nebula-1.1.2.tgz --prefix <profile>
 
 # 2) profile package.json: add the dependency AND the bundle
 #    dependencies:        "dsh-memory-nebula": "file:../../plugins/dsh-memory-nebula"
@@ -148,4 +148,4 @@ test/                 node:test unit suite (84 assertions)
 
 ## Version
 
-`1.1.1`, MIT. Front-end changes apply on refresh; everything under `lib/host/` is hot-reloaded, while `http.js` / `index.js` / `live.js` need a DSH restart (or the standalone engine).
+`1.1.2`, MIT. Front-end changes apply on refresh; everything under `lib/host/` is hot-reloaded, while `http.js` / `index.js` / `live.js` need a DSH restart (or the standalone engine).

package/README.md

@@ -73,7 +73,7 @@
 已经装好的机器上（`profiles/desktop`），只要打开「记忆星图」页面就有一张图；页面每 12 秒比一次 `sig`，别的会话写了新记忆会自动长出新星点。库里还没有记忆时，页面会给三条接入路径，也可以先点「看示例」；`…/memory-nebula/?demo=1` 可直达示例星系，逛完退出不会动真实库。
 
 从插件市场装（推荐）：在 DSH 的「插件 → 发现」里添加索引
-`https://cdn.jsdelivr.net/npm/dsh-memory-nebula@1.1.1/market/index.json`，搜索「记忆星图」安装。
+`https://unpkg.com/dsh-memory-nebula@1.1.2/market/index.json`，搜索「记忆星图」安装。
 装完按下面的 2.5 跑一次 doctor。
 
 也可以从仓库装到 DSH：
@@ -81,7 +81,7 @@
 ```powershell
 # 1) 装进 profile（二选一）
 New-Item -ItemType Junction -Path "<profile>\node_modules\dsh-memory-nebula" -Target "<plugin>"
-npm install <path>\dsh-memory-nebula-1.1.1.tgz --prefix <profile>
+npm install <path>\dsh-memory-nebula-1.1.2.tgz --prefix <profile>
 
 # 2) profile 的 package.json 里加依赖与 bundles（只加 dependencies 不会重组）
 #    dependencies:         "dsh-memory-nebula": "file:../../plugins/dsh-memory-nebula"
@@ -181,6 +181,6 @@ sidecar.ps1           Windows 独立引擎开关
 
 ## 版本与许可
 
-- 当前版本 `1.1.1`，MIT。
+- 当前版本 `1.1.2`，MIT。
 - 打包内容：`lib/`、`scripts/`、`docs/`、`cordis.patch.yml`、`README.md` / `README.en.md`、`CHANGELOG.md`、`LICENSE`、`sidecar.ps1`（`test/` 与 `dev/` 不进包）。
 - 前端改动刷新页面即生效；`lib/host/` 下所有引擎文件（含 `adapters/`、`bridge.js`、`ledger.js`）由 LiveEngine 热载，`http.js` / `index.js` / `live.js` 改动要重启 DSH 或走独立引擎。

package/docs/compatibility.md

@@ -1,6 +1,6 @@
 # 兼容矩阵
 
-最后更新：2026-11-10（1.1.1 增加市场索引；运行前提未变）
+最后更新：2026-11-10（1.1.2 市场索引改由 unpkg 托管；运行前提未变）
 
 这张表只写**真的跑过**的东西。「预期可用」是依据运行前提做的判断，不是实测；
 没有出现的组合就是没有验证，不应当作支持承诺。
@@ -22,7 +22,7 @@
 | 适配器 | meow / notes / sqlite / ingest | 单测 + `sources`/`ingest` 冒烟 |
 | 独立引擎 | `lib/standalone.mjs`，127.0.0.1:7890/7891 | 0.5–0.9 的 `smoke` 与 `--diagnose` |
 | 安装体检 | 真实 desktop profile + 运行中实例 | 1.1.0：9 项全过（含 endpoint ping） |
-| 市场索引 | `@deepseek-ai/dsh-plugin-market-index` format 1 校验 | 1.1.1：`parseMarketIndex` 通过，图标与安装参数可解析 |
+| 市场索引 | `@deepseek-ai/dsh-plugin-market-index` format 1 校验 | 1.1.2：unpkg 索引 200，`parseMarketIndex` 通过，图标/详情/图片可解析 |
 
 ## 预期可用，但未实测
 

package/docs/distribution.md

@@ -19,7 +19,7 @@
 在「插件 → 发现」里添加索引：
 
 ```text
-https://cdn.jsdelivr.net/npm/dsh-memory-nebula@1.1.1/market/index.json
+https://unpkg.com/dsh-memory-nebula@1.1.2/market/index.json
 ```
 
 搜索「记忆星图」安装。索引与详情页都在 npm 包里，由 jsDelivr 提供；不需要第二个站点。
@@ -29,16 +29,16 @@ https://cdn.jsdelivr.net/npm/dsh-memory-nebula@1.1.1/market/index.json
 
 ```powershell
 # Windows PowerShell
-Get-FileHash .\dsh-memory-nebula-1.1.1.tgz -Algorithm SHA256
+Get-FileHash .\dsh-memory-nebula-1.1.2.tgz -Algorithm SHA256
 # 对照同目录的 .sha256 文件；不一致就停，不要装
 
-npm install .\dsh-memory-nebula-1.1.1.tgz --prefix <profile>
+npm install .\dsh-memory-nebula-1.1.2.tgz --prefix <profile>
 ```
 
 ```bash
 # macOS / Linux
-sha256sum -c dsh-memory-nebula-1.1.1.tgz.sha256
-npm install ./dsh-memory-nebula-1.1.1.tgz --prefix <profile>
+sha256sum -c dsh-memory-nebula-1.1.2.tgz.sha256
+npm install ./dsh-memory-nebula-1.1.2.tgz --prefix <profile>
 ```
 
 然后确认 `<profile>/package.json` 里两处都在：
@@ -129,7 +129,7 @@ Remove-Item -Recurse -Force "$env:USERPROFILE\.dsh\memory-nebula"
 npm run verify          # 单测 + release-check；prepack 也会跑
 node scripts/smoke.mjs http://127.0.0.1:7891/api/memory-nebula
 npm pack                # 生成 tgz
-sha256sum dsh-memory-nebula-1.1.1.tgz > dsh-memory-nebula-1.1.1.tgz.sha256
+sha256sum dsh-memory-nebula-1.1.2.tgz > dsh-memory-nebula-1.1.2.tgz.sha256
 npm publish             # 需要维护者账号；本仓库不自动发布
 ```
 

package/docs/http-api.md

@@ -17,7 +17,7 @@
 ```json
 {
   "ok": true,
-  "version": "1.1.1",
+  "version": "1.1.2",
   "build": 1794143336695,
   "startedAt": 1794143336695,
   "mode": "host",
@@ -47,7 +47,7 @@
 ```text
 memory-nebula diagnostics
 # 不含记忆正文；可以整段贴给支持
-version: 1.1.1
+version: 1.1.2
 build: 1794276294467
 started: 2026-11-10T02:20:37.399Z
 mode: standalone

package/docs/market.en.md

@@ -1,109 +1,109 @@
-# Memory Nebula
-
-Turn memories scattered across workspaces into a star map you can walk, query and carry away.
-
-Memories in DSH often live across dozens of workspaces and thousands of sessions. Search returns a list, but not **why** those memories relate; switch projects and the thread disappears. Memory Nebula reads the same stores (meow-memory / OpenViking) and builds a graph: nodes are memories, edges are shared keywords, entities, projects, session co-activation and manual curation. It never copies or rewrites your stores. Everything is computed locally.
-
-![Memory Nebula: a thousand memories and their links](assets/preview.png)
-
-## Three ways to look
-
-- **Nebula**: every memory in one cloud, clustered by workspace and coloured by level and link strength. Good for “what does my memory look like as a whole?”
-- **Galaxy**: grouped by workspace; open a cluster to enter that project. Good for “what have I thought about here?”
-- **Time**: memories arranged as a river by creation/update time, with a scrubber. Good for “how did this grow?”
-
-All three share the same filters: workspace, level, edge type, minimum strength and importance. Switching views keeps the selection.
-
-## From one memory to another
-
-Select a node and the right panel shows what it connects to and why: same project, shared keywords, same entity, repeatedly co-mentioned in sessions, or a link you made yourself. Every neighbour carries a weight and a reason, not an evidence-free “similarity 0.87”.
-
-Shift-click a second node to find an **evidence chain**: every hop, its reason and the chain strength. If there is no path, it says so instead of inventing one.
-
-![A selected memory: body, neighbours and the reason for each edge](assets/demo.jpg)
-
-## Search and spread
-
-Search covers titles, bodies and keywords. Results mark direct hits versus nodes reached by spreading, and cross-workspace hits are called out so nothing hides behind the current filter.
-
-“Spread from here” is not a second search box: it starts at this memory and walks existing links up to two hops, answering “what else is attached to this?”
-
-## Hygiene, including the ugly parts
-
-Six read-only checks:
-
-- near-duplicate memories;
-- oversized memories;
-- stale memories that have not changed in a long time;
-- memories stored but never read;
-- many edges but no strong relation (keyword noise);
-- completely isolated memories.
-
-Diagnosis only. It never merges, deletes or rewires anything; that stays your call.
-
-## Demo galaxy and first-visit guide
-
-An empty store no longer means a black screen: open the **demo galaxy**, 64 fictional memories that are read-only and never touch your stores. First time in a real store, a three-step guide helps you search a word, open one memory and see its evidence, then Shift-click two nodes to see how they connect. Skip it anytime; press `?` to replay.
-
-![First time? Three steps is enough](assets/onboard.jpg)
-
-## Share a slice, not the whole store
-
-Export the whole map, the current workspace, or “two hops around this memory” as a **single-file HTML**:
-
-- double-click to open, no DSH and no network;
-- search, neighbours and evidence chains run locally;
-- read-only, cannot touch your stores;
-- optionally hide the Diary workspace before export; `scope`/`hide` affect the export only;
-- the file is a point-in-time copy **including the text** — check the scope before sending it.
-
-![An exported offline snapshot: only the selected slice](assets/snapshot.jpg)
-
-## Ten thousand memories, still opens
-
-Same-machine benchmark with 10,000 memories and 78,593 links:
-
-- graph build 1.2 s;
-- first paint about 0.6 s;
-- recall 4.9 ms;
-- single-file snapshot 122 ms / 21 MB;
-- heap 145 MB.
-
-Above 3,000 memories it stays in the 2D nebula and asks you to filter before switching to 3D: software rendering would choke, and the map does not pretend otherwise.
-
-![A 10,000-memory 2D nebula](assets/bench-10k.jpg)
-
-## Install
-
-Add this index in the DSH plugin market:
-
-```text
-https://cdn.jsdelivr.net/npm/dsh-memory-nebula@1.1.1/market/index.json
-```
-
-Then search for “Memory Nebula” and install. Or install the npm package directly:
-
-```bash
-npm install dsh-memory-nebula@1.1.1
-```
-
-Run the read-only doctor after installing:
-
-```bash
-npx memory-nebula-doctor --profile <your DSH profile directory>
-```
-
-It checks the Node version, package files, profile dependency, bundle declaration, memory sources and an optional `/ping` endpoint. Nothing is uploaded.
-
-## Boundaries
-
-- single-user, local tool: no accounts, do not expose it to the public internet;
-- the stores are read-only: the map never edits memories for you;
-- 3D needs WebGL2;
-- snapshots contain the text inside the selected scope; treat them as plain-text backups;
-- server-side diagnostics and logs are mostly Chinese; the UI is bilingual.
-
-## More
-
-- architecture, data model, HTTP API, compatibility and distribution docs ship in `docs/`
-- License: MIT
+# Memory Nebula
+
+Turn memories scattered across workspaces into a star map you can walk, query and carry away.
+
+Memories in DSH often live across dozens of workspaces and thousands of sessions. Search returns a list, but not **why** those memories relate; switch projects and the thread disappears. Memory Nebula reads the same stores (meow-memory / OpenViking) and builds a graph: nodes are memories, edges are shared keywords, entities, projects, session co-activation and manual curation. It never copies or rewrites your stores. Everything is computed locally.
+
+![Memory Nebula: a thousand memories and their links](assets/preview.png)
+
+## Three ways to look
+
+- **Nebula**: every memory in one cloud, clustered by workspace and coloured by level and link strength. Good for “what does my memory look like as a whole?”
+- **Galaxy**: grouped by workspace; open a cluster to enter that project. Good for “what have I thought about here?”
+- **Time**: memories arranged as a river by creation/update time, with a scrubber. Good for “how did this grow?”
+
+All three share the same filters: workspace, level, edge type, minimum strength and importance. Switching views keeps the selection.
+
+## From one memory to another
+
+Select a node and the right panel shows what it connects to and why: same project, shared keywords, same entity, repeatedly co-mentioned in sessions, or a link you made yourself. Every neighbour carries a weight and a reason, not an evidence-free “similarity 0.87”.
+
+Shift-click a second node to find an **evidence chain**: every hop, its reason and the chain strength. If there is no path, it says so instead of inventing one.
+
+![A selected memory: body, neighbours and the reason for each edge](assets/demo.jpg)
+
+## Search and spread
+
+Search covers titles, bodies and keywords. Results mark direct hits versus nodes reached by spreading, and cross-workspace hits are called out so nothing hides behind the current filter.
+
+“Spread from here” is not a second search box: it starts at this memory and walks existing links up to two hops, answering “what else is attached to this?”
+
+## Hygiene, including the ugly parts
+
+Six read-only checks:
+
+- near-duplicate memories;
+- oversized memories;
+- stale memories that have not changed in a long time;
+- memories stored but never read;
+- many edges but no strong relation (keyword noise);
+- completely isolated memories.
+
+Diagnosis only. It never merges, deletes or rewires anything; that stays your call.
+
+## Demo galaxy and first-visit guide
+
+An empty store no longer means a black screen: open the **demo galaxy**, 64 fictional memories that are read-only and never touch your stores. First time in a real store, a three-step guide helps you search a word, open one memory and see its evidence, then Shift-click two nodes to see how they connect. Skip it anytime; press `?` to replay.
+
+![First time? Three steps is enough](assets/onboard.jpg)
+
+## Share a slice, not the whole store
+
+Export the whole map, the current workspace, or “two hops around this memory” as a **single-file HTML**:
+
+- double-click to open, no DSH and no network;
+- search, neighbours and evidence chains run locally;
+- read-only, cannot touch your stores;
+- optionally hide the Diary workspace before export; `scope`/`hide` affect the export only;
+- the file is a point-in-time copy **including the text** — check the scope before sending it.
+
+![An exported offline snapshot: only the selected slice](assets/snapshot.jpg)
+
+## Ten thousand memories, still opens
+
+Same-machine benchmark with 10,000 memories and 78,593 links:
+
+- graph build 1.2 s;
+- first paint about 0.6 s;
+- recall 4.9 ms;
+- single-file snapshot 122 ms / 21 MB;
+- heap 145 MB.
+
+Above 3,000 memories it stays in the 2D nebula and asks you to filter before switching to 3D: software rendering would choke, and the map does not pretend otherwise.
+
+![A 10,000-memory 2D nebula](assets/bench-10k.jpg)
+
+## Install
+
+Add this index in the DSH plugin market:
+
+```text
+https://unpkg.com/dsh-memory-nebula@1.1.2/market/index.json
+```
+
+Then search for “Memory Nebula” and install. Or install the npm package directly:
+
+```bash
+npm install dsh-memory-nebula@1.1.2
+```
+
+Run the read-only doctor after installing:
+
+```bash
+npx memory-nebula-doctor --profile <your DSH profile directory>
+```
+
+It checks the Node version, package files, profile dependency, bundle declaration, memory sources and an optional `/ping` endpoint. Nothing is uploaded.
+
+## Boundaries
+
+- single-user, local tool: no accounts, do not expose it to the public internet;
+- the stores are read-only: the map never edits memories for you;
+- 3D needs WebGL2;
+- snapshots contain the text inside the selected scope; treat them as plain-text backups;
+- server-side diagnostics and logs are mostly Chinese; the UI is bilingual.
+
+## More
+
+- architecture, data model, HTTP API, compatibility and distribution docs ship in `docs/`
+- License: MIT

package/docs/market.zh.md

@@ -1,109 +1,109 @@
-# 记忆星图
-
-把散落在工作区里的记忆，连成一张能走、能查、能带走的星图。
-
-DSH 里的记忆常常躺在几十个工作区、几千条会话里。搜索能给出一串结果，但看不到它们**为什么**相关；换一个项目，之前那条线索又断了。记忆星图读同一批记忆库（meow-memory / OpenViking），把它们建成一张图：节点是记忆，边是关键词、实体、项目、会话共现和人工策展。它不复制、不改写你的库，所有计算都在本机完成。
-
-![记忆星图：一千多条记忆与它们的连接](assets/preview.png)
-
-## 三种看法
-
-- **星云**：全部记忆铺成星云，按工作区聚成星团，按层级与连接强度着色。适合看「我的记忆整体长什么样」。
-- **星系**：按工作区分组，点开一团就进入那个项目。适合回答「这个项目里我都想过什么」。
-- **时间**：按创建/更新时间排成一条河，拖动时间轴回放。适合看「这件事是怎么一步步长出来的」。
-
-三种视图共用同一套筛选：工作区、层级、连接类型、最小强度、最低重要度。切视图不丢选中。
-
-## 从一条记忆走到另一条
-
-选中节点，右侧会告诉你它和谁相连、为什么相连：同项目、共享关键词、同一实体、被同一批会话反复一起提到，或者你亲手连的。每条邻居都带权重和理由，不是「相似度 0.87」这种没有证据的数字。
-
-Shift 点第二个节点，会找两点之间的**证据链**：每一跳是哪条边、什么理由、链强度多少。找不到就明说找不到，不编一条路。
-
-![选中一条记忆：正文、邻居、每条边的理由](assets/demo.jpg)
-
-## 搜索与扩散
-
-搜索是全文的：标题、正文、关键词都算。结果里会标出是直接命中，还是从别的节点扩散过来的；跨工作区命中会单独标出来，避免「明明有却搜不到」。
-
-「以它为种子扩散」不是搜索的替代品：它从这条记忆出发，沿已有连接看两跳内的邻居，告诉你「还有哪些东西和它连着」。
-
-## 体检：不只看漂亮的那面
-
-星图会做六类只读体检：
-
-- 近似重复的记忆；
-- 过长的记忆；
-- 很久没更新、可能过时的记忆；
-- 存了但从没被读过的记忆；
-- 边很多但没有一条强关系（关键词撞出来的）；
-- 完全孤立的记忆。
-
-只诊断，不自动改库。要不要合并、删掉、补连接，由你决定。
-
-## 示例星系与首访引导
-
-第一次打开时，如果库还是空的，不会只给你一片黑：可以进「示例星系」逛一圈。示例是 64 条**虚构**记忆，只读，不写盘，退出后回到你自己的库。第一次进真实库还有三步引导：搜一个词、点开一条看证据、Shift 点两条看它们怎么连。跳过随时可以按 `?` 重看。
-
-![第一次来？三步就够](assets/onboard.jpg)
-
-## 分享一片，而不是整座库
-
-可以把整张图、当前工作区、或「以某条记忆为中心 2 跳」导出成一个**单文件 HTML**：
-
-- 双击就能看，不需要 DSH、不需要网络；
-- 搜索、邻居、证据链都在本地算；
-- 只读，不能改你的库；
-- 导出前可以勾选「隐去日记工作区」；`scope`/`hide` 只影响导出，不动原库；
-- 分享出去的是时间点副本，正文会一起带走——发给别人前先确认范围。
-
-![导出的离线快照：文件里只有被选中的那一片](assets/snapshot.jpg)
-
-## 一万条也敢开
-
-10,000 条记忆、78,593 条连接的同机基准：
-
-- 建图 1.2 秒；
-- 首屏约 0.6 秒；
-- 检索 4.9 毫秒；
-- 单文件快照 122 毫秒 / 21MB；
-- 堆内存 145MB。
-
-超过 3,000 条时默认留在二维星云，并提示先筛选再切三维：软件渲染下三维会卡，星图不假装自己扛得住。
-
-![一万条记忆的二维星云](assets/bench-10k.jpg)
-
-## 安装
-
-在 DSH 的插件市场里添加索引：
-
-```text
-https://cdn.jsdelivr.net/npm/dsh-memory-nebula@1.1.1/market/index.json
-```
-
-然后搜索「记忆星图」安装。也可以直接装 npm 包：
-
-```bash
-npm install dsh-memory-nebula@1.1.1
-```
-
-装完先体检：
-
-```bash
-npx memory-nebula-doctor --profile <你的 DSH profile 目录>
-```
-
-它会检查 Node 版本、包文件、profile 依赖、bundle 声明、记忆源和可选的 `/ping` 端点。只读，不上传任何东西。
-
-## 边界
-
-- 本机单用户工具，没有账号系统，不要把它暴露到公网；
-- 记忆库是只读的：星图不会替你改记忆；
-- 三维视图需要 WebGL2；
-- 单文件快照包含被选中范围内的正文，按明文备份对待；
-- 服务端诊断与日志仍以中文为主，界面中英双语。
-
-## 更多
-
-- 架构、数据模型、HTTP 接口、兼容矩阵与分发说明都随包发布，见 `docs/`
-- 许可：MIT
+# 记忆星图
+
+把散落在工作区里的记忆，连成一张能走、能查、能带走的星图。
+
+DSH 里的记忆常常躺在几十个工作区、几千条会话里。搜索能给出一串结果，但看不到它们**为什么**相关；换一个项目，之前那条线索又断了。记忆星图读同一批记忆库（meow-memory / OpenViking），把它们建成一张图：节点是记忆，边是关键词、实体、项目、会话共现和人工策展。它不复制、不改写你的库，所有计算都在本机完成。
+
+![记忆星图：一千多条记忆与它们的连接](assets/preview.png)
+
+## 三种看法
+
+- **星云**：全部记忆铺成星云，按工作区聚成星团，按层级与连接强度着色。适合看「我的记忆整体长什么样」。
+- **星系**：按工作区分组，点开一团就进入那个项目。适合回答「这个项目里我都想过什么」。
+- **时间**：按创建/更新时间排成一条河，拖动时间轴回放。适合看「这件事是怎么一步步长出来的」。
+
+三种视图共用同一套筛选：工作区、层级、连接类型、最小强度、最低重要度。切视图不丢选中。
+
+## 从一条记忆走到另一条
+
+选中节点，右侧会告诉你它和谁相连、为什么相连：同项目、共享关键词、同一实体、被同一批会话反复一起提到，或者你亲手连的。每条邻居都带权重和理由，不是「相似度 0.87」这种没有证据的数字。
+
+Shift 点第二个节点，会找两点之间的**证据链**：每一跳是哪条边、什么理由、链强度多少。找不到就明说找不到，不编一条路。
+
+![选中一条记忆：正文、邻居、每条边的理由](assets/demo.jpg)
+
+## 搜索与扩散
+
+搜索是全文的：标题、正文、关键词都算。结果里会标出是直接命中，还是从别的节点扩散过来的；跨工作区命中会单独标出来，避免「明明有却搜不到」。
+
+「以它为种子扩散」不是搜索的替代品：它从这条记忆出发，沿已有连接看两跳内的邻居，告诉你「还有哪些东西和它连着」。
+
+## 体检：不只看漂亮的那面
+
+星图会做六类只读体检：
+
+- 近似重复的记忆；
+- 过长的记忆；
+- 很久没更新、可能过时的记忆；
+- 存了但从没被读过的记忆；
+- 边很多但没有一条强关系（关键词撞出来的）；
+- 完全孤立的记忆。
+
+只诊断，不自动改库。要不要合并、删掉、补连接，由你决定。
+
+## 示例星系与首访引导
+
+第一次打开时，如果库还是空的，不会只给你一片黑：可以进「示例星系」逛一圈。示例是 64 条**虚构**记忆，只读，不写盘，退出后回到你自己的库。第一次进真实库还有三步引导：搜一个词、点开一条看证据、Shift 点两条看它们怎么连。跳过随时可以按 `?` 重看。
+
+![第一次来？三步就够](assets/onboard.jpg)
+
+## 分享一片，而不是整座库
+
+可以把整张图、当前工作区、或「以某条记忆为中心 2 跳」导出成一个**单文件 HTML**：
+
+- 双击就能看，不需要 DSH、不需要网络；
+- 搜索、邻居、证据链都在本地算；
+- 只读，不能改你的库；
+- 导出前可以勾选「隐去日记工作区」；`scope`/`hide` 只影响导出，不动原库；
+- 分享出去的是时间点副本，正文会一起带走——发给别人前先确认范围。
+
+![导出的离线快照：文件里只有被选中的那一片](assets/snapshot.jpg)
+
+## 一万条也敢开
+
+10,000 条记忆、78,593 条连接的同机基准：
+
+- 建图 1.2 秒；
+- 首屏约 0.6 秒；
+- 检索 4.9 毫秒；
+- 单文件快照 122 毫秒 / 21MB；
+- 堆内存 145MB。
+
+超过 3,000 条时默认留在二维星云，并提示先筛选再切三维：软件渲染下三维会卡，星图不假装自己扛得住。
+
+![一万条记忆的二维星云](assets/bench-10k.jpg)
+
+## 安装
+
+在 DSH 的插件市场里添加索引：
+
+```text
+https://unpkg.com/dsh-memory-nebula@1.1.2/market/index.json
+```
+
+然后搜索「记忆星图」安装。也可以直接装 npm 包：
+
+```bash
+npm install dsh-memory-nebula@1.1.2
+```
+
+装完先体检：
+
+```bash
+npx memory-nebula-doctor --profile <你的 DSH profile 目录>
+```
+
+它会检查 Node 版本、包文件、profile 依赖、bundle 声明、记忆源和可选的 `/ping` 端点。只读，不上传任何东西。
+
+## 边界
+
+- 本机单用户工具，没有账号系统，不要把它暴露到公网；
+- 记忆库是只读的：星图不会替你改记忆；
+- 三维视图需要 WebGL2；
+- 单文件快照包含被选中范围内的正文，按明文备份对待；
+- 服务端诊断与日志仍以中文为主，界面中英双语。
+
+## 更多
+
+- 架构、数据模型、HTTP 接口、兼容矩阵与分发说明都随包发布，见 `docs/`
+- 许可：MIT

package/docs/operations.md

@@ -68,7 +68,7 @@ memory-nebula --help          # 装成 bin 之后
 
 ## 发布信息
 
-- 版本 `1.1.1`，`license: MIT`，无运行时依赖。
+- 版本 `1.1.2`，`license: MIT`，无运行时依赖。
 - `files`：`lib/`、`cordis.patch.yml`、`README.md`、`README.en.md`、`docs/`、`CHANGELOG.md`、
   `LICENSE`、`scripts/`、`sidecar.ps1`。
 - 装完先体检：`node scripts/doctor.mjs --profile <profile> [--root DIR] [--base URL] [--json]`；

package/docs/release.md

@@ -59,8 +59,8 @@ npm pack --dry-run
 npm pack                       # 生成 dsh-memory-nebula-<version>.tgz
 ```
 
-1.1.1 实测：**72 个文件**，包 2.3 MB、解开 2.9 MB（`docs/assets/` 的预览图与基准脚本占大头）。
-干净目录安装后 `--version` = 1.1.1，导出 `apply,inject,name,version`；
+1.1.2 实测：**72 个文件**，包 2.3 MB、解开 2.9 MB（`docs/assets/` 的预览图与基准脚本占大头）。
+干净目录安装后 `--version` = 1.1.2，导出 `apply,inject,name,version`；
 安装包起 7896 实测 `/share?scope=ws:DSHLive2D&hide=level:lesson` → 200、601,177 字节。
 清单里应含 `sidecar.ps1`、`scripts/`、`test/` 不含（单测只在仓库里跑）、`dev/` 与截图不含。
 
@@ -71,7 +71,7 @@ npm pack                       # 生成 dsh-memory-nebula-<version>.tgz
 New-Item -ItemType Junction -Path "<profile>\node_modules\dsh-memory-nebula" -Target "<plugin>"
 
 # 方式 B：tarball
-npm install <path>\dsh-memory-nebula-1.1.1.tgz --prefix <profile>
+npm install <path>\dsh-memory-nebula-1.1.2.tgz --prefix <profile>
 ```
 
 然后确认 profile `package.json` 里：

package/docs/verified.md

@@ -2,6 +2,19 @@
 
 > 下面的数字都是**当时实测**的值，记录的是「这个结论被什么证据支撑过」，不是当前规模——记忆每天在长，规模类数字请以 `node scripts/smoke.mjs` 和界面顶栏为准。
 
+## 1.1.2（2026-11-10）换个门牌
+
+- **CDN 选择**：本机 `curl` 与 Node fetch 访问 jsDelivr 均为 `ECONNRESET`；unpkg 的 Node fetch 200。
+- **远程索引**：`https://unpkg.com/dsh-memory-nebula@1.1.2/market/index.json` 经
+  `parseMarketIndex` 解析通过，1 个插件、版本 1.1.2；`marketInstallTarget` 返回
+  `dsh-memory-nebula@1.1.2`。
+- **详情资产**：icon.svg（image/svg+xml）、locale/zh.json、docs/market.zh.md、
+  docs/assets/demo.jpg 均 200。
+- **本包**：87/87 单测；release-check 1.1.2 全绿；干净安装 `--version` = 1.1.2。
+- **发布**：npm `latest` = 1.1.2；registry 安装 `dsh-memory-nebula@1.1.2` 后 doctor 通过。
+- **市场端**：DSH 索引配置后可在发现页搜索「记忆星图」并安装；若宿主网络无法访问 unpkg，
+  改用本机 `file://` 索引或 npm 直装，不影响包本身。
+
 ## 1.1.1（2026-11-10）上架
 
 - **市场索引校验**：`market/index.json` 经 `@deepseek-ai/dsh-plugin-market-index` 的

package/lib/web/app.js

@@ -1,7 +1,7 @@
 /* 记忆星图 —— 界面逻辑
    数据：/api/memory-nebula/*（同源）。渲染交给 render.js。 */
 ;(function () {
-  const APP_VERSION = '1.1.1'  // 和 package.json 同步（release-check 会核对）
+  const APP_VERSION = '1.1.2'  // 和 package.json 同步（release-check 会核对）
   'use strict'
   /** 中文即源文：英文模式查 i18n.js 的表，查不到就原样返回中文。 */
   const T = (zh, vars) => (window.NebulaI18n ? window.NebulaI18n.t(zh, vars) : zh)