tavern-harness 0.1.13 → 0.1.15

package/CHANGELOG.md

@@ -3,6 +3,120 @@
 > 本文件由 `scripts/changelog.mjs` 从 `changelog/v*.md` 生成，请勿手动编辑。
 > 版本对比链接见文末；源文件位于仓库 `changelog/` 目录。
 
+## [0.1.15] - 2026-10-25
+
+### 中文
+
+#### 新增
+
+- 本地命令支持管道与重定向：可用 `|`、`<`、`>`、`>>`、`2>`、`2>&1` 组合命令，例如 `cat log.txt | grep error`；读写路径仍限制在会话工作区内，变量展开、命令替换、heredoc、后台执行与交互式命令依旧不支持
+- 工作区文件管理器新增「在文件管理器中打开」：一键在运行 Tavern Harness 的设备上打开当前会话的工作目录；通过局域网从手机访问时会提示仅限本机
+
+#### 改进
+
+- HTML 预览支持多文件应用：HTML 现在可以加载同目录下的脚本、样式、图片和字体，沙箱内的指针锁定等交互也可正常使用
+- 大工具参数的流式生成更流畅：参数较大时自动降低草稿刷新与本地写入频率，只保留最新快照，减少界面卡顿和写入积压
+- 画中画预览更易操作：可从四角任意缩放、拖动标题栏移动窗口，移动端初始窗口尺寸更合适
+
+#### 修复
+
+- 修复模型输出中的原始 HTML 被当成页面结构渲染的问题：现在按纯文本显示，消息排版不再被意外改变
+- 修复生成大型工具参数（例如写入大文件）耗时较长时，已建立的响应可能因 60 秒无数据而中断的问题；现在只有建立连接受 20 秒超时限制，生成过程可随时手动停止
+
+#### 内部
+
+- 新增 `/preview_read` 端点与 Vite 预览中间件：按会话工作区读取预览资源，上限 8MB，并按扩展名返回正确的 MIME 类型
+- 命令执行改为流式连接管道各段与重定向文件，超时或出错时结束整个进程组
+- 补充管道、重定向与预览端点的测试，并修正生成技能的工具与参数说明
+
+### English
+
+#### Added
+
+- Local commands now support pipes and redirection: combine commands with `|`, `<`, `>`, `>>`, `2>`, and `2>&1`, for example `cat log.txt | grep error`; read and write paths stay inside the session workspace, and expansion, command substitution, heredocs, background jobs, and interactive commands remain unsupported
+- The workspace file manager has an "Open in file manager" button that opens the current conversation's workspace on the machine running Tavern Harness; when the app is opened over the LAN from a phone, it explains that this only works on the host device
+
+#### Changed
+
+- HTML previews now support multi-file apps: HTML can load same-directory scripts, styles, images, and fonts, and sandboxed interactions such as pointer lock work
+- Smoother streaming for large tool arguments: once an argument gets large, draft updates and local writes happen less often and only the latest snapshot is kept, reducing UI stalls and write backlogs
+- Picture-in-picture previews are easier to use: resize from any corner, drag the title bar to move the window, and the initial window size fits mobile screens better
+
+#### Fixed
+
+- Fixed raw HTML in model output being rendered as page structure: it now appears as plain text, so message layout is no longer altered unexpectedly
+- Fixed connected responses being interrupted after 60 seconds without data while generating large tool arguments (such as writing a big file); only establishing the connection is still subject to a 20-second timeout, and generation can be stopped manually at any time
+
+#### Internal
+
+- Added a `/preview_read` endpoint and Vite preview middleware that reads preview assets from the session workspace, caps them at 8MB, and serves the correct MIME type by extension
+- Command execution now streams between pipeline stages and to and from redirected files, ending the whole process group on timeout or failure
+- Added tests for pipes, redirection, and the preview endpoint, and clarified generated-skill tool and parameter descriptions
+
+## [0.1.14] - 2026-10-24
+
+### 中文
+
+#### 新增
+
+- 设置 → 生成参数新增「重置」按钮，一键恢复温度、Top P、Top K 等默认值；已是默认值时按钮不可用
+
+#### 改进
+
+- 启动体验：界面先于数据加载显示，启动期间展示加载动画而非空白页；启动数据合并为一次快照请求，长列表通过增量渲染与缓存保持流畅
+- 成就页重新排版：奖杯陈列柜置顶，生涯统计与「重置统计」分区展示
+- 统一弹窗尺寸（紧凑 / 标准 / 宽屏三档）并适配移动端；弹窗不再响应点击遮罩关闭，避免编辑内容时误触丢失
+- 流式生成更稳定：自动跟随最新内容，向上回看历史时不会被拉回底部；键盘弹出时思考与回复内容不再被输入框遮挡
+- 移动端消息编辑器：键盘开合过渡更平滑，消息操作与编辑时的附件按钮更易用
+- 聊天附件、思考面板与工具卡片的样式更清晰，浅色主题配色更柔和统一
+
+#### 修复
+
+- 修复生成失败或模型返回空内容时界面停留在「生成中」的问题：现在会回滚本轮并提示；已收到的部分正文会保留并标注「生成已中断」
+- 修复通过局域网地址（侧栏二维码）在手机上打开时，回合可能一直卡住没有回复的问题；连接超时会自动改用同源代理，全部失败时明确提示网络错误
+- 修复刷新或切换标签页时短暂闪出错误主题的问题，首帧即应用正确的深浅色背景
+- 修复首页角色卡片不显示自定义头像的问题
+- 修复移动端消息操作按钮、编辑附件动作与聊天附件样式的问题
+- 修复启动时模型服务列表可能为空、旧会话消息加载较慢的问题
+
+#### 内部
+
+- 持久化服务新增启动快照与按字段查询接口，减少启动请求数量，并补充相应测试
+- 新增头像缩略图与行指纹缓存，降低长列表重复渲染开销
+- 统一弹窗组件与尺寸定义，移除重复的遮罩点击逻辑
+- 补全 yolo 集成测试所需的持久化接口 mock
+
+### English
+
+#### Added
+
+- New "Reset" button in Settings → Generation parameters restores the default temperature, Top P, Top K, and related values; it is disabled when everything is already at defaults
+
+#### Changed
+
+- Startup: the UI mounts before data loads and shows a loading indicator instead of a blank page; startup data is fetched in a single snapshot request, and long lists stay smooth through incremental rendering and caching
+- Achievements page reorganized: the trophy shelf comes first, with career stats and a separate "Reset stats" section below
+- Modals now share three sizes (compact, standard, wide) and adapt to mobile; clicking the backdrop no longer dismisses them, preventing accidental loss of edits
+- Streaming is more stable: the view follows new content automatically but stops following when you scroll up to read history, and the keyboard no longer covers thinking or reply content
+- Mobile message editor: smoother keyboard transitions and easier access to attachment actions while editing
+- Clearer styling for chat attachments, thinking panels, and tool cards, with softer, more consistent light-theme colors
+
+#### Fixed
+
+- Fixed the UI getting stuck in "generating" when generation failed or the model returned no content: the turn now rolls back with a clear message, while partial text is kept and marked as interrupted
+- Fixed turns occasionally hanging without a reply when the app is opened on a phone through its LAN address (sidebar QR code); connection timeouts now fall back to the same-origin proxy and report a network error when every candidate fails
+- Fixed a brief flash of the wrong theme when refreshing or switching tabs; the correct light/dark background is applied on the first frame
+- Fixed character cards on the home page not showing custom avatars
+- Fixed mobile message action buttons, attachment actions while editing, and chat attachment styling
+- Fixed the model provider list sometimes being empty at startup and slow loading of older session messages
+
+#### Internal
+
+- Added startup snapshot and list-by-field persistence endpoints to reduce startup requests, with tests
+- Added avatar thumbnail and row fingerprint caches to cut repeated rendering work in long lists
+- Unified the modal component and sizing, removing duplicate backdrop-click logic
+- Filled in the persistence endpoint mocks required by the yolo integration test
+
 ## [0.1.13] - 2026-10-21
 
 ### 中文
@@ -330,6 +444,8 @@
 - Vite + React + Zustand + Dexie stack with browser-side local persistence
 
 <!-- 链接定义 -->
+[0.1.15]: https://github.com/Rannichan/Tavern-Harness/compare/v0.1.14...v0.1.15
+[0.1.14]: https://github.com/Rannichan/Tavern-Harness/compare/v0.1.13...v0.1.14
 [0.1.13]: https://github.com/Rannichan/Tavern-Harness/compare/v0.1.12...v0.1.13
 [0.1.12]: https://github.com/Rannichan/Tavern-Harness/compare/v0.1.11...v0.1.12
 [0.1.11]: https://github.com/Rannichan/Tavern-Harness/compare/v0.1.10...v0.1.11

package/index.html

@@ -14,7 +14,7 @@
           const systemDark = window.matchMedia('(prefers-color-scheme: dark)').matches;
           const dark = mode === 'dark' || (mode !== 'light' && systemDark);
           const root = document.documentElement;
-          const background = dark ? '#20160E' : '#F7F2E9';
+          const background = dark ? '#20160E' : '#F7F0E4';
           root.dataset.theme = dark ? 'dark' : 'light';
           root.style.colorScheme = dark ? 'dark' : 'light';
           root.style.backgroundColor = background;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tavern-harness",
-  "version": "0.1.13",
+  "version": "0.1.15",
   "description": "Tavern Harness — 酒馆模式下的本地 AI Agent 助手（MyAgent-Android 的 Web 复刻）",
   "private": false,
   "type": "module",

package/persistence-server.mjs

@@ -152,6 +152,18 @@ function saveRecord(collection, key, value) {
 
 const getRecord = database.prepare('SELECT value_json, updated_at FROM app_records WHERE collection = ? AND record_key = ?');
 const listRecords = database.prepare('SELECT record_key, value_json, updated_at FROM app_records WHERE collection = ? ORDER BY record_key');
+const listRecordsByJsonField = database.prepare(`
+  SELECT record_key, value_json, updated_at
+  FROM app_records
+  WHERE collection = ? AND json_extract(value_json, ?) = ?
+  ORDER BY record_key
+`);
+const listStartupRecords = database.prepare(`
+  SELECT collection, record_key, value_json, updated_at
+  FROM app_records
+  WHERE collection IN ('settings', 'providers', 'npcs', 'sessions', 'worldBooks', 'tools', 'achievementUnlocks')
+  ORDER BY collection, record_key
+`);
 const deleteRecord = database.prepare('DELETE FROM app_records WHERE collection = ? AND record_key = ?');
 const clearCollection = database.prepare('DELETE FROM app_records WHERE collection = ?');
 const nextNumericRecordKey = database.prepare(`
@@ -220,6 +232,31 @@ const server = createServer(async (req, res) => {
       json(res, 200, { ok: true, leaseId: writeLock.leaseId, lock: lockState(clientId) });
       return;
     }
+    if (pathname === '/startup/snapshot') {
+      if (!clientId || clientId.length > 255) {
+        json(res, 400, { ok: false, message: 'A valid clientId is required' });
+        return;
+      }
+      if (!writeLock) setWriteLock({ clientId, leaseId: randomBytes(24).toString('base64url') });
+      const collections = {
+        settings: [],
+        providers: [],
+        npcs: [],
+        sessions: [],
+        worldBooks: [],
+        tools: [],
+        achievementUnlocks: [],
+      };
+      for (const row of listStartupRecords.all()) {
+        collections[row.collection].push({
+          key: row.record_key,
+          value: JSON.parse(row.value_json),
+          updatedAt: row.updated_at,
+        });
+      }
+      json(res, 200, { ok: true, lock: lockState(clientId), collections });
+      return;
+    }
     if (!pathname.startsWith('/records/')) {
       json(res, 404, { ok: false, message: `Unknown persistence endpoint: ${pathname}` });
       return;
@@ -231,8 +268,23 @@ const server = createServer(async (req, res) => {
       return;
     }
 
-    if (pathname === '/records/list') {
-      const records = listRecords.all(collection).map((row) => ({
+    if (pathname === '/records/list' || pathname === '/records/list-by-field') {
+      let rows;
+      if (pathname === '/records/list-by-field') {
+        const { field, value } = body;
+        if (typeof field !== 'string' || !/^[A-Za-z][A-Za-z0-9_]{0,63}$/.test(field)) {
+          json(res, 400, { ok: false, message: 'Invalid record field' });
+          return;
+        }
+        if (typeof value !== 'string' && typeof value !== 'number' && typeof value !== 'boolean') {
+          json(res, 400, { ok: false, message: 'Invalid record field value' });
+          return;
+        }
+        rows = listRecordsByJsonField.all(collection, `$.${field}`, value);
+      } else {
+        rows = listRecords.all(collection);
+      }
+      const records = rows.map((row) => ({
         key: row.record_key,
         value: JSON.parse(row.value_json),
         updatedAt: row.updated_at,