fix(local): remux aac into .m4a so the extracted audio is bit-exact (#149)

Extraction always used Matroska (.mka) because it takes essentially any
audio codec. That is right for most codecs but wrong for AAC: MP4 records
the AAC encoder priming (the ~1000 warm-up samples every AAC encoder emits)
in an edit list, and the edit list does not survive into Matroska. The
remuxed track then decodes ~23 ms longer than the source, with the priming
samples played at the head instead of discarded.

Measured on a 5s 640x480 fixture: source audio decodes to 962980 bytes of
PCM, the .mka to 967440 — 4460 bytes / ~23 ms extra, peaking at -66 dBFS.
Inaudible in practice, but it also puts the track fractionally out of step
with its own reported duration, for no reason.

Pick the container by codec instead: aac -> .m4a (keeps the edit list),
everything else -> .mka as before. If the preferred container refuses the
codec, retry into .mka before falling back to keeping the whole video. AAC
is worth the special case because mp4 / mov / m4v — what people actually
upload — almost always carry it.

Adds the strongest available test of the "lossless" claim: decode the audio
straight out of the source mp4, decode the stored extract, assert the PCM is
byte-for-byte equal. Forcing .mka fails it with exactly the 4460-byte delta.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

README.md

@@ -32,7 +32,7 @@
 - **游客模式（免登录点歌，默认关闭）** — 管理员可选择允许访客**无需账号密码**进入 WebUI 点歌，并逐项配置游客权限（8 个开关，默认仅「添加到队列末尾」开启）与可控机器人白名单；游客无法查看 / 修改任何设置、管理机器人或访问用户管理。开启后登录页出现 **「以游客身份进入」**。详见下文 **「游客模式 / Guest mode」** 小节
 - **本地收藏歌单** — 在首页 / 搜索 / 歌单页一键收藏，收藏内容按用户存储，登录后跨设备同步
 - **保存/加载播放清单 + 重启后自动恢复队列（可选，默认关闭）** — 管理员在 设置 → 行为设置 开启后，可在网页「已存队列」页或聊天命令（`!save` / `!load` / `!queues`）把当前队列保存为清单，随时**替换**加载或**追加**到队列末尾；同时机器人重启后会自动恢复并继续播放上次的队列。网页保存可选「共享」，聊天保存进入共享清单。**说明**：重启只能从当前曲目的开头恢复（不记忆播放进度）；Spotify 自动恢复为尽力而为（依赖 sidecar 可用）。详见 [使用说明](#使用说明)
-- **本地音频上传播放** — 在搜索页拖拽或选择本地音频上传，上传后可直接播放 / 下一首播放 / 加入队列；管理员可在 设置 → 行为设置 开关此功能，播放结束或停止/清空/替换队列时会清理服务端接收的本地文件
+- **本地音视频上传播放** — 在搜索页拖拽或选择本地文件上传，音频（mp3 / flac / wav / m4a / ogg / opus 等）和视频（mp4 / mov / avi / mkv / flv / wmv 等）都支持，视频上传后只保留其中的音轨；上传后可直接播放 / 下一首播放 / 加入队列；管理员可在 设置 → 行为设置 开关此功能，播放结束或停止/清空/替换队列时会清理服务端接收的本地文件
 - **专属链接（单机器人锁定）** — 通过 `/bot/<id>` 专属链接打开 WebUI 时锁定到单个机器人，刷新后保持，适合把某台机器人的控制页分享给特定用户
 - **频道无人时自动暂停** — 机器人所在频道没有其他人时自动暂停播放，有人加入后自动恢复（**默认关闭**，可在设置中开启）
 - **Jellyfin 音源（可选）** — 连接自建 [Jellyfin](https://jellyfin.org/) 服务器作为额外音源：搜索（歌曲 / 专辑 / 歌单）、懒解析直传播放、同步歌词、收藏 Instant Mix 电台（`!fm -j`）、首页「最近添加 / 播放最多 / 收藏 / 流派」，并把播放进度回报给 Jellyfin（PlayCount / 播放状态）。**默认关闭**，在 设置 → Jellyfin 音乐库 一键开启。详见 [可选：Jellyfin 音源](#可选jellyfin-音源)
@@ -884,7 +884,26 @@ A：本项目内置 `/login` 限流（每 IP 每分钟 5 次），但生产部
 
 > 完整历史请查看 [git log](https://github.com/ZHANGTIANYAO1/teamspeak-music-bot/commits/main) 或 [Releases](https://github.com/ZHANGTIANYAO1/teamspeak-music-bot/releases)。这里只列出重要变更和面向用户的破坏性改动。
 
-### 最新版本 — v1.12.0：网站图标 / 移动端交互 / 安装脚本按 ABI 自愈
+### 最新版本 — v1.13.0：本地视频上传播放 / 头像上传时机
+
+处理了 2 个社区反馈的 issue。**没有配置变化，升级无需任何操作**；原有的本地音频上传行为完全不变。
+
+**本地视频上传播放（[#149](https://github.com/ZHANGTIANYAO1/teamspeak-music-bot/issues/149)，[PR #151](https://github.com/ZHANGTIANYAO1/teamspeak-music-bot/pull/151)，感谢 [@LadenceE](https://github.com/LadenceE)）**
+
+- 搜索页的本地上传现在也收**视频文件**：mp4 / mov / avi / mkv / flv / wmv / m4v / mpg / mpeg / 3gp / ts / m2ts / ogv。上传后当普通歌曲用——直接播放、下一首播放、加入队列都一样。
+- 视频**只保留音轨**：上传后立刻把音频流原样搬进一个音频容器（不重编码、无损），再删掉原视频。720p 素材实测落盘只剩原文件的 14%，不然十几个视频就把 5 GiB 的上传目录配额占满了。
+- 没有音轨的视频会在**上传时**就被拒绝并说明原因，而不是排进队列后静默跳过。
+- 单文件上限从 200 MB 提到 **500 MB**；超限时的报错从 Express 默认的 HTML 错误页（带堆栈和服务器绝对路径）换成正常的中文提示，浏览器端也会在开传前就拦下超大文件。
+- 上传进度按文件显示百分比，传完切到「服务端处理中」——视频比音频大得多，原先那句静止的「正在上传」看着像卡死。
+- 说明：这里做的是「把你本地磁盘上的文件传上来播放」。让机器人直接读取**服务器**磁盘上任意路径的文件没有做——那等于开一个全盘任意文件读取的口子，而「播放服务器上已有的媒体库」用 Jellyfin 音源即可。
+
+**初始化阶段不再发起注定失败的头像上传（[#148](https://github.com/ZHANGTIANYAO1/teamspeak-music-bot/issues/148)，[PR #150](https://github.com/ZHANGTIANYAO1/teamspeak-music-bot/pull/150)，感谢 [@shenmu-rua](https://github.com/shenmu-rua)）**
+
+- 机器人构造阶段读出已保存的自定义头像后会立刻发起文件传输，但那时 TeamSpeak 还没连上，这次传输必定失败。现在构造阶段只把头像数据装入内存，实际上传交给连接成功后的 `onConnect()`。
+- **影响范围说明**：头像本身一直是能正常显示的（连接成功后本来就会重新应用一次），所以这不是「头像丢了」。真正的代价是每次启动 / 重启都会多一次注定失败的请求和一条 `Profile update failed` 警告日志——现在没有了。
+- 顺带修掉一个边角：头像文件写到一半崩溃会留下 0 字节文件，原先这会再触发两个同样注定失败的请求。
+
+### v1.12.0：网站图标 / 移动端交互 / 安装脚本按 ABI 自愈
 
 一次性处理了 6 个社区反馈的 issue。**没有配置变化，升级无需任何操作**；`!play id:<id>` 等旧写法全部继续可用。
 

pr57.diff

@@ -0,0 +1,591 @@
+diff --git a/src/music/netease.test.ts b/src/music/netease.test.ts
+index b2adb3d..ecc2202 100644
+--- a/src/music/netease.test.ts
++++ b/src/music/netease.test.ts
+@@ -1,5 +1,5 @@
+ import { describe, it, expect } from "vitest";
+-import { parseLyrics } from "./netease.js";
++import { parseLyrics, mapNeteaseAlbums } from "./netease.js";
+ 
+ describe("NetEase adapter", () => {
+   it("parses LRC format lyrics", () => {
+@@ -28,4 +28,32 @@ describe("NetEase adapter", () => {
+     expect(lines[0].text).toBe("Hello world");
+     expect(lines[0].translation).toBe("你好世界");
+   });
++
++  it("mapNeteaseAlbums maps raw cloudsearch albums to Album shape", () => {
++    const raw = [
++      {
++        id: 42,
++        name: "Album A",
++        picUrl: "https://x/p.jpg",
++        artists: [{ name: "Artist X" }, { name: "Featured Y" }],
++        size: 12,
++      },
++      {
++        id: 99,
++        name: "Album B",
++        picUrl: "",
++        artists: [],
++      },
++    ];
++    expect(mapNeteaseAlbums(raw)).toEqual([
++      { id: "42", name: "Album A", artist: "Artist X / Featured Y", coverUrl: "https://x/p.jpg", songCount: 12, platform: "netease" },
++      { id: "99", name: "Album B", artist: "", coverUrl: "", songCount: 0, platform: "netease" },
++    ]);
++  });
++
++  it("mapNeteaseAlbums returns [] for empty/null input", () => {
++    expect(mapNeteaseAlbums([])).toEqual([]);
++    expect(mapNeteaseAlbums(null as any)).toEqual([]);
++    expect(mapNeteaseAlbums(undefined as any)).toEqual([]);
++  });
+ });
+diff --git a/src/music/netease.ts b/src/music/netease.ts
+index 8aaeab6..4a863db 100644
+--- a/src/music/netease.ts
++++ b/src/music/netease.ts
+@@ -8,6 +8,7 @@ import type {
+   SearchResult,
+   QrCodeResult,
+   AuthStatus,
++  Album,
+ } from "./provider.js";
+ 
+ export function parseLyrics(lrc: string, tlyric?: string): LyricLine[] {
+@@ -55,6 +56,18 @@ export function parseLyrics(lrc: string, tlyric?: string): LyricLine[] {
+   return lines.sort((a, b) => a.time - b.time);
+ }
+ 
++export function mapNeteaseAlbums(raw: any[] | null | undefined): Album[] {
++  if (!Array.isArray(raw)) return [];
++  return raw.map((a) => ({
++    id: String(a.id),
++    name: a.name ?? "",
++    artist: (a.artists ?? []).map((x: any) => x.name).join(" / "),
++    coverUrl: a.picUrl ?? "",
++    songCount: a.size ?? 0,
++    platform: "netease",
++  }));
++}
++
+ // NetEase quality levels: standard(128k) higher(192k) exhigh(320k) lossless(flac) hires(hi-res) jyeffect jymaster
+ export const NETEASE_QUALITY_LEVELS = [
+   { value: "standard", label: "标准 (128kbps)", bitrate: 128 },
+@@ -91,7 +104,7 @@ export class NeteaseProvider implements MusicProvider {
+   }
+ 
+   async search(query: string, limit = 20): Promise<SearchResult> {
+-    const [songRes, playlistRes] = await Promise.all([
++    const [songRes, playlistRes, albumRes] = await Promise.all([
+       this.api.get("/cloudsearch", {
+         params: { keywords: query, type: 1, limit, ...this.cookieParams },
+       }),
+@@ -103,6 +116,9 @@ export class NeteaseProvider implements MusicProvider {
+           ...this.cookieParams,
+         },
+       }),
++      this.api.get("/cloudsearch", {
++        params: { keywords: query, type: 10, limit: 5, ...this.cookieParams },
++      }),
+     ]);
+ 
+     const songs: Song[] = (songRes.data?.result?.songs ?? []).map(
+@@ -127,7 +143,9 @@ export class NeteaseProvider implements MusicProvider {
+       platform: "netease",
+     }));
+ 
+-    return { songs, playlists, albums: [] };
++    const albums = mapNeteaseAlbums(albumRes.data?.result?.albums);
++
++    return { songs, playlists, albums };
+   }
+ 
+   async getSongUrl(songId: string, quality?: string): Promise<string | null> {
+diff --git a/src/music/qq.test.ts b/src/music/qq.test.ts
+new file mode 100644
+index 0000000..4f606cf
+--- /dev/null
++++ b/src/music/qq.test.ts
+@@ -0,0 +1,43 @@
++import { describe, it, expect } from "vitest";
++import { mapQqAlbums } from "./qq.js";
++
++describe("QQ adapter", () => {
++  it("mapQqAlbums maps albumMID-style raw entries", () => {
++    const raw = [
++      {
++        albumMID: "abc",
++        albumName: "Aero",
++        singerName: "Singer A",
++      },
++      {
++        albumMID: "xyz",
++        albumName: "Beta",
++        singer: [{ name: "Singer B" }, { name: "Singer C" }],
++      },
++    ];
++    const out = mapQqAlbums(raw);
++    expect(out).toHaveLength(2);
++    expect(out[0]).toMatchObject({
++      id: "abc",
++      name: "Aero",
++      artist: "Singer A",
++      platform: "qq",
++    });
++    expect(out[0].coverUrl).toContain("T002R300x300M000abc.jpg");
++    expect(out[1].artist).toBe("Singer B / Singer C");
++    expect(out[1].coverUrl).toContain("xyz");
++  });
++
++  it("mapQqAlbums returns [] for empty/null input", () => {
++    expect(mapQqAlbums([])).toEqual([]);
++    expect(mapQqAlbums(null as any)).toEqual([]);
++    expect(mapQqAlbums(undefined as any)).toEqual([]);
++  });
++
++  it("mapQqAlbums falls back to albumPic when no albumMID", () => {
++    const raw = [{ albumName: "C", albumPic: "https://x/p.jpg", singerName: "S" }];
++    const out = mapQqAlbums(raw);
++    expect(out[0].coverUrl).toBe("https://x/p.jpg");
++    expect(out[0].id).toBe("");
++  });
++});
+diff --git a/src/music/qq.ts b/src/music/qq.ts
+index 9c0360d..1e7a8ae 100644
+--- a/src/music/qq.ts
++++ b/src/music/qq.ts
+@@ -8,6 +8,7 @@ import type {
+   SearchResult,
+   QrCodeResult,
+   AuthStatus,
++  Album,
+ } from "./provider.js";
+ import { parseLyrics } from "./netease.js";
+ 
+@@ -27,6 +28,26 @@ const qqFavApi = axios.create({
+   headers: { referer: "https://y.qq.com/" },
+ });
+ 
++export function mapQqAlbums(raw: any[] | null | undefined): Album[] {
++  if (!Array.isArray(raw)) return [];
++  return raw.map((a) => {
++    const id = String(a.albumMID ?? a.mid ?? a.albumID ?? "");
++    const artist = a.singerName
++      ?? (Array.isArray(a.singer) ? a.singer.map((s: any) => s.name).join(" / ") : "");
++    const coverUrl = id
++      ? `https://y.gtimg.cn/music/photo_new/T002R300x300M000${id}.jpg`
++      : (a.albumPic ?? "");
++    return {
++      id,
++      name: a.albumName ?? a.title ?? "",
++      artist,
++      coverUrl,
++      songCount: a.song_count ?? a.songCount ?? 0,
++      platform: "qq" as const,
++    };
++  });
++}
++
+ function computeGtk(pSkey: string): number {
+   let hash = 5381;
+   for (let i = 0; i < pSkey.length; i++) {
+@@ -65,11 +86,12 @@ export class QQMusicProvider implements MusicProvider {
+       req_0: {
+         module: "music.search.SearchCgiService",
+         method: "DoSearchForQQMusicDesktop",
+-        param: {
+-          searchid: "1",
+-          query,
+-          num_per_page: Math.min(limit, 50),
+-        },
++        param: { searchid: "1", query, num_per_page: Math.min(limit, 50), search_type: 0 },
++      },
++      req_album: {
++        module: "music.search.SearchCgiService",
++        method: "DoSearchForQQMusicDesktop",
++        param: { searchid: "1", query, num_per_page: 5, search_type: 8 },
+       },
+     });
+     const res = await qqDirectApi.get("/cgi-bin/musicu.fcg", {
+@@ -90,7 +112,10 @@ export class QQMusicProvider implements MusicProvider {
+       platform: "qq",
+     }));
+ 
+-    return { songs, playlists: [], albums: [] };
++    const albumList: any[] = res.data?.req_album?.data?.body?.album?.list ?? [];
++    const albums = mapQqAlbums(albumList);
++
++    return { songs, playlists: [], albums };
+   }
+ 
+   async getSongUrl(songId: string, quality?: string): Promise<string | null> {
+diff --git a/src/web/api/music.ts b/src/web/api/music.ts
+index b08f9a2..edf9c04 100644
+--- a/src/web/api/music.ts
++++ b/src/web/api/music.ts
+@@ -52,14 +52,20 @@ export function createMusicRouter(
+       ]);
+ 
+       const songs = [
+-        ...(neteaseResult.status === "fulfilled"
+-          ? neteaseResult.value.songs
+-          : []),
++        ...(neteaseResult.status === "fulfilled" ? neteaseResult.value.songs : []),
+         ...(qqResult.status === "fulfilled" ? qqResult.value.songs : []),
+         ...(bilibiliResult.status === "fulfilled" ? bilibiliResult.value.songs : []),
+       ];
++      const albums = [
++        ...(neteaseResult.status === "fulfilled" ? neteaseResult.value.albums : []),
++        ...(qqResult.status === "fulfilled" ? qqResult.value.albums : []),
++      ];
++      const playlists = [
++        ...(neteaseResult.status === "fulfilled" ? neteaseResult.value.playlists : []),
++        ...(qqResult.status === "fulfilled" ? qqResult.value.playlists : []),
++      ];
+ 
+-      res.json({ songs });
++      res.json({ songs, albums, playlists });
+     } catch (err) {
+       logger.error({ err }, "Unified search failed");
+       res.status(500).json({ error: (err as Error).message });
+diff --git a/src/web/api/player.ts b/src/web/api/player.ts
+index a9af600..4f0930b 100644
+--- a/src/web/api/player.ts
++++ b/src/web/api/player.ts
+@@ -313,6 +313,78 @@ export function createPlayerRouter(
+     }
+   });
+ 
++  // Play an album by ID — mirrors play-playlist but calls getAlbumSongs
++  router.post("/:botId/play-album", async (req, res) => {
++    try {
++      const bot = (req as any).bot;
++      const { albumId, platform } = req.body;
++      const provider = bot.getProviderFor(
++        platform === "bilibili" || platform === "qq" || platform === "youtube"
++          ? platform
++          : "netease"
++      );
++
++      // Stop current playback
++      bot.getPlayer().stop();
++      bot.getPlayer().resetFailures();
++
++      const songs = await provider.getAlbumSongs(albumId);
++      if (songs.length === 0) {
++        res.json({ message: "Album is empty" });
++        return;
++      }
++
++      // QQ-specific optimization: batch-resolve playable IDs to avoid
++      // wasting retries on region/copyright-restricted tracks.
++      let queueable: { id: string }[] = songs;
++      const totalCount = songs.length;
++      const qqLike = provider as { getPlayableSongIds?: (ids: string[]) => Promise<Set<string> | null> };
++      if (typeof qqLike.getPlayableSongIds === "function") {
++        const playable = await qqLike.getPlayableSongIds(songs.map((s: { id: string }) => s.id));
++        if (playable !== null) {
++          queueable = songs.filter((s: { id: string }) => playable.has(s.id));
++        }
++      }
++      if (queueable.length === 0) {
++        res.json({ ok: false, message: `专辑 ${totalCount} 首歌曲均无版权可播放（区域/版权限制）` });
++        return;
++      }
++
++      const queue = bot.getQueueManager();
++      queue.clear();
++      for (const song of queueable) {
++        queue.add({ ...song, platform: provider.platform });
++      }
++
++      const mode = queue.getMode();
++      let first;
++      if (mode === "random" || mode === "rloop") {
++        const idx = Math.floor(Math.random() * queue.size());
++        first = queue.playAt(idx);
++      } else {
++        first = queue.play();
++      }
++
++      let started = first ? await bot.resolveAndPlay(first) : false;
++      if (first && !started) {
++        started = await bot.playNext(20);
++      }
++
++      const playing = queue.current();
++      const loadedMsg = queueable.length < totalCount
++        ? `已加载 ${queueable.length}/${totalCount} 首（其余区域/版权限制）`
++        : `已加载 ${queueable.length} 首`;
++      if (started && playing) {
++        res.json({ ok: true, message: `${loadedMsg}，正在播放：${playing.name}` });
++      } else {
++        res.json({ ok: false, message: `${loadedMsg}，但无法开始播放。` });
++      }
++    } catch (err) {
++      logger.error({ err }, "play-album failed");
++      res.status(500).json({ error: (err as Error).message });
++    }
++  });
++
+   // Play a single song by ID — resolves URL on demand
+   router.post("/:botId/play-song", async (req, res) => {
+     try {
+diff --git a/web/src/router/index.ts b/web/src/router/index.ts
+index cc060f5..d62afcd 100644
+--- a/web/src/router/index.ts
++++ b/web/src/router/index.ts
+@@ -22,6 +22,13 @@ const router = createRouter({
+       path: '/playlist/:id',
+       name: 'playlist',
+       component: () => import('../views/Playlist.vue'),
++      meta: { kind: 'playlist' },
++    },
++    {
++      path: '/album/:id',
++      name: 'album',
++      component: () => import('../views/Playlist.vue'),
++      meta: { kind: 'album' },
+     },
+     {
+       path: '/lyrics',
+diff --git a/web/src/stores/player.ts b/web/src/stores/player.ts
+index 9bc817d..083262c 100644
+--- a/web/src/stores/player.ts
++++ b/web/src/stores/player.ts
+@@ -322,6 +322,16 @@ export const usePlayerStore = defineStore('player', {
+       this._syncAfterAction();
+     },
+ 
++    async playAlbum(albumId: string, platform = 'netease') {
++      if (!this.activeBotId) return;
++      const res = await axios.post(`/api/player/${this.activeBotId}/play-album`, { albumId, platform });
++      if (res.data?.message) {
++        this.notify(res.data.message, res.data.ok === false ? 'error' : 'info');
++      }
++      this._setTiming(this.activeBotId, { serverElapsed: 0 });
++      this._syncAfterAction();
++    },
++
+     async pause() {
+       if (!this.activeBotId) return;
+       // Freeze elapsed at current interpolated value
+diff --git a/web/src/views/Playlist.vue b/web/src/views/Playlist.vue
+index d5c9f8d..d00101d 100644
+--- a/web/src/views/Playlist.vue
++++ b/web/src/views/Playlist.vue
+@@ -38,7 +38,7 @@
+       </div>
+     </template>
+ 
+-    <div v-else class="loading">歌单不存在或加载失败</div>
++    <div v-else class="loading">{{ kind === 'album' ? '专辑' : '歌单' }}不存在或加载失败</div>
+   </div>
+ </template>
+ 
+@@ -64,6 +64,8 @@ interface PlaylistDetail {
+   songCount: number;
+ }
+ 
++const kind = (route.meta.kind as string) ?? 'playlist'; // 'playlist' | 'album'
++
+ const playlist = ref<PlaylistDetail | null>(null);
+ const songs = ref<Song[]>([]);
+ const loading = ref(true);
+@@ -71,20 +73,32 @@ const loading = ref(true);
+ async function playAll() {
+   const id = route.params.id as string;
+   const platform = (route.query.platform as string) || 'netease';
+-  await store.playPlaylist(id, platform);
++  if (kind === 'album') {
++    await store.playAlbum(id, platform);
++  } else {
++    await store.playPlaylist(id, platform);
++  }
+ }
+ 
+ onMounted(async () => {
+   const id = route.params.id as string;
+   const platform = (route.query.platform as string) || 'netease';
+ 
++  const detailUrl = kind === 'album'
++    ? `/api/music/album/${id}/detail`
++    : `/api/music/playlist/${id}/detail`;
++  const songsUrl = kind === 'album'
++    ? `/api/music/album/${id}`
++    : `/api/music/playlist/${id}`;
++
+   // allSettled, not Promise.all — if detail 404s but songs is fine
+   // (e.g., a QQ playlist whose detail endpoint flaked but the song
+   // list resolved), we still want to show the songs rather than
+-  // the "歌单不存在" empty state.
++  // the "不存在" empty state. For albums, detail always 404s — that
++  // is intentional; the fallback stub below handles it.
+   const [detailRes, songsRes] = await Promise.allSettled([
+-    axios.get(`/api/music/playlist/${id}/detail`, { params: { platform } }),
+-    axios.get(`/api/music/playlist/${id}`, { params: { platform } }),
++    axios.get(detailUrl, { params: { platform } }),
++    axios.get(songsUrl, { params: { platform } }),
+   ]);
+ 
+   const detail = detailRes.status === 'fulfilled' ? detailRes.value.data?.playlist : null;
+@@ -96,7 +110,7 @@ onMounted(async () => {
+     // Fall back to a stub built from the route + first song's cover.
+     playlist.value = {
+       id,
+-      name: '歌单',
++      name: kind === 'album' ? '专辑' : '歌单',
+       description: '',
+       coverUrl: songList[0]?.coverUrl ?? '',
+       songCount: songList.length,
+@@ -104,7 +118,7 @@ onMounted(async () => {
+   } else {
+     playlist.value = null;
+     if (detailRes.status === 'rejected') {
+-      console.error('Failed to load playlist detail:', (detailRes.reason as any)?.response?.status, (detailRes.reason as any)?.message);
++      console.error('Failed to load detail:', (detailRes.reason as any)?.response?.status, (detailRes.reason as any)?.message);
+     }
+   }
+   songs.value = songList;
+diff --git a/web/src/views/Search.vue b/web/src/views/Search.vue
+index 0536e36..8822277 100644
+--- a/web/src/views/Search.vue
++++ b/web/src/views/Search.vue
+@@ -20,22 +20,54 @@
+ 
+     <div v-if="loading" class="loading">搜索中...</div>
+ 
+-    <div v-else-if="results.length > 0" class="results">
+-      <SongCard
+-        v-for="(song, i) in results"
+-        :key="`${song.platform}-${song.id}`"
+-        :song="song"
+-        :index="i + 1"
+-        :active="store.currentSong?.id === song.id"
+-        @play="store.playSong(song)"
+-        @playNext="store.playNextSong(song)"
+-        @add="store.addSong(song)"
+-      />
+-    </div>
++    <template v-else-if="songs.length || albums.length || playlists.length">
++      <section v-if="albums.length" class="result-section">
++        <h2 class="section-title">专辑</h2>
++        <div class="card-grid">
++          <router-link
++            v-for="al in albums"
++            :key="`${al.platform}-${al.id}`"
++            :to="`/album/${al.id}?platform=${al.platform}`"
++            class="card hover-scale"
++          >
++            <CoverArt :url="al.coverUrl" :size="160" :radius="10" :show-shadow="true" />
++            <div class="card-name">{{ al.name }}</div>
++            <div class="card-sub">{{ al.artist }}</div>
++          </router-link>
++        </div>
++      </section>
++
++      <section v-if="playlists.length" class="result-section">
++        <h2 class="section-title">歌单</h2>
++        <div class="card-grid">
++          <router-link
++            v-for="pl in playlists"
++            :key="`${pl.platform}-${pl.id}`"
++            :to="`/playlist/${pl.id}?platform=${pl.platform}`"
++            class="card hover-scale"
++          >
++            <CoverArt :url="pl.coverUrl" :size="160" :radius="10" :show-shadow="true" />
++            <div class="card-name">{{ pl.name }}</div>
++          </router-link>
++        </div>
++      </section>
++
++      <section v-if="songs.length" class="result-section">
++        <h2 class="section-title">单曲</h2>
++        <SongCard
++          v-for="(song, i) in songs"
++          :key="`${song.platform}-${song.id}`"
++          :song="song"
++          :index="i + 1"
++          :active="store.currentSong?.id === song.id"
++          @play="store.playSong(song)"
++          @playNext="store.playNextSong(song)"
++          @add="store.addSong(song)"
++        />
++      </section>
++    </template>
+ 
+-    <div v-else-if="searched" class="empty">
+-      未找到相关结果
+-    </div>
++    <div v-else-if="searched" class="empty">未找到相关结果</div>
+   </div>
+ </template>
+ 
+@@ -45,15 +77,21 @@ import { useRoute } from 'vue-router';
+ import { Icon } from '@iconify/vue';
+ import axios from 'axios';
+ import { usePlayerStore } from '../stores/player.js';
++import type { Song } from '../stores/player.js';
+ import SongCard from '../components/SongCard.vue';
++import CoverArt from '../components/CoverArt.vue';
+ 
+ const store = usePlayerStore();
+ const route = useRoute();
+ 
+ const query = ref((route.query.q as string) || '');
+-import { Song } from '../stores/player.js';
+ 
+-const results = ref<Song[]>([]);
++interface Album { id: string; name: string; artist: string; coverUrl: string; songCount?: number; platform: string; }
++interface Playlist { id: string; name: string; coverUrl: string; songCount?: number; platform: string; }
++
++const songs = ref<Song[]>([]);
++const albums = ref<Album[]>([]);
++const playlists = ref<Playlist[]>([]);
+ const loading = ref(false);
+ const searched = ref(false);
+ 
+@@ -62,12 +100,12 @@ async function doSearch() {
+   loading.value = true;
+   searched.value = true;
+   try {
+-    const res = await axios.get('/api/music/search/all', {
+-      params: { q: query.value },
+-    });
+-    results.value = res.data.songs;
++    const res = await axios.get('/api/music/search/all', { params: { q: query.value } });
++    songs.value = res.data.songs ?? [];
++    albums.value = res.data.albums ?? [];
++    playlists.value = res.data.playlists ?? [];
+   } catch {
+-    results.value = [];
++    songs.value = []; albums.value = []; playlists.value = [];
+   } finally {
+     loading.value = false;
+   }
+@@ -141,4 +179,23 @@ onMounted(() => {
+   flex-direction: column;
+   gap: 2px;
+ }
++
++.result-section {
++  margin-bottom: 32px;
++  .section-title { font-size: 18px; margin: 0 0 12px; opacity: 0.85; }
++}
++.card-grid {
++  display: grid;
++  grid-template-columns: repeat(auto-fill, minmax(140px, 1fr));
++  gap: 16px;
++}
++.card {
++  display: flex;
++  flex-direction: column;
++  gap: 6px;
++  text-decoration: none;
++  color: inherit;
++  .card-name { font-size: 14px; line-height: 1.3; max-height: 2.6em; overflow: hidden; }
++  .card-sub  { font-size: 12px; opacity: 0.6; }
++}
+ </style>

scripts/smoke_issue51.sh

@@ -0,0 +1,96 @@
+#!/usr/bin/env bash
+# Smoke test for issue #51 — run AFTER you start the bot from temp/preview-merge
+# (or from main once both PRs are merged).
+#
+# Usage: ./scripts/smoke_issue51.sh [HOST]
+# Default HOST is http://127.0.0.1:3000
+
+set -e
+HOST="${1:-http://127.0.0.1:3000}"
+PASS=0
+FAIL=0
+note() { echo -e "\n=== $* ==="; }
+ok()   { echo "  [PASS] $*"; PASS=$((PASS+1)); }
+bad()  { echo "  [FAIL] $*"; FAIL=$((FAIL+1)); }
+
+# ---- Album search ----------------------------------------------------------
+note "1. /api/music/search/all returns {songs,albums,playlists}"
+
+RES=$(curl.exe -s "$HOST/api/music/search/all?q=%E5%91%A8%E6%9D%B0%E4%BC%A6") # 周杰伦
+KEYS=$(echo "$RES" | python3 -c "import json,sys;d=json.load(sys.stdin);print(','.join(sorted(d.keys())))")
+if [ "$KEYS" = "albums,playlists,songs" ]; then ok "keys = $KEYS"; else bad "keys = $KEYS (expected albums,playlists,songs)"; fi
+
+NA=$(echo "$RES" | python3 -c "import json,sys;d=json.load(sys.stdin);print(len(d.get('albums',[])))")
+NS=$(echo "$RES" | python3 -c "import json,sys;d=json.load(sys.stdin);print(len(d.get('songs',[])))")
+NP=$(echo "$RES" | python3 -c "import json,sys;d=json.load(sys.stdin);print(len(d.get('playlists',[])))")
+echo "  songs=$NS, albums=$NA, playlists=$NP"
+if [ "$NA" -gt 0 ]; then ok "albums populated"; else bad "albums empty (expected >0 for 周杰伦)"; fi
+if [ "$NS" -gt 0 ]; then ok "songs populated"; fi
+
+# ---- Album detail playback path -------------------------------------------
+note "2. /api/music/album/:id returns songs"
+
+if [ "$NA" -gt 0 ]; then
+  ALBUM_ID=$(echo "$RES" | python3 -c "import json,sys;d=json.load(sys.stdin);a=d['albums'][0];print(a['id'])")
+  PLATFORM=$(echo "$RES" | python3 -c "import json,sys;d=json.load(sys.stdin);a=d['albums'][0];print(a['platform'])")
+  echo "  testing album id=$ALBUM_ID platform=$PLATFORM"
+  ASONGS=$(curl.exe -s "$HOST/api/music/album/$ALBUM_ID?platform=$PLATFORM" | python3 -c "import json,sys;d=json.load(sys.stdin);print(len(d.get('songs',[])))" 2>/dev/null || echo 0)
+  if [ "$ASONGS" -gt 0 ]; then ok "album returned $ASONGS songs"; else bad "album endpoint returned 0 songs"; fi
+else
+  echo "  (skipped — no albums to test)"
+fi
+
+# ---- Avatar API ------------------------------------------------------------
+note "3. avatar GET 404 on bot with no avatar"
+
+BOT_ID=$(curl.exe -s "$HOST/api/bot" | python3 -c "import json,sys;d=json.load(sys.stdin);bots=d.get('bots',[]);print(bots[0]['id'] if bots else '')")
+if [ -z "$BOT_ID" ]; then bad "no bot found — create a bot first"; exit 1; fi
+echo "  using bot $BOT_ID"
+
+curl.exe -s -o /dev/null -w "%{http_code}" "$HOST/api/bot/$BOT_ID/avatar" > /tmp/code
+CODE=$(cat /tmp/code)
+if [ "$CODE" = "404" ] || [ "$CODE" = "200" ]; then ok "GET initial state = $CODE"; else bad "unexpected GET status $CODE"; fi
+
+note "4. avatar PUT 200 + GET 200 round-trip"
+# 1×1 transparent PNG (67 bytes)
+TINY_PNG_B64="iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNkYAAAAAYAAjCB0C8AAAAASUVORK5CYII="
+PUT_RES=$(curl.exe -s -X PUT "$HOST/api/bot/$BOT_ID/avatar" -H "Content-Type: application/json" \
+  -d "{\"dataUrl\":\"data:image/png;base64,$TINY_PNG_B64\"}")
+echo "  PUT response: $PUT_RES"
+GOT_PATH=$(echo "$PUT_RES" | python3 -c "import json,sys;d=json.load(sys.stdin);print(d.get('path',''))" 2>/dev/null || echo "")
+if [ "$GOT_PATH" = "$BOT_ID.png" ]; then ok "PUT returned path=$GOT_PATH"; else bad "PUT path = $GOT_PATH (expected $BOT_ID.png)"; fi
+
+curl.exe -s -o /tmp/avatar_check.png -w "%{http_code}" "$HOST/api/bot/$BOT_ID/avatar" > /tmp/code
+CODE=$(cat /tmp/code)
+SIZE=$(wc -c < /tmp/avatar_check.png)
+if [ "$CODE" = "200" ] && [ "$SIZE" -gt 60 ]; then ok "GET returned 200, $SIZE bytes"; else bad "GET status=$CODE size=$SIZE"; fi
+
+note "5. avatar DELETE 204 + GET 404"
+curl.exe -s -X DELETE "$HOST/api/bot/$BOT_ID/avatar" -o /dev/null -w "%{http_code}" > /tmp/code
+CODE=$(cat /tmp/code)
+if [ "$CODE" = "204" ]; then ok "DELETE returned 204"; else bad "DELETE status = $CODE"; fi
+
+curl.exe -s -o /dev/null -w "%{http_code}" "$HOST/api/bot/$BOT_ID/avatar" > /tmp/code
+CODE=$(cat /tmp/code)
+if [ "$CODE" = "404" ]; then ok "GET after DELETE returned 404"; else bad "GET after DELETE = $CODE"; fi
+
+note "6. avatar PUT rejects oversize (>200KB)"
+BIG_B64=$(node -e "console.log(Buffer.alloc(210*1024,7).toString('base64'))")
+curl.exe -s -o /dev/null -w "%{http_code}" -X PUT "$HOST/api/bot/$BOT_ID/avatar" \
+  -H "Content-Type: application/json" -d "{\"dataUrl\":\"data:image/png;base64,$BIG_B64\"}" > /tmp/code
+CODE=$(cat /tmp/code)
+if [ "$CODE" = "413" ]; then ok "oversize rejected with 413"; else bad "oversize status = $CODE (expected 413)"; fi
+
+note "7. avatar PUT rejects bad MIME (image/gif)"
+GIF_B64="R0lGODlhAQABAAAAACw="  # tiny gif
+curl.exe -s -o /dev/null -w "%{http_code}" -X PUT "$HOST/api/bot/$BOT_ID/avatar" \
+  -H "Content-Type: application/json" -d "{\"dataUrl\":\"data:image/gif;base64,$GIF_B64\"}" > /tmp/code
+CODE=$(cat /tmp/code)
+if [ "$CODE" = "400" ]; then ok "bad MIME rejected with 400"; else bad "bad MIME status = $CODE (expected 400)"; fi
+
+# ---------------------------------------------------------------------------
+echo ""
+echo "============================================="
+echo "SMOKE RESULT: $PASS passed, $FAIL failed"
+echo "============================================="
+[ "$FAIL" -eq 0 ]

src/bot/instance.ts

@@ -295,7 +295,13 @@ export class BotInstance extends EventEmitter {
       const relPath = this.database.getCustomAvatarPath(this.id);
       if (relPath) {
         const buf = this.avatarStore.read(relPath);
-        if (buf) this.profileManager.setCustomAvatar(buf);
+        // loadCustomAvatar, NOT setCustomAvatar (#148): we are still in the
+        // constructor, so tsClient has not connected. setCustomAvatar would
+        // start a file transfer right here and fail. profileManager.onConnect()
+        // uploads it for real once the handshake completes.
+        // `length > 0` because avatarStore.write is delete-then-write, so a
+        // crash mid-write leaves a 0-byte file that is truthy as a Buffer.
+        if (buf && buf.length > 0) this.profileManager.loadCustomAvatar(buf);
       }
     } catch (err) {
       this.logger.warn({ err }, "Failed to load custom avatar — skipping");

src/bot/profile.test.ts

@@ -145,3 +145,60 @@ describe("BotProfileManager custom avatar precedence", () => {
     expect(ts.clearCalls).toBe(0);
   });
 });
+
+// #148: the persisted avatar is loaded in the BotInstance constructor, before
+// tsClient.connect() has run. Loading it must not touch the wire at all.
+describe("BotProfileManager loadCustomAvatar (pre-connect load, #148)", () => {
+  let ts: ReturnType<typeof makeMockTs>;
+  beforeEach(() => { ts = makeMockTs(); });
+
+  it("does not upload or clear anything when called before connect", async () => {
+    const pm = new BotProfileManager(ts as any, noopLogger, cfgOn, "Bot");
+    pm.loadCustomAvatar(Buffer.from([7, 7, 7]));
+    await flush();
+    expect(ts.uploadCalls.length).toBe(0);
+    expect(ts.clearCalls).toBe(0);
+    expect(ts.fileTransferInitUpload).not.toHaveBeenCalled();
+  });
+
+  it("the loaded avatar is uploaded once onConnect fires", async () => {
+    const pm = new BotProfileManager(ts as any, noopLogger, cfgOn, "Bot");
+    pm.loadCustomAvatar(Buffer.from([7, 7, 7]));
+    await flush();
+    pm.onConnect();
+    await flush();
+    expect(ts.uploadCalls.length).toBe(1);
+    expect(ts.uploadCalls[0].equals(Buffer.from([7, 7, 7]))).toBe(true);
+  });
+
+  it("survives a reconnect: onConnect re-applies the loaded avatar every time", async () => {
+    const pm = new BotProfileManager(ts as any, noopLogger, cfgOn, "Bot");
+    pm.loadCustomAvatar(Buffer.from([8]));
+    pm.onConnect();
+    await flush();
+    pm.onConnect();
+    await flush();
+    expect(ts.uploadCalls.length).toBe(2);
+  });
+
+  it("loading null leaves the wire untouched and onConnect stays quiet", async () => {
+    const pm = new BotProfileManager(ts as any, noopLogger, cfgOn, "Bot");
+    pm.loadCustomAvatar(null);
+    pm.onConnect();
+    await flush();
+    expect(ts.uploadCalls.length).toBe(0);
+    expect(ts.clearCalls).toBe(0);
+  });
+
+  it("setCustomAvatar still uploads immediately after connect (post-connect edit unchanged)", async () => {
+    const pm = new BotProfileManager(ts as any, noopLogger, cfgOn, "Bot");
+    pm.loadCustomAvatar(Buffer.from([1]));
+    pm.onConnect();
+    await flush();
+    ts.uploadCalls.length = 0;
+    pm.setCustomAvatar(Buffer.from([2, 2]));
+    await flush();
+    expect(ts.uploadCalls.length).toBe(1);
+    expect(ts.uploadCalls[0].equals(Buffer.from([2, 2]))).toBe(true);
+  });
+});

src/bot/profile.ts

@@ -65,6 +65,20 @@ export class BotProfileManager {
 
   // --- Public API ---
 
+  /**
+   * Store a persisted custom avatar WITHOUT touching TeamSpeak (#148).
+   *
+   * Used during BotInstance construction, when the TS connection does not
+   * exist yet: setCustomAvatar would immediately fire the three-step file
+   * transfer (fileTransferInitUpload → uploadFileData → clientupdate) against
+   * a client that has not connected, so the upload always failed and the
+   * saved avatar never appeared. onConnect() re-applies this.customAvatar
+   * once the handshake completes, so loading it silently here loses nothing.
+   */
+  loadCustomAvatar(buffer: Buffer | null): void {
+    this.customAvatar = buffer;
+  }
+
   /**
    * Set/clear the persistent idle avatar. Pass null to remove.
    *

src/music/local-extract-fallback.test.ts

@@ -0,0 +1,109 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from "vitest";
+// unlinkSync/rmdirSync are NOT mocked below, so the test's own fixture
+// teardown is unaffected by the simulated lock on *.mp4.
+import { mkdtempSync, statSync, existsSync, readFileSync, unlinkSync, readdirSync, rmdirSync } from "node:fs";
+import { spawnSync } from "node:child_process";
+import { createRequire } from "node:module";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+/**
+ * #149: when the audio track is extracted successfully but the source video
+ * cannot be deleted (Windows keeps files locked briefly — rmSync with
+ * force:true still throws EBUSY/EPERM), the record must fall back to the
+ * ORIGINAL container completely: both the path AND the recorded size.
+ *
+ * Committing the size before the delete succeeded would leave the record
+ * claiming the small extracted size while still holding the whole video, so
+ * totalBytes() under-counts and the upload directory grows past its quota.
+ *
+ * This lives in its own file because it partially mocks node:fs, which would
+ * otherwise leak into every other test in local.test.ts.
+ */
+vi.mock("node:fs", async (importOriginal) => {
+  const actual = await importOriginal<typeof import("node:fs")>();
+  return {
+    ...actual,
+    default: actual,
+    rmSync: (path: string, opts?: object) => {
+      // Simulate the lock on the source video only; every other delete
+      // (the discarded .mka, temp dirs, the reject path) behaves normally.
+      if (typeof path === "string" && path.endsWith(".mp4")) {
+        const err = new Error("EBUSY: resource busy or locked") as NodeJS.ErrnoException;
+        err.code = "EBUSY";
+        throw err;
+      }
+      return actual.rmSync(path, opts as never);
+    },
+  };
+});
+
+const { LocalMusicProvider } = await import("./local.js");
+
+const ffmpeg: string | null = (() => {
+  try {
+    return createRequire(import.meta.url)("ffmpeg-static") as string;
+  } catch {
+    return null;
+  }
+})();
+const have = !!ffmpeg && spawnSync(ffmpeg, ["-version"], { stdio: "ignore" }).status === 0;
+
+let dir: string;
+beforeEach(() => { dir = mkdtempSync(join(tmpdir(), "local-extract-fallback-")); });
+afterEach(() => {
+  // Recursive teardown without rmSync (mocked above for *.mp4).
+  for (const f of readdirSync(dir)) {
+    try { unlinkSync(join(dir, f)); } catch { /* best effort */ }
+  }
+  try { rmdirSync(dir); } catch { /* best effort */ }
+});
+
+describe("LocalMusicProvider: source video cannot be deleted after extraction (#149)", () => {
+  it.runIf(have)("keeps the original container AND its real size, not the extracted size", async () => {
+    const src = join(dir, "fixture.mp4");
+    const r = spawnSync(ffmpeg!, [
+      "-y", "-hide_banner", "-loglevel", "error",
+      "-f", "lavfi", "-i", "testsrc=s=320x240:r=25:d=3",
+      "-f", "lavfi", "-i", "sine=f=440:d=3",
+      "-c:v", "libx264", "-b:v", "800k", "-c:a", "aac", "-shortest", src,
+    ], { stdio: "ignore" });
+    expect(r.status).toBe(0);
+
+    const bytes = readFileSync(src);
+    unlinkSync(src); // uploadAudio writes its own copy under a uuid name
+
+    const p = new LocalMusicProvider(dir);
+    const song = await p.uploadAudio({
+      buffer: bytes, originalName: "fixture.mp4", mimeType: "video/mp4",
+    });
+
+    const resolved = await p.getSongUrl(song.id);
+    expect(resolved).not.toBeNull();
+
+    // Fell back to the original container — the extract was discarded.
+    expect(resolved!.url.endsWith(".mp4")).toBe(true);
+    expect(existsSync(resolved!.url)).toBe(true);
+    expect(existsSync(resolved!.url.replace(/\.mp4$/, ".m4a"))).toBe(false);
+    expect(existsSync(resolved!.url.replace(/\.mp4$/, ".mka"))).toBe(false);
+
+    const onDisk = statSync(resolved!.url).size;
+    expect(onDisk).toBe(bytes.length);
+
+    // The RECORDED size drives the quota (totalBytes()), so it must describe
+    // the file actually retained. It is not exposed through search()/toSong,
+    // but it is persisted to index.json — read it back from there.
+    const record = (JSON.parse(readFileSync(join(dir, "index.json"), "utf8")) as Array<{
+      id: string; size: number; filePath: string;
+    }>).find((r) => r.id === song.id);
+    expect(record).toBeDefined();
+    expect(record!.filePath.endsWith(".mp4")).toBe(true);
+    // Before the fix this was the (much smaller) .mka size while the whole
+    // .mp4 stayed on disk, so the quota under-counted the retained bytes.
+    expect(record!.size).toBe(bytes.length);
+
+    // Sanity: the extract really is much smaller, so a wrong commit order
+    // would have been clearly observable rather than a rounding error.
+    expect(onDisk).toBeGreaterThan(50_000);
+  }, 60000);
+});

src/music/local.test.ts

@@ -1,8 +1,11 @@
 import { describe, it, expect, beforeEach, afterEach } from "vitest";
-import { mkdtempSync, rmSync, existsSync, writeFileSync } from "node:fs";
+import { mkdtempSync, rmSync, existsSync, writeFileSync, readFileSync, readdirSync, statSync } from "node:fs";
+import { spawnSync } from "node:child_process";
+import { createRequire } from "node:module";
+import { buildFfmpegArgs } from "../audio/player.js";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
-import { LocalMusicProvider } from "./local.js";
+import { LocalMusicProvider, parseMediaProbe } from "./local.js";
 
 let dir: string;
 
@@ -149,6 +152,115 @@ describe("LocalMusicProvider upload validation", () => {
       p.uploadAudio({ buffer: Buffer.alloc(0), originalName: "a.mp3" }),
     ).rejects.toThrow();
   });
+
+  // #149: video containers are accepted; only their audio track is kept.
+  it("still rejects a non-media extension after video was allowed", async () => {
+    const p = new LocalMusicProvider(dir);
+    for (const name of ["evil.exe", "evil.html", "evil.mp4.txt", "notes.pdf"]) {
+      await expect(
+        p.uploadAudio({ buffer: Buffer.from("x"), originalName: name, mimeType: "video/mp4" }),
+      ).rejects.toThrow();
+    }
+  });
+
+  it("accepts every supported video extension at the extension gate", async () => {
+    const p = new LocalMusicProvider(dir);
+    // Junk content: ffmpeg cannot open it, so it is "unrecognised" rather than
+    // "no audio track" and must be accepted exactly like a truncated .mp3
+    // always has been. The extension allowlist is what is under test here.
+    // .m4v is excluded on purpose — see the next test.
+    for (const ext of [".mp4", ".mov", ".avi", ".mkv", ".flv", ".wmv", ".mpg", ".mpeg", ".3gp", ".ts", ".m2ts", ".ogv"]) {
+      const song = await p.uploadAudio({
+        buffer: Buffer.from("not really a video"),
+        originalName: `clip${ext}`,
+        mimeType: "video/mp4",
+      });
+      expect(song.platform).toBe("local");
+      expect(song.name).toBe("clip");
+    }
+  });
+
+  it("refuses a .m4v raw video elementary stream, which by definition has no audio", async () => {
+    // .m4v is not a container — ffmpeg's rawvideo demuxer opens arbitrary
+    // bytes as an MPEG-4 video elementary stream, so it IS recognised and
+    // genuinely carries no audio track. Refusing it is the correct outcome,
+    // and it is the one case that distinguishes `recognized` from `probed`.
+    const p = new LocalMusicProvider(dir);
+    await expect(
+      p.uploadAudio({
+        buffer: Buffer.from("not really a video"),
+        originalName: "clip.m4v",
+        mimeType: "video/x-m4v",
+      }),
+    ).rejects.toThrow(/音轨/);
+  });
+
+  it("the error message names both audio and video formats", async () => {
+    const p = new LocalMusicProvider(dir);
+    await expect(
+      p.uploadAudio({ buffer: Buffer.from("x"), originalName: "a.exe" }),
+    ).rejects.toThrow(/视频/);
+  });
+});
+
+describe("parseMediaProbe (#149)", () => {
+  const mp4Banner = `Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'clip.mp4':
+  Duration: 00:03:27.15, start: 0.000000, bitrate: 1105 kb/s
+  Stream #0:0[0x1](und): Video: h264 (High), yuv420p, 1280x720, 30 fps
+  Stream #0:1[0x2](und): Audio: aac (LC), 48000 Hz, stereo, fltp, 192 kb/s`;
+
+  it("reads duration and detects the audio stream in a video container", () => {
+    const r = parseMediaProbe(mp4Banner);
+    expect(r.durationSeconds).toBe(3 * 60 + 27);
+    expect(r.hasAudio).toBe(true);
+  });
+
+  it("reports hasAudio false for a video with only a video stream", () => {
+    const silent = `Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'silent.mp4':
+  Duration: 00:00:02.00, start: 0.000000, bitrate: 29 kb/s
+  Stream #0:0[0x1](und): Video: h264 (High 4:4:4 Predictive), yuv444p, 160x120, 10 fps`;
+    const r = parseMediaProbe(silent);
+    expect(r.durationSeconds).toBe(2);
+    expect(r.hasAudio).toBe(false);
+  });
+
+  it("detects a plain audio file", () => {
+    const r = parseMediaProbe(`Input #0, mp3, from 'a.mp3':
+  Duration: 00:00:30.02, start: 0.000000, bitrate: 128 kb/s
+  Stream #0:0: Audio: mp3, 44100 Hz, stereo, fltp, 128 kb/s`);
+    expect(r.durationSeconds).toBe(30);
+    expect(r.hasAudio).toBe(true);
+  });
+
+  it("does not mistake an attached cover image for an audio stream", () => {
+    const r = parseMediaProbe(`Input #0, mp3, from 'cover.mp3':
+  Duration: 00:00:10.00, start: 0.000000, bitrate: 130 kb/s
+  Stream #0:0: Audio: mp3, 44100 Hz, stereo, fltp, 128 kb/s
+  Stream #0:1: Video: mjpeg (Baseline), yuvj420p(pc), 100x100 [attached pic]`);
+    expect(r.hasAudio).toBe(true);
+  });
+
+  it("returns zeros on unparseable output rather than throwing", () => {
+    const r = parseMediaProbe("ffmpeg: command exploded");
+    expect(r.durationSeconds).toBe(0);
+    expect(r.hasAudio).toBe(false);
+    expect(r.recognized).toBe(false);
+  });
+
+  // The distinction that decides whether an upload is refused: ffmpeg opened
+  // the file and found no audio (refuse) vs ffmpeg could not open it at all
+  // (accept, as it always has for truncated audio).
+  it("marks a readable container recognized and unreadable bytes not", () => {
+    expect(parseMediaProbe(mp4Banner).recognized).toBe(true);
+    expect(parseMediaProbe(`[mov,mp4,m4a,3gp,3g2,mj2 @ 0x1] moov atom not found
+[in#0 @ 0x2] Error opening input: Invalid data found when processing input
+Error opening input file junk.mp4.`).recognized).toBe(false);
+  });
+
+  it("rounds fractional durations", () => {
+    expect(parseMediaProbe("Duration: 00:00:03.60,").durationSeconds).toBe(4);
+    expect(parseMediaProbe("Duration: 01:02:03.10,").durationSeconds).toBe(3723);
+  });
 });
 
 describe("LocalMusicProvider quota", () => {
@@ -233,3 +345,146 @@ describe("LocalMusicProvider filename handling", () => {
     expect(await p.getSongUrl(song.id)).not.toBeNull();
   });
 });
+
+// #149 end-to-end: build real containers with the bundled ffmpeg and push
+// them through the actual upload path. Skipped automatically if the binary is
+// unavailable, so the suite still runs on a machine without it.
+describe("LocalMusicProvider video upload, end to end (#149)", () => {
+  const ffmpeg: string | null = (() => {
+    try {
+      return createRequire(import.meta.url)("ffmpeg-static") as string;
+    } catch {
+      return null;
+    }
+  })();
+
+  const have = !!ffmpeg && spawnSync(ffmpeg, ["-version"], { stdio: "ignore" }).status === 0;
+
+  /** Render a real container into the temp dir and return its bytes. */
+  function render(name: string, args: string[]): Buffer {
+    const out = join(dir, name);
+    const r = spawnSync(ffmpeg!, ["-y", "-hide_banner", "-loglevel", "error", ...args, out], {
+      stdio: "ignore",
+    });
+    if (r.status !== 0) throw new Error(`fixture render failed: ${name}`);
+    const buf = readFileSync(out);
+    rmSync(out, { force: true }); // upload writes its own copy
+    return buf;
+  }
+
+  const withAudio = (dur: number, vcodec: string, acodec: string) => [
+    "-f", "lavfi", "-i", `testsrc=s=160x120:r=10:d=${dur}`,
+    "-f", "lavfi", "-i", `sine=f=440:d=${dur}`,
+    "-c:v", vcodec, "-c:a", acodec, "-shortest",
+  ];
+
+  it.runIf(have)("accepts an mp4, reads its duration, and keeps only the audio", async () => {
+    const p = new LocalMusicProvider(dir);
+    const mp4 = render("src.mp4", withAudio(3, "libx264", "aac"));
+    const song = await p.uploadAudio({
+      buffer: mp4, originalName: "My Clip.mp4", mimeType: "video/mp4",
+    });
+
+    expect(song.name).toBe("My Clip");
+    expect(song.platform).toBe("local");
+    expect(song.duration).toBe(3);
+
+    const resolved = await p.getSongUrl(song.id);
+    expect(resolved).not.toBeNull();
+    // The video container is gone; what remains is the extracted audio track.
+    // AAC (what libx264+aac mp4s carry) goes to .m4a so the encoder-priming
+    // edit list survives — see extractedAudioExt.
+    expect(resolved!.url.endsWith(".m4a")).toBe(true);
+    expect(existsSync(join(dir, `${song.id}.mp4`))).toBe(false);
+    expect(existsSync(resolved!.url)).toBe(true);
+    expect(statSync(resolved!.url).size).toBeGreaterThan(0);
+    expect(statSync(resolved!.url).size).toBeLessThan(mp4.length);
+  }, 60000);
+
+  it.runIf(have)("extracted audio is still decodable by the player's ffmpeg args", async () => {
+    const p = new LocalMusicProvider(dir);
+    const song = await p.uploadAudio({
+      buffer: render("src2.mp4", withAudio(2, "libx264", "aac")),
+      originalName: "clip.mp4",
+      mimeType: "video/mp4",
+    });
+    const url = (await p.getSongUrl(song.id))!.url;
+
+    const decoded = spawnSync(
+      ffmpeg!,
+      [...buildFfmpegArgs(url, 0).slice(0, -1), "-"],
+      { maxBuffer: 64 * 1024 * 1024 },
+    );
+    expect(decoded.status).toBe(0);
+    // 2s of 48 kHz stereo s16le ≈ 384000 bytes; allow codec priming slack.
+    expect(decoded.stdout.length).toBeGreaterThan(300000);
+  }, 60000);
+
+  it.runIf(have)("aac extraction decodes bit-for-bit identically to the audio inside the video", async () => {
+    // The strongest statement of "lossless": decode the audio track straight
+    // out of the source mp4, decode the stored extract, compare the PCM.
+    // A Matroska remux would NOT pass this — it loses the MP4 edit list that
+    // discards AAC encoder priming, so it decodes ~23 ms longer.
+    const p = new LocalMusicProvider(dir);
+    const bytes = render("bitexact.mp4", withAudio(4, "libx264", "aac"));
+    const sourceCopy = join(dir, "source-kept.mp4");
+    writeFileSync(sourceCopy, bytes);
+
+    const song = await p.uploadAudio({
+      buffer: bytes, originalName: "bitexact.mp4", mimeType: "video/mp4",
+    });
+    const url = (await p.getSongUrl(song.id))!.url;
+
+    const toPcm = (input: string, pre: string[] = []) => spawnSync(
+      ffmpeg!,
+      ["-hide_banner", "-loglevel", "error", "-i", input, ...pre,
+       "-f", "s16le", "-ar", "48000", "-ac", "2", "-acodec", "pcm_s16le", "-"],
+      { maxBuffer: 128 * 1024 * 1024 },
+    );
+
+    const fromVideo = toPcm(sourceCopy, ["-vn", "-map", "0:a:0"]);
+    const fromExtract = toPcm(url);
+    expect(fromVideo.status).toBe(0);
+    expect(fromExtract.status).toBe(0);
+    expect(fromExtract.stdout.length).toBe(fromVideo.stdout.length);
+    expect(fromExtract.stdout.equals(fromVideo.stdout)).toBe(true);
+  }, 90000);
+
+  it.runIf(have)("refuses a video that genuinely has no audio track", async () => {
+    const p = new LocalMusicProvider(dir);
+    const silent = render("silent.mp4", [
+      "-f", "lavfi", "-i", "testsrc=s=160x120:r=10:d=2", "-an",
+    ]);
+    await expect(
+      p.uploadAudio({ buffer: silent, originalName: "silent.mp4", mimeType: "video/mp4" }),
+    ).rejects.toThrow(/音轨/);
+    // The rejected upload must not leave its bytes behind.
+    expect(readdirSync(dir).filter((f) => f.endsWith(".mp4"))).toEqual([]);
+  }, 60000);
+
+  it.runIf(have)("extracts losslessly from avi/mkv/flv too, not just mp4", async () => {
+    const p = new LocalMusicProvider(dir);
+    const cases: Array<[string, string[]]> = [
+      ["a.avi", withAudio(2, "mpeg4", "libmp3lame")],
+      ["a.mkv", withAudio(2, "libx264", "libopus")],
+      ["a.flv", withAudio(2, "flv", "libmp3lame")],
+    ];
+    for (const [name, args] of cases) {
+      const song = await p.uploadAudio({
+        buffer: render(`src-${name}`, args), originalName: name, mimeType: "video/x-msvideo",
+      });
+      const url = (await p.getSongUrl(song.id))!.url;
+      expect(url.endsWith(".mka")).toBe(true);
+      expect(statSync(url).size).toBeGreaterThan(0);
+    }
+  }, 120000);
+
+  it.runIf(have)("a plain audio upload is untouched — no extraction, original extension kept", async () => {
+    const p = new LocalMusicProvider(dir);
+    const mp3 = render("src.mp3", ["-f", "lavfi", "-i", "sine=f=440:d=2", "-c:a", "libmp3lame"]);
+    const song = await p.uploadAudio({ buffer: mp3, originalName: "tune.mp3", mimeType: "audio/mpeg" });
+    const url = (await p.getSongUrl(song.id))!.url;
+    expect(url.endsWith(".mp3")).toBe(true);
+    expect(statSync(url).size).toBe(mp3.length); // byte-identical, not remuxed
+  }, 60000);
+});

src/music/local.ts

@@ -1,5 +1,5 @@
 import { spawn } from "node:child_process";
-import { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, readFileSync, rmSync, statSync, writeFileSync } from "node:fs";
 import { createRequire } from "node:module";
 import path from "node:path";
 import crypto from "node:crypto";
@@ -34,6 +34,56 @@ const AUDIO_EXTENSIONS = new Set([
   ".ape",
 ]);
 
+/** Video containers accepted for upload (#149). Only the audio track is ever
+ *  used — the bot has no video output. Playback would work straight from the
+ *  container (ffmpeg selects the audio stream), but we extract the audio on
+ *  upload so a 200 MB clip does not sit on disk for a 3 MB song; see
+ *  extractAudioTrack. `.webm` is deliberately absent: it is already in
+ *  AUDIO_EXTENSIONS and both audio-only and video .webm are handled there. */
+const VIDEO_EXTENSIONS = new Set([
+  ".mp4",
+  ".mov",
+  ".avi",
+  ".mkv",
+  ".flv",
+  ".wmv",
+  ".m4v",
+  ".mpg",
+  ".mpeg",
+  ".3gp",
+  ".ts",
+  ".m2ts",
+  ".ogv",
+]);
+
+/** Fallback container for an extracted audio track. Matroska takes
+ *  essentially any audio codec, so `-c:a copy` works without knowing what the
+ *  source used — no re-encode, no codec/extension table. */
+const EXTRACTED_AUDIO_EXT = ".mka";
+
+/**
+ * Container to remux an extracted track into, chosen by its codec.
+ *
+ * AAC gets .m4a rather than the Matroska fallback. MP4 stores the AAC encoder
+ * priming (the ~1000 warm-up samples every AAC encoder emits) in an edit list,
+ * and that edit list does NOT survive into Matroska — so an aac→.mka remux
+ * decodes ~23 ms longer than the source, with the priming samples audible at
+ * the head instead of discarded. Measured: −66 dBFS, i.e. inaudible, but the
+ * track is then fractionally out of step with its own reported duration for
+ * no reason. Copying aac into .m4a keeps the edit list and decodes
+ * byte-for-byte identical to the audio inside the original video.
+ *
+ * AAC is worth special-casing because it is what mp4 / mov / m4v — the
+ * formats people actually upload — almost always carry.
+ */
+function extractedAudioExt(codec: string | null): string {
+  return codec === "aac" ? ".m4a" : EXTRACTED_AUDIO_EXT;
+}
+
+function isSupportedUploadExt(ext: string): boolean {
+  return AUDIO_EXTENSIONS.has(ext) || VIDEO_EXTENSIONS.has(ext);
+}
+
 const DEFAULT_MAX_FILES = 200;
 const DEFAULT_MAX_TOTAL_BYTES = 5 * 1024 * 1024 * 1024; // 5 GiB
 
@@ -71,37 +121,128 @@ function titleFromFileName(name: string): string {
   return safeFileName(name).replace(/\.[^.]+$/, "") || "本地音频";
 }
 
-async function probeDurationSeconds(filePath: string): Promise<number> {
+export interface MediaProbe {
+  /** Rounded seconds, 0 when the probe failed or the container has no duration. */
+  durationSeconds: number;
+  /** True when ffmpeg reported at least one audio stream. Only meaningful
+   *  together with `recognized` — see the comment there. */
+  hasAudio: boolean;
+  /** Lowercased codec name of the first audio stream ("aac", "mp3", "opus",
+   *  …), or null when there is none. Picks the remux container. */
+  audioCodec: string | null;
+  /**
+   * True when ffmpeg actually opened the container and printed its
+   * `Input #0, <format>, from '...'` header.
+   *
+   * This is what separates "ffmpeg looked inside and there is genuinely no
+   * audio track" from "ffmpeg could not make sense of these bytes at all".
+   * Both produce hasAudio === false, but only the first is a file we should
+   * refuse. Unreadable bytes have always been accepted here (a truncated mp3
+   * uploads fine and simply reports duration 0), and that stays true.
+   */
+  recognized: boolean;
+  /** False when ffmpeg could not be run or timed out, so nothing else in this
+   *  object is meaningful and the caller must not reject the file on it. */
+  probed: boolean;
+}
+
+/** Parse `Duration: HH:MM:SS.ss`, the `Input #0,` header and
+ *  `Stream #0:N...: Audio:` out of the banner ffmpeg prints on stderr when
+ *  asked to open a file with no output. */
+export function parseMediaProbe(stderr: string): Omit<MediaProbe, "probed"> {
+  const match = stderr.match(/Duration:\s*(\d+):(\d+):(\d+(?:\.\d+)?)/);
+  let durationSeconds = 0;
+  if (match) {
+    const total = Number(match[1]) * 3600 + Number(match[2]) * 60 + Number(match[3]);
+    durationSeconds = Number.isFinite(total) ? Math.round(total) : 0;
+  }
+  // e.g. "  Stream #0:1[0x2](und): Audio: aac (LC) ..." — the stream index and
+  // the bracketed id/language vary, so match on the "Audio:" tag itself. An
+  // embedded cover image is a separate "Video: mjpeg ... [attached pic]" line
+  // and never matches this.
+  const audioMatch = stderr.match(/Stream #\d+:\d+[^\n]*:\s*Audio:\s*([A-Za-z0-9_]+)/);
+  const hasAudio = audioMatch !== null;
+  const audioCodec = audioMatch ? audioMatch[1].toLowerCase() : null;
+  // "Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'clip.mp4':" — absent entirely
+  // when ffmpeg bails with "Error opening input: Invalid data found ...".
+  const recognized = /^Input #\d+,/m.test(stderr);
+  return { durationSeconds, hasAudio, audioCodec, recognized };
+}
+
+async function probeMedia(filePath: string): Promise<MediaProbe> {
   return new Promise((resolve) => {
     const ffmpeg = spawn(ffmpegPath || "ffmpeg", ["-hide_banner", "-i", filePath], {
       stdio: ["ignore", "ignore", "pipe"],
     });
     let stderr = "";
+    let settled = false;
+    const done = (probe: MediaProbe) => {
+      if (settled) return;
+      settled = true;
+      resolve(probe);
+    };
+    // Video containers are much larger than the audio files this used to see,
+    // and the probe only reads headers — but a network/USB path can still be
+    // slow, so allow more than the old 5s before giving up.
     const timeout = setTimeout(() => {
       ffmpeg.kill("SIGKILL");
-      resolve(0);
+      done({ durationSeconds: 0, hasAudio: false, audioCodec: null, recognized: false, probed: false });
-    }, 5000);
+    }, 20000);
     ffmpeg.stderr.on("data", (chunk) => {
       stderr += chunk.toString("utf8");
     });
     ffmpeg.on("error", () => {
       clearTimeout(timeout);
-      resolve(0);
+      done({ durationSeconds: 0, hasAudio: false, audioCodec: null, recognized: false, probed: false });
     });
     ffmpeg.on("close", () => {
       clearTimeout(timeout);
-      const match = stderr.match(/Duration:\s*(\d+):(\d+):(\d+(?:\.\d+)?)/);
+      done({ ...parseMediaProbe(stderr), probed: true });
-      if (!match) {
+    });
-        resolve(0);
+  });
-        return;
+}
+
+/**
+ * Remux the first audio stream of `source` into `target` (#149).
+ *
+ * `-c:a copy` — the audio is moved bit-for-bit into a Matroska audio
+ * container, so this is fast, lossless, and codec-agnostic. Nothing is
+ * re-encoded, so a 200 MB .mp4 becomes a few MB .mka with the original audio
+ * intact. Video, subtitle and data streams are dropped.
+ *
+ * Returns true only if ffmpeg exited 0 AND produced a non-empty file, so a
+ * partial/zero-byte result can never be mistaken for a successful extraction.
+ * Callers fall back to keeping the original container, which plays fine.
+ */
+async function extractAudioTrack(source: string, target: string): Promise<boolean> {
+  const ok = await new Promise<boolean>((resolve) => {
+    const ffmpeg = spawn(
+      ffmpegPath || "ffmpeg",
+      ["-hide_banner", "-loglevel", "error", "-y", "-i", source,
+       "-vn", "-sn", "-dn", "-map", "0:a:0", "-c:a", "copy", target],
+      { stdio: ["ignore", "ignore", "ignore"] },
+    );
+    let settled = false;
+    const done = (v: boolean) => {
+      if (settled) return;
+      settled = true;
+      resolve(v);
+    };
+    // Remuxing is I/O bound, but a multi-GB input on a slow disk still takes
+    // a while. Cap it so a pathological file cannot wedge the upload request.
+    const timeout = setTimeout(() => {
+      ffmpeg.kill("SIGKILL");
+      done(false);
+    }, 120000);
+    ffmpeg.on("error", () => { clearTimeout(timeout); done(false); });
+    ffmpeg.on("close", (code) => { clearTimeout(timeout); done(code === 0); });
+  });
+  if (!ok) return false;
+  try {
+    return statSync(target).size > 0;
+  } catch {
+    return false;
   }
-      const hours = Number(match[1]);
-      const minutes = Number(match[2]);
-      const seconds = Number(match[3]);
-      const total = hours * 3600 + minutes * 60 + seconds;
-      resolve(Number.isFinite(total) ? Math.round(total) : 0);
-    });
-  });
 }
 
 export class LocalMusicProvider implements MusicProvider {
@@ -171,33 +312,91 @@ export class LocalMusicProvider implements MusicProvider {
     const ext = path.extname(originalName).toLowerCase();
     // Validate by the (sanitised) file extension only — never trust the
     // client-supplied Content-Type. This also guarantees the STORED extension
-    // is one of the known audio types, so a spoofed header cannot persist an
+    // is one of the known audio/video types, so a spoofed header cannot
-    // arbitrary-extension blob on disk.
+    // persist an arbitrary-extension blob on disk.
-    if (!AUDIO_EXTENSIONS.has(ext)) {
+    if (!isSupportedUploadExt(ext)) {
-      throw new Error("只支持常见音频文件，如 mp3、flac、wav、m4a、ogg、opus、aac、webm 等");
+      throw new Error(
+        "只支持常见音频文件（mp3、flac、wav、m4a、ogg、opus、aac、webm 等）" +
+        "和视频文件（mp4、mov、avi、mkv、flv、wmv 等，仅取其中的音轨播放）",
+      );
     }
     if (!input.buffer || input.buffer.length === 0) {
       throw new Error("上传文件为空");
     }
 
     const id = crypto.randomUUID();
-    const storedName = `${id}${ext}`;
+    const isVideo = VIDEO_EXTENSIONS.has(ext);
-    const filePath = path.join(this.uploadDir, storedName);
+    let filePath = path.join(this.uploadDir, `${id}${ext}`);
     writeFileSync(filePath, input.buffer);
 
-    const duration = await probeDurationSeconds(filePath);
+    let probe: MediaProbe;
+    try {
+      probe = await probeMedia(filePath);
+    } catch {
+      probe = { durationSeconds: 0, hasAudio: false, audioCodec: null, recognized: false, probed: false };
+    }
+
+    // Reject a video with no audio track up front (#149). Left to playback it
+    // would produce a silent, zero-byte stream that just looks like a broken
+    // song. Require `recognized` as well as `probed`: bytes ffmpeg cannot open
+    // at all report hasAudio false for a different reason, and those have
+    // always been accepted (a truncated upload lands with duration 0) — this
+    // change must not start rejecting them.
+    if (isVideo && probe.probed && probe.recognized && !probe.hasAudio) {
+      rmSync(filePath, { force: true });
+      throw new Error("这个视频里没有音轨，无法播放");
+    }
+
+    let size = input.buffer.length;
+    if (isVideo) {
+      // Keep only the audio. The video bytes are dead weight against the
+      // upload-directory quota and would never be used.
+      // Preferred container first; if that remux fails (a codec the container
+      // will not take), retry into Matroska, which takes almost anything.
+      const preferredExt = extractedAudioExt(probe.audioCodec);
+      let extracted = path.join(this.uploadDir, `${id}${preferredExt}`);
+      let ok = await extractAudioTrack(filePath, extracted);
+      if (!ok && preferredExt !== EXTRACTED_AUDIO_EXT) {
+        rmSync(extracted, { force: true });
+        extracted = path.join(this.uploadDir, `${id}${EXTRACTED_AUDIO_EXT}`);
+        ok = await extractAudioTrack(filePath, extracted);
+      }
+      if (ok) {
+        try {
+          // Commit filePath and size TOGETHER, and only after the source is
+          // actually gone. rmSync(force) still throws EBUSY/EPERM on Windows,
+          // and assigning size first would leave the record claiming the
+          // small extracted size while still pointing at the whole video —
+          // which makes totalBytes() under-count and lets the upload
+          // directory grow past its quota.
+          const extractedSize = statSync(extracted).size;
+          rmSync(filePath, { force: true });
+          filePath = extracted;
+          size = extractedSize;
+        } catch {
+          // Could not stat/remove (Windows lock) — keep playing the original
+          // container and drop the half-finished extract.
+          rmSync(extracted, { force: true });
+        }
+      } else {
+        // Extraction failed (exotic codec Matroska won't take, timeout, …).
+        // The original container still plays: ffmpeg picks its audio stream.
+        rmSync(extracted, { force: true });
+      }
+    }
+
     const song: LocalSongRecord = {
       id,
       name: titleFromFileName(originalName),
       artist: "本地上传",
       album: "本地音乐",
-      duration,
+      duration: probe.durationSeconds,
       coverUrl: "",
       platform: "local",
       filePath,
       originalName,
       uploadedAt: new Date().toISOString(),
-      size: input.buffer.length,
+      size,
       mimeType: input.mimeType || "application/octet-stream",
     };
 

src/web/api/music.test.ts

@@ -14,7 +14,7 @@ import { createSessionStore } from "../../data/sessions.js";
 import { createPermissionStore } from "../../data/permissions.js";
 import { createRequireAuth } from "../middleware/requireAuth.js";
 import { SESSION_COOKIE_NAME } from "../auth/validateSession.js";
-import { createMusicRouter } from "./music.js";
+import { createMusicRouter, createLocalUploadBody } from "./music.js";
 
 const empty: SearchResult = { songs: [], albums: [], playlists: [] };
 
@@ -295,3 +295,146 @@ describe("music router POST /quality — persistence (#125)", () => {
     expect(config.audioQuality.netease).toBe("lossless");
   });
 });
+
+// #149: video containers must survive the transport layer. Before this the
+// express.raw type filter only matched audio/*, video/webm and
+// application/octet-stream, so a browser-sent video/mp4 body was never parsed
+// and the handler answered 400 "raw audio body is required".
+describe("music router POST /local/upload — content types and size cap (#149)", () => {
+  let app: express.Express;
+  let botDb: BotDatabase;
+  let cookie: string;
+  let uploadAudio: ReturnType<typeof vi.fn>;
+
+  beforeEach(async () => {
+    botDb = createDatabase(":memory:");
+    const users = createUserStore(botDb.db);
+    const sessions = createSessionStore(botDb.db);
+    const admin = await users.createUser("admin", "pw-admin", "admin");
+    cookie = `${SESSION_COOKIE_NAME}=${sessions.createSession(admin.id).token}`;
+
+    uploadAudio = vi.fn(async (input: { originalName: string }) => ({
+      id: "local-1", name: input.originalName, artist: "本地上传", album: "本地音乐",
+      duration: 1, coverUrl: "", platform: "local",
+    }));
+    const local = { platform: "local", search: vi.fn().mockResolvedValue(empty), uploadAudio } as unknown as MusicProvider;
+
+    app = express();
+    app.use(express.json());
+    app.use(cookieParser());
+    app.use("/api", createRequireAuth(sessions, createPermissionStore(botDb.db), () => getDefaultConfig().guestMode));
+    app.use("/api/music", createMusicRouter(
+      fakeProvider("netease"), fakeProvider("qq"), fakeProvider("bilibili"),
+      pino({ level: "silent" }), local, getDefaultConfig(),
+    ));
+  });
+
+  afterEach(() => botDb.close());
+
+  const post = (contentType: string, body: Buffer, name = "clip.mp4") =>
+    request(app)
+      .post("/api/music/local/upload")
+      .set("Cookie", cookie)
+      .set("Content-Type", contentType)
+      .set("X-Filename", encodeURIComponent(name))
+      .send(body);
+
+  it("accepts the video MIME types browsers actually send", async () => {
+    // These are what Chrome/Firefox put on a File for .mp4/.mov/.avi/.mkv.
+    for (const ct of ["video/mp4", "video/quicktime", "video/x-msvideo", "video/x-matroska", "video/webm"]) {
+      uploadAudio.mockClear();
+      const res = await post(ct, Buffer.from("fake video bytes"));
+      expect(res.status, `content-type ${ct}`).toBe(200);
+      expect(uploadAudio).toHaveBeenCalledOnce();
+      expect(res.body.song.platform).toBe("local");
+    }
+  });
+
+  it("still accepts audio and octet-stream bodies", async () => {
+    for (const ct of ["audio/mpeg", "audio/flac", "application/octet-stream"]) {
+      uploadAudio.mockClear();
+      const res = await post(ct, Buffer.from("fake audio"), "tune.mp3");
+      expect(res.status, `content-type ${ct}`).toBe(200);
+      expect(uploadAudio).toHaveBeenCalledOnce();
+    }
+  });
+
+  it("passes the decoded filename and the content type through to the provider", async () => {
+    await post("video/mp4", Buffer.from("bytes"), "我的 视频.mp4");
+    expect(uploadAudio).toHaveBeenCalledWith(
+      expect.objectContaining({ originalName: "我的 视频.mp4", mimeType: "video/mp4" }),
+    );
+  });
+
+  it("surfaces a provider rejection as a 400 with its message", async () => {
+    uploadAudio.mockRejectedValueOnce(new Error("这个视频里没有音轨，无法播放"));
+    const res = await post("video/mp4", Buffer.from("bytes"));
+    expect(res.status).toBe(400);
+    expect(res.body.error).toBe("这个视频里没有音轨，无法播放");
+  });
+
+  it("requires authentication", async () => {
+    const res = await request(app)
+      .post("/api/music/local/upload")
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.from("bytes"));
+    expect(res.status).toBe(401);
+  });
+
+  it("rejects an oversize body as JSON, not an HTML stack trace", async () => {
+    // Same middleware the route mounts, built with a small limit so the test
+    // does not have to allocate half a gigabyte to reach the cap.
+    const tiny = express();
+    const reached = vi.fn();
+    tiny.post("/u", createLocalUploadBody("1kb"), (_req, res) => { reached(); res.json({ ok: true }); });
+
+    const res = await request(tiny)
+      .post("/u")
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.alloc(4096, 1));
+
+    expect(res.status).toBe(413);
+    expect(res.headers["content-type"]).toMatch(/application\/json/);
+    expect(res.body.error).toContain("文件太大");
+    // The HTML default handler leaked absolute server paths and a stack.
+    expect(res.text).not.toMatch(/node_modules|<\/pre>|at read/);
+    expect(reached).not.toHaveBeenCalled();
+  });
+
+  it("lets a body under the cap through the same middleware", async () => {
+    const tiny = express();
+    tiny.post("/u", createLocalUploadBody("1kb"), (req, res) => {
+      res.json({ bytes: (req.body as Buffer).length });
+    });
+    const res = await request(tiny)
+      .post("/u")
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.alloc(512, 1));
+    expect(res.status).toBe(200);
+    expect(res.body.bytes).toBe(512);
+  });
+
+  it("rejects local uploads when the feature is switched off", async () => {
+    const off = getDefaultConfig();
+    off.localAudioEnabled = false;
+    const users = createUserStore(botDb.db);
+    const sessions = createSessionStore(botDb.db);
+    const a2 = await users.createUser("admin2", "pw-admin2", "admin");
+    const c2 = `${SESSION_COOKIE_NAME}=${sessions.createSession(a2.id).token}`;
+    const app2 = express();
+    app2.use(cookieParser());
+    app2.use("/api", createRequireAuth(sessions, createPermissionStore(botDb.db), () => getDefaultConfig().guestMode));
+    app2.use("/api/music", createMusicRouter(
+      fakeProvider("netease"), fakeProvider("qq"), fakeProvider("bilibili"),
+      pino({ level: "silent" }),
+      { platform: "local", search: vi.fn(), uploadAudio } as unknown as MusicProvider, off,
+    ));
+    const res = await request(app2)
+      .post("/api/music/local/upload")
+      .set("Cookie", c2)
+      .set("Content-Type", "video/mp4")
+      .send(Buffer.from("bytes"));
+    expect(res.status).toBe(403);
+    expect(uploadAudio).not.toHaveBeenCalled();
+  });
+});

src/web/api/music.ts

@@ -7,6 +7,56 @@ import { requirePermission } from "../middleware/requirePermission.js";
 import { requireNotGuest } from "../middleware/requireNotGuest.js";
 import { authorize } from "../middleware/authorize.js";
 
+/**
+ * Body cap for a local upload. express.raw buffers the whole body in memory,
+ * so this is also the peak RAM one upload can cost — raised from 200mb for
+ * video (#149), which is far bigger than audio for the same song, but kept
+ * well short of "any video file at all" for that reason. Only the audio track
+ * survives to disk.
+ */
+export const LOCAL_UPLOAD_LIMIT = "500mb";
+
+/**
+ * Body parser for the local-upload route.
+ *
+ * `type` includes "video/*" (#149): the browser sends the File's own MIME
+ * type, so an .mp4 arrives as video/mp4 and used to be rejected by this
+ * filter before ever reaching the provider. Only the audio track is kept —
+ * uploadAudio remuxes it out on the way in.
+ *
+ * express.raw hands an oversize body to the default error handler, which
+ * answers with an HTML page carrying a stack trace and absolute server paths
+ * (unless NODE_ENV=production, which this project never sets). Video makes
+ * hitting the cap far more likely than audio did, so that one case is
+ * translated into the same JSON shape the rest of this route returns. Any
+ * other body-parser error is passed on untouched.
+ *
+ * Exported as a factory so tests can drive the identical path with a small
+ * limit instead of allocating half a gigabyte.
+ */
+export function createLocalUploadBody(limit: string): express.RequestHandler {
+  const raw = express.raw({
+    type: ["audio/*", "video/*", "application/octet-stream"],
+    limit,
+  });
+  return (req, res, next) => {
+    raw(req, res, (err?: unknown) => {
+      if (!err) {
+        next();
+        return;
+      }
+      const e = err as { type?: string; status?: number };
+      if (e?.type === "entity.too.large" || e?.status === 413) {
+        res.status(413).json({ error: `文件太大，单个文件上限 ${limit}` });
+        return;
+      }
+      next(err);
+    });
+  };
+}
+
+const localUploadBody = createLocalUploadBody(LOCAL_UPLOAD_LIMIT);
+
 export function createMusicRouter(
   neteaseProvider: MusicProvider,
   qqProvider: MusicProvider,
@@ -65,10 +115,7 @@ export function createMusicRouter(
       }
       next();
     },
-    express.raw({
+    localUploadBody,
-      type: ["audio/*", "video/webm", "application/octet-stream"],
-      limit: "200mb",
-    }),
     async (req, res) => {
       try {
         if (!localProvider) {

src/web/api/session.test.ts

@@ -167,7 +167,13 @@ describe("session router", () => {
     expect(meB.status).toBe(401);
 
     expect(u.id).toBe(meA.body.id);
-  });
+    // 20s, not the 5s default: this case runs SIX bcryptjs rounds (one hash to
+    // create the user, four verifies, one hash for the new password), and
+    // bcryptjs is pure JS. It takes ~4.5s on an idle machine — close enough to
+    // the default that it tipped over whenever the full suite saturated the
+    // CPU, which made it look like a real intermittent failure. The work is
+    // genuinely slow, not hung, so the timeout is what was wrong.
+  }, 20000);
 });
 
 describe("session router — guest mode", () => {

test-ts6-version.cjs

@@ -0,0 +1,92 @@
+const { Client, generateIdentity } = require('@honeybbq/teamspeak-client');
+
+function escapeTS3(value) {
+  return value
+    .replace(/\\/g, "\\\\")
+    .replace(/\//g, "\\/")
+    .replace(/ /g, "\\s")
+    .replace(/\|/g, "\\p")
+    .replace(/\n/g, "\\n")
+    .replace(/\r/g, "\\r")
+    .replace(/\t/g, "\\t");
+}
+
+function replaceField(cmd, key, value) {
+  const escaped = escapeTS3(value);
+  const regex = new RegExp(key + "=\\S*");
+  if (regex.test(cmd)) return cmd.replace(regex, key + "=" + escaped);
+  return cmd;
+}
+
+// Version to test - passed via env
+const VERSIONS = [
+  {
+    name: "6.0.0-beta2",
+    version: "6.0.0-beta2 [Build: 1737468425]",
+    platform: "Windows",
+    sign: "b5mySiqYAd4Lv5TZNflW+G5Gk8y7Woo9YnJfHRjmzhUyMdGfF1O7fSGJzmx2Hpe/PKaY2aDuKKD6lSxbLYlKCw==",
+  },
+  {
+    name: "3.?.? wildcard",
+    version: "3.?.? [Build: 5680278000]",
+    platform: "Windows",
+    sign: "DX5NIYLvfJEUjuIbCidnoeozxIDRRkpq3I9vVMBmE9L2qnekOoBzSenkzsg2lC9CMv8K5hkEzhr2TYUYSwUXCg==",
+  },
+  {
+    name: "5.0.0-beta77",
+    version: "5.0.0-beta77 [Build: 1702382332]",
+    platform: "Windows",
+    sign: "Ee6DzP16MUXpdKWjiSY0NGb4thN22/Ks0hwNcaMrWoaadgkM6c5477X0IbGFWVjzTWfjFTEad5noYLUPDWSgCQ==",
+  },
+  {
+    name: "3.6.2 (corrected sign)",
+    version: "3.6.2 [Build: 1695203293]",
+    platform: "Windows",
+    sign: "4BdaZpdgUSMCuIs8qcloJPNxNlJ4o7QKnxMCRO60mSOTtJZyKjOrGLAmeAEtLIJjcjmdSpycMbQOIV92K2vXAw==",
+  },
+];
+
+const idx = parseInt(process.env.VERSION_IDX || "0");
+const V = VERSIONS[idx];
+
+const identity = generateIdentity(8);
+const client = new Client(identity, "localhost:9987", "MusicBot", {
+  logger: {
+    debug: () => {},
+    info: (m) => console.log("[INFO]", m),
+    warn: (m) => console.log("[WARN]", m),
+    error: (m) => console.log("[ERROR]", m),
+  },
+});
+
+console.log("Testing version: " + V.name + " -> " + V.version);
+client.connect().then(() => {
+  const handler = client.handler;
+  const origSendPacket = handler.sendPacket.bind(handler);
+  handler.sendPacket = (pType, data, flags) => {
+    if (pType === 2) {
+      let str = Buffer.from(data).toString("utf-8");
+      if (str.startsWith("clientinit ")) {
+        str = replaceField(str, "client_version", V.version);
+        str = replaceField(str, "client_platform", V.platform);
+        str = replaceField(str, "client_version_sign", V.sign);
+        console.log("[PATCHED] version=" + V.version);
+        origSendPacket(pType, Buffer.from(str), flags);
+        return;
+      }
+    }
+    origSendPacket(pType, data, flags);
+  };
+  return client.waitConnected();
+}).then(() => {
+  console.log("SUCCESS! Connected with clientId = " + client.clientID());
+  client.disconnect();
+  process.exit(0);
+}).catch((err) => {
+  console.log("ERROR:", err && err.message || err);
+  process.exit(1);
+});
+setTimeout(() => {
+  console.log("TIMEOUT - version rejected");
+  process.exit(2);
+}, 10000);

web/src/views/Search.vue

@@ -28,18 +28,18 @@
       >
         <Icon icon="mdi:tray-arrow-up" class="upload-icon" />
         <div class="upload-copy">
-          <div class="upload-title">拖拽本地音频到这里上传</div>
+          <div class="upload-title">拖拽本地音频 / 视频到这里上传</div>
-          <div class="upload-subtitle">支持 mp3、flac、wav、m4a、ogg、opus、aac、webm 等格式，上传后可直接播放或加入队列</div>
+          <div class="upload-subtitle">音频支持 mp3、flac、wav、m4a、ogg、opus、aac、webm 等，视频支持 mp4、mov、avi、mkv、flv、wmv 等（只取其中的音轨播放）；上传后可直接播放或加入队列</div>
         </div>
         <button class="upload-btn" :disabled="uploading" @click="fileInput?.click()">
-          {{ uploading ? '上传中...' : '选择音频' }}
+          {{ uploading ? '上传中...' : '选择文件' }}
         </button>
         <input
           ref="fileInput"
           class="file-input"
           type="file"
           multiple
-          accept="audio/*,.mp3,.flac,.wav,.m4a,.aac,.ogg,.opus,.webm,.wma,.alac,.aiff,.ape"
+          accept="audio/*,video/*,.mp3,.flac,.wav,.m4a,.aac,.ogg,.opus,.webm,.wma,.alac,.aiff,.ape,.mp4,.mov,.avi,.mkv,.flv,.wmv,.m4v,.mpg,.mpeg,.3gp,.ts,.m2ts,.ogv"
           @change="handleFileSelect"
         />
       </div>
@@ -404,8 +404,17 @@ async function doSearch() {
 }
 
 
-function isAudioFile(file: File): boolean {
+/** Must match LOCAL_UPLOAD_LIMIT in src/web/api/music.ts. */
-  return file.type.startsWith('audio/') || /\.(mp3|flac|wav|m4a|aac|ogg|opus|webm|wma|alac|aiff|ape)$/i.test(file.name);
+const UPLOAD_MAX_MB = 500;
+const UPLOAD_MAX_BYTES = UPLOAD_MAX_MB * 1024 * 1024;
+
+// Video is accepted too (#149) — the server keeps only the audio track.
+// Keep the extension list in sync with AUDIO_EXTENSIONS / VIDEO_EXTENSIONS in
+// src/music/local.ts; the server re-validates, this just avoids a round-trip.
+function isMediaFile(file: File): boolean {
+  return file.type.startsWith('audio/')
+    || file.type.startsWith('video/')
+    || /\.(mp3|flac|wav|m4a|aac|ogg|opus|webm|wma|alac|aiff|ape|mp4|mov|avi|mkv|flv|wmv|m4v|mpg|mpeg|3gp|ts|m2ts|ogv)$/i.test(file.name);
 }
 
 async function uploadLocalFiles(fileList: File[]) {
@@ -414,10 +423,22 @@ async function uploadLocalFiles(fileList: File[]) {
     uploadMessage.value = '本地音频播放已关闭';
     return;
   }
-  const files = fileList.filter(isAudioFile);
+  const candidates = fileList.filter(isMediaFile);
+  if (candidates.length === 0) {
+    uploadMessageType.value = 'error';
+    uploadMessage.value = '没有找到可上传的音频 / 视频文件';
+    return;
+  }
+
+  // Reject oversize files before spending minutes uploading them (#149).
+  // The server enforces the same cap (LOCAL_UPLOAD_LIMIT in
+  // src/web/api/music.ts) and answers 413 — this only saves the round-trip,
+  // which matters now that a single video can be hundreds of megabytes.
+  const files = candidates.filter((f) => f.size <= UPLOAD_MAX_BYTES);
+  const oversize = candidates.filter((f) => f.size > UPLOAD_MAX_BYTES);
   if (files.length === 0) {
     uploadMessageType.value = 'error';
-    uploadMessage.value = '没有找到可上传的音频文件';
+    uploadMessage.value = `文件太大，单个文件上限 ${UPLOAD_MAX_MB} MB：${oversize[0].name}`;
     return;
   }
 
@@ -426,8 +447,19 @@ async function uploadLocalFiles(fileList: File[]) {
   uploadMessage.value = `正在上传 ${files.length} 个文件...`;
 
   const uploaded: Song[] = [];
-  const failed: string[] = [];
+  const failed: string[] = oversize.map((f) => `${f.name}: 超过 ${UPLOAD_MAX_MB} MB 上限`);
-  for (const file of files) {
+  for (const [i, file] of files.entries()) {
+    // Videos are orders of magnitude bigger than the audio files this used to
+    // handle (#149), so a silent "正在上传..." can sit there for minutes and
+    // look hung. Report per-file percentage while the bytes are in flight, and
+    // switch to a processing note once the server takes over (it still has to
+    // probe the file and remux the audio track out).
+    const label = files.length > 1 ? `（${i + 1}/${files.length}）` : '';
+    const setProgress = (text: string) => {
+      uploadMessageType.value = 'info';
+      uploadMessage.value = `${text}${label}：${file.name}`;
+    };
+    setProgress('正在上传');
     try {
       const res = await axios.post('/api/music/local/upload', file, {
         headers: {
@@ -435,6 +467,11 @@ async function uploadLocalFiles(fileList: File[]) {
           'X-Filename': encodeURIComponent(file.name),
         },
         maxBodyLength: Infinity,
+        onUploadProgress: (e) => {
+          if (!e.total) return;
+          const pct = Math.round((e.loaded / e.total) * 100);
+          setProgress(pct >= 100 ? '服务端处理中' : `正在上传 ${pct}%`);
+        },
       });
       if (res.data?.song) uploaded.push(res.data.song as Song);
     } catch (err: any) {