feat(subtitles): 字幕站(SubHD、字幕库)在线搜索与下载;rar / 7z 解压;验证码与页面解析在沙箱里

Co-Authored-By: Claude Opus 5.5 <[email protected]>

apps/server/src/modules/sources-identity/routes/sources.ts

@@ -25,6 +25,7 @@ import {
   SiteUserDataPoint,
   SiteUserDataQuery,
   type SourceRelease,
+  SUBTITLE_SITES,
 } from "@umt/contracts";
 import { type FeedSourceRow, sql } from "@umt/db";
 import { z } from "zod";
@@ -92,19 +93,35 @@ export function sourcesRoutes(app: ModuleApp, rt: () => SourcesRuntime) {
       const rows = await db().selectFrom("site").select(["id", "key"]).where("deleted_at", "is", null).execute();
       const byKey = new Map(rows.map((r) => [r.key, r.id]));
       return {
-        items: listSiteDefinitions().map((d) => ({
-          key: d.id,
-          name: d.umt.kind === "nexusphp" ? (d.umt.displayName ?? d.name) : d.umt.displayName,
-          kind: definitionKind(d),
-          adapter: adapterOf(d),
-          baseUrl: d.domain,
-          shippedEnabled: d.enabled !== false,
-          definitionVersion: d.umt.version,
-          capabilities: d.umt.capabilities,
-          defaultRateLimit: d.umt.rateLimit,
-          authSupported: ("authSupported" in d.umt ? d.umt.authSupported : undefined) ?? [],
-          siteId: byKey.get(d.id) ?? null,
-        })),
+        items: [
+          ...listSiteDefinitions().map((d) => ({
+            key: d.id,
+            name: d.umt.kind === "nexusphp" ? (d.umt.displayName ?? d.name) : d.umt.displayName,
+            kind: definitionKind(d),
+            adapter: adapterOf(d),
+            baseUrl: d.domain,
+            shippedEnabled: d.enabled !== false,
+            definitionVersion: d.umt.version,
+            capabilities: d.umt.capabilities,
+            defaultRateLimit: d.umt.rateLimit,
+            authSupported: ("authSupported" in d.umt ? d.umt.authSupported : undefined) ?? [],
+            siteId: byKey.get(d.id) ?? null,
+          })),
+          // 内置字幕站:不是资源适配器,只用于补字幕(subtitles 模块)
+          ...SUBTITLE_SITES.map((d) => ({
+            key: d.key,
+            name: d.name,
+            kind: "subtitle" as const,
+            adapter: "subtitle" as const,
+            baseUrl: d.baseUrl,
+            shippedEnabled: true,
+            definitionVersion: "builtin",
+            capabilities: { list: false, search: false, detail: false, download: true, rss: false },
+            defaultRateLimit: { minIntervalMs: 2000, burst: 2 },
+            authSupported: [],
+            siteId: byKey.get(d.key) ?? null,
+          })),
+        ],
       };
     },
   );
@@ -159,6 +176,9 @@ export function sourcesRoutes(app: ModuleApp, rt: () => SourcesRuntime) {
         throw new ApiException("bad_request", `注册表中没有站点 ${body.key}；未适配站点只能以 generic_rss 登记`);
       }
       if (def && definitionKind(def) !== "pt") throw new ApiException("bad_request", `${body.key} 不是 PT 站`);
+    } else if (kind === "subtitle") {
+      if (!SUBTITLE_SITES.some((d) => d.key === body.key) || body.adapter !== "subtitle")
+        throw new ApiException("bad_request", "字幕站要从内置字幕站里选（SubHD、字幕库）");
     } else if (publicDef) {
       if (body.adapter !== adapterOf(publicDef))
         throw new ApiException("bad_request", `站点 ${body.key} 的适配器类型应为 ${adapterOf(publicDef)}`);

apps/server/src/modules/sources-identity/service/account.ts

@@ -810,7 +810,20 @@ export async function checkSiteHealth(
   let message: string | null = null;
   let detail: SiteHealthDetail | null = null;
   const def = getSiteDefinition(site.key);
-  if (!def) {
+  if (site.kind === "subtitle") {
+    // 字幕站:用站点出口(代理 / UA / Cloudflare 处理)打开首页,2xx 即可用
+    const t0 = Date.now();
+    try {
+      const res = await rt.siteFetch(site)(site.base_url, { signal: AbortSignal.timeout(20_000) });
+      await res.body?.cancel();
+      ok = res.ok;
+      status = res.ok ? "ok" : "unreachable";
+      latencyMs = Date.now() - t0;
+      message = res.ok ? null : `首页返回 HTTP ${res.status}`;
+    } catch (e) {
+      message = (e as Error).message.slice(0, 300);
+    }
+  } else if (!def) {
     message = "站点没有适配器定义，无法检测";
     status = "unsupported";
   } else {

apps/server/src/modules/subtitles/index.ts

@@ -1,6 +1,9 @@
-import { Api, type SubtitleApplyRequest } from "@umt/contracts";
+import { Api, type SubtitleApplyRequest, type SubtitleFetchRequest, type SubtitleSource } from "@umt/contracts";
+import { ApiException } from "../../http/errors.ts";
 import { route } from "../agent/http.ts";
+import { sourcesIdentityModule } from "../sources-identity/index.ts";
 import type { ServerModule } from "../types.ts";
+import { downloadOnline, SubtitleSiteError, searchOnline, subtitleSearchQuery } from "./online/index.ts";
 import { applySubtitles, parseSubtitles, subtitleTargets, subtitleWrites, undoSubtitleWrite } from "./service.ts";
 
 /** 补字幕(docs/plan/subtitles.md):上传 / 在线下载的字幕解析、配对、写入、历史与撤销 */
@@ -9,8 +12,39 @@ export function createSubtitlesModule(): ServerModule {
     name: "subtitles",
     registerRoutes(app, ctx) {
       const tag = { tag: "subtitles" };
+      const rt = () => sourcesIdentityModule.runtime(ctx);
+      const siteErr = async <T>(fn: () => Promise<T>): Promise<T> => {
+        try {
+          return await fn();
+        } catch (e) {
+          if (e instanceof SubtitleSiteError) throw new ApiException("upstream_unavailable", e.message);
+          throw e;
+        }
+      };
       route(app, Api.subtitles.targets, ({ query }) => subtitleTargets(ctx, query.entityId), tag);
       route(app, Api.subtitles.parse, ({ body }) => parseSubtitles(ctx, body.files, body.entityId), tag);
+      route(
+        app,
+        Api.subtitles.search,
+        ({ body }) => siteErr(async () => searchOnline(ctx, rt(), await subtitleSearchQuery(ctx, body.entityId, body))),
+        tag,
+      );
+      route(
+        app,
+        Api.subtitles.fetch,
+        ({ body }) =>
+          siteErr(async () => {
+            const b = body as SubtitleFetchRequest;
+            const f = await downloadOnline(ctx, rt(), b.siteId, b.ref);
+            const parsed = await parseSubtitles(ctx, [{ name: f.name, base64: f.base64 }], b.entityId);
+            // 在线来源:每项带上字幕站与条目地址,写入时记进历史
+            return {
+              ...parsed,
+              items: parsed.items.map((it) => ({ ...it, source: f.siteKey as SubtitleSource, sourceRef: f.pageUrl })),
+            };
+          }),
+        tag,
+      );
       route(
         app,
         Api.subtitles.apply,

apps/server/src/modules/subtitles/online/index.ts

@@ -0,0 +1,359 @@
+// 在线字幕(docs/plan/subtitles.md 第 2 步):在已启用的字幕站(site.kind = subtitle)搜索、下载。
+// 请求走站点出口 rt.siteFetch(站点代理、UA、Cloudflare 挑战走浏览器);页面解析与验证码识别在沙箱里(ADR-005)。
+// 流程对照 MP 插件 online_subtitles/providers/{subhd,zimuku}.py,页面结构 2026-10-03 实测。
+import type { SubtitleSearchResponse, SubtitleSearchResult } from "@umt/contracts";
+import type { SiteRow } from "@umt/db";
+import { subtitleEpisode } from "@umt/domain";
+import { runSandbox } from "../../../sandbox/client.ts";
+import { b64encode } from "../../../sandbox/jobs.ts";
+import type { SourcesRuntime } from "../../sources-identity/runtime.ts";
+import type { ServerContext } from "../../types.ts";
+import type { PageEntry, PageKind, PageResult } from "./parse.ts";
+
+const MAX_DOWNLOAD_BYTES = 30 * 1024 * 1024;
+
+/** 站内操作失败(给用户看的原因) */
+export class SubtitleSiteError extends Error {
+  override name = "SubtitleSiteError";
+}
+
+// ── 请求:站点出口 + 每站一个 Cookie 罐(字幕库的验证 Cookie、SubHD 的会话) ──
+
+const jars = new Map<string, Map<string, string>>();
+
+interface Env {
+  rt: SourcesRuntime;
+  site: SiteRow;
+}
+
+async function request(
+  env: Env,
+  url: string,
+  init: { referer?: string; json?: unknown; signal?: AbortSignal } = {},
+): Promise<Response> {
+  const jar = jars.get(env.site.id) ?? new Map<string, string>();
+  jars.set(env.site.id, jar);
+  const headers: Record<string, string> = { Accept: "text/html,application/json;q=0.9,*/*;q=0.8" };
+  if (jar.size) headers.cookie = [...jar].map(([k, v]) => `${k}=${v}`).join("; ");
+  if (init.referer) headers.referer = init.referer;
+  if (init.json !== undefined) headers["content-type"] = "application/json";
+  let res: Response;
+  try {
+    res = await env.rt.siteFetch(env.site)(url, {
+      method: init.json === undefined ? "GET" : "POST",
+      headers,
+      ...(init.json !== undefined ? { body: JSON.stringify(init.json) } : {}),
+      signal: init.signal ?? AbortSignal.timeout(30_000),
+    });
+  } catch (e) {
+    const n = (e as Error).name;
+    if (n === "TimeoutError" || n === "AbortError")
+      throw new SubtitleSiteError(`连接${env.site.name}超时，稍后再试（站点可能在限流或网络不通）`);
+    throw new SubtitleSiteError(`连接${env.site.name}失败：${(e as Error).message.slice(0, 200)}`);
+  }
+  for (const c of res.headers.getSetCookie()) {
+    const kv = c.split(";")[0] ?? "";
+    const i = kv.indexOf("=");
+    if (i > 0) jar.set(kv.slice(0, i).trim(), kv.slice(i + 1).trim());
+  }
+  return res;
+}
+
+async function getText(env: Env, url: string, referer?: string): Promise<string> {
+  const res = await request(env, url, referer ? { referer } : {});
+  if (!res.ok) throw new SubtitleSiteError(`${env.site.name} 返回 HTTP ${res.status}`);
+  return await res.text();
+}
+
+const page = <K extends PageKind>(kind: K, html: string) =>
+  runSandbox("subtitle.page", { kind, html }, { timeoutMs: 15_000 }) as Promise<PageResult<K>>;
+
+const abs = (env: Env, path: string) => new URL(path, env.site.base_url).href;
+
+/** content-disposition 里的文件名:常见是 UTF-8 字节被当成 latin1 发出,按字节还原 */
+function fileNameOf(res: Response, url: string): string {
+  const cd = res.headers.get("content-disposition") ?? "";
+  const star = /filename\*=UTF-8''([^;]+)/i.exec(cd)?.[1];
+  if (star) return decodeURIComponent(star);
+  const plain = /filename="?([^";]+)"?/i.exec(cd)?.[1];
+  if (plain) {
+    const bytes = Uint8Array.from(plain, (c) => c.charCodeAt(0) & 0xff);
+    const utf8 = new TextDecoder("utf-8", { fatal: false }).decode(bytes);
+    return utf8.includes("�") ? plain : utf8;
+  }
+  return decodeURIComponent(new URL(url).pathname.split("/").pop() || "subtitle.zip");
+}
+
+async function getFile(env: Env, url: string, referer: string): Promise<{ name: string; bytes: Uint8Array }> {
+  const res = await request(env, url, { referer });
+  if (!res.ok) throw new SubtitleSiteError(`下载返回 HTTP ${res.status}`);
+  const len = Number(res.headers.get("content-length") ?? 0);
+  if (len > MAX_DOWNLOAD_BYTES) throw new SubtitleSiteError("字幕包超过 30 MB");
+  const bytes = new Uint8Array(await res.arrayBuffer());
+  if (bytes.length > MAX_DOWNLOAD_BYTES) throw new SubtitleSiteError("字幕包超过 30 MB");
+  const head = new TextDecoder().decode(bytes.subarray(0, 200)).toLowerCase();
+  if (/<!doctype html|<html/.test(head)) throw new SubtitleSiteError("下载地址返回的是网页，不是字幕文件");
+  return { name: fileNameOf(res, res.url || url), bytes };
+}
+
+// ── 搜索词与匹配 ──
+
+export interface SearchQuery {
+  keywords: string[];
+  doubanId: string | null;
+  season: number | null;
+  episode: number | null;
+}
+
+const norm = (s: string) => s.toLowerCase().replace(/[\s\p{P}\p{S}]/gu, "");
+
+/** 标题里含任一搜索词(去掉空白与标点比较) */
+function relevant(title: string, keywords: string[]): boolean {
+  const t = norm(title);
+  return keywords.some((k) => {
+    const n = norm(k);
+    return n.length >= 2 && t.includes(n);
+  });
+}
+
+/** 季集不符的丢掉;合集 / 季包保留 */
+function episodeOk(title: string, q: SearchQuery): { ok: boolean; season: number | null; episode: number | null } {
+  const ep = subtitleEpisode(`${title}.srt`);
+  if (q.season !== null && !ep.seasonInferred && ep.season !== null && ep.season !== q.season)
+    return { ok: false, season: ep.season, episode: ep.episode };
+  if (q.episode !== null && ep.episode !== null && ep.episode !== q.episode)
+    return { ok: false, season: ep.season, episode: ep.episode };
+  return { ok: true, season: ep.seasonInferred ? null : ep.season, episode: ep.episode };
+}
+
+const LANG_SCORE: Record<string, number> = { 双语: 3, 简体: 2, 繁体: 1 };
+
+function toResult(env: Env, e: PageEntry, q: SearchQuery): SubtitleSearchResult | null {
+  const ep = episodeOk(e.title, q);
+  if (!ep.ok) return null;
+  const lang = Math.max(0, ...e.languages.map((l) => LANG_SCORE[l] ?? 0));
+  const exact = q.episode !== null && ep.episode === q.episode ? 2 : 0;
+  return {
+    siteId: env.site.id,
+    siteKey: env.site.key,
+    siteName: env.site.name,
+    ref: e.ref,
+    title: e.title,
+    pageUrl: abs(env, e.ref),
+    languages: e.languages,
+    format: e.format,
+    downloads: e.downloads,
+    uploader: e.uploader,
+    publishedAt: e.publishedAt,
+    season: ep.season,
+    episode: ep.episode,
+    score: lang * 10 + exact * 5 + Math.log10((e.downloads ?? 0) + 1),
+  };
+}
+
+// ── SubHD ──
+
+async function subhdSearch(env: Env, q: SearchQuery): Promise<PageEntry[]> {
+  // SubHD 的条目 ID 就是豆瓣条目 ID:有豆瓣 ID 直接打开条目页
+  if (q.doubanId) {
+    const d = await page("subhd.list", await getText(env, abs(env, `/d/${q.doubanId}`), env.site.base_url));
+    if (d.entries.length) return d.entries;
+  }
+  for (const kw of q.keywords.slice(0, 3)) {
+    const html = await getText(env, abs(env, `/search/${encodeURIComponent(kw)}`), env.site.base_url);
+    const r = await page("subhd.list", html);
+    const hits = r.entries.filter((e) => relevant(e.title, q.keywords));
+    if (hits.length) return hits;
+    for (const d of r.details.filter((x) => relevant(x.title, q.keywords)).slice(0, 2)) {
+      const dd = await page("subhd.list", await getText(env, abs(env, `/d/${d.id}`), env.site.base_url));
+      if (dd.entries.length) return dd.entries;
+    }
+  }
+  return [];
+}
+
+async function subhdDownload(env: Env, ref: string): Promise<{ name: string; bytes: Uint8Array }> {
+  const subUrl = abs(env, ref);
+  const sub = await page("subhd.sub", await getText(env, subUrl, env.site.base_url));
+  let down = sub.down;
+  if (!down && sub.dataSid) {
+    const prep = (await (
+      await request(env, abs(env, "/api/sub/prepare-download"), { referer: subUrl, json: { sid: sub.dataSid } })
+    ).json()) as { success?: boolean; url?: string };
+    if (prep.success && prep.url?.includes("/down/")) down = prep.url;
+  }
+  if (!down) throw new SubtitleSiteError("SubHD 字幕页没有下载按钮");
+  const downUrl = abs(env, down);
+  const sid = new URL(downUrl).pathname.split("/").pop() ?? "";
+  await (await request(env, downUrl, { referer: subUrl })).body?.cancel();
+  const api = async (cap: string) =>
+    (await (await request(env, abs(env, "/api/sub/down"), { referer: downUrl, json: { sid, cap } })).json()) as {
+      success?: boolean;
+      pass?: boolean;
+      msg?: string;
+      url?: string;
+    };
+  let r = await api("");
+  if (r.pass === false) {
+    // SVG 文字验证码:沙箱里识别后带上再提交一次
+    const { code } = await runSandbox("captcha.solve", { kind: "subhd", svg: String(r.msg ?? "") });
+    r = await api(code ?? "");
+  }
+  if (!r.success || !r.url)
+    throw new SubtitleSiteError("SubHD 下载验证没通过：可在站点的原站视图里打开这条字幕手动下载后上传");
+  return await getFile(env, abs(env, r.url), downUrl);
+}
+
+// ── 字幕库 ──
+
+/** 打开页面;遇到云锁验证码页就在沙箱里识别、提交验证后重开(最多 3 次) */
+async function zimukuPage(env: Env, url: string, referer?: string): Promise<string> {
+  for (let i = 0; i < 3; i++) {
+    const res = await request(env, url, referer ? { referer } : {});
+    const html = await res.text();
+    const cap = await page("zimuku.captcha", html);
+    if (!cap.bmpBase64) {
+      if (!res.ok) throw new SubtitleSiteError(`${env.site.name} 返回 HTTP ${res.status}`);
+      return html;
+    }
+    const { code } = await runSandbox("captcha.solve", { kind: "zimuku", bmpBase64: cap.bmpBase64 });
+    if (!code) break;
+    const hex = [...code].map((c) => c.charCodeAt(0).toString(16)).join("");
+    await (await request(env, `${url}${url.includes("?") ? "&" : "?"}security_verify_img=${hex}`)).body?.cancel();
+  }
+  throw new SubtitleSiteError("字幕库的验证码没能自动通过：可在站点的原站视图里打开一次再试");
+}
+
+async function zimukuSearch(env: Env, q: SearchQuery): Promise<PageEntry[]> {
+  for (const kw of q.keywords.slice(0, 3)) {
+    const s = await page("zimuku.search", await zimukuPage(env, abs(env, `/search?q=${encodeURIComponent(kw)}`)));
+    const subs = s.subs.filter((x) => !x.title || relevant(x.title, q.keywords)).slice(0, 3);
+    const out: PageEntry[] = [];
+    for (const x of subs) {
+      const p = await page("zimuku.subs", await zimukuPage(env, abs(env, x.path), env.site.base_url));
+      out.push(...p.entries);
+    }
+    if (out.length) return out;
+  }
+  return [];
+}
+
+async function zimukuDownload(env: Env, ref: string): Promise<{ name: string; bytes: Uint8Array }> {
+  const detailUrl = abs(env, ref);
+  const d = await page("zimuku.detail", await zimukuPage(env, detailUrl, env.site.base_url));
+  if (!d.dl) throw new SubtitleSiteError("字幕库字幕页没有下载入口");
+  const dlUrl = new URL(d.dl, detailUrl).href;
+  const links = (await page("zimuku.dl", await zimukuPage(env, dlUrl, detailUrl))).links;
+  let last = "字幕库没有可用的下载线路";
+  for (const l of links) {
+    try {
+      const f = await getFile(env, new URL(l, dlUrl).href, dlUrl);
+      if (f.bytes.length > 64) return f;
+    } catch (e) {
+      last = (e as Error).message;
+    }
+  }
+  throw new SubtitleSiteError(last);
+}
+
+const PROVIDERS: Record<
+  string,
+  {
+    search(env: Env, q: SearchQuery): Promise<PageEntry[]>;
+    download(env: Env, ref: string): Promise<{ name: string; bytes: Uint8Array }>;
+  }
+> = {
+  subhd: { search: subhdSearch, download: subhdDownload },
+  zimuku: { search: zimukuSearch, download: zimukuDownload },
+};
+
+// ── 对外 ──
+
+export async function subtitleSearchQuery(
+  ctx: ServerContext,
+  entityId: string,
+  opts: { season?: number | undefined; episode?: number | undefined; keyword?: string | undefined },
+): Promise<SearchQuery> {
+  const e = await ctx.db
+    .selectFrom("media_entity")
+    .select(["title", "original_title"])
+    .where("id", "=", entityId)
+    .executeTakeFirst();
+  if (!e) throw new SubtitleSiteError("作品不存在");
+  const aliases = await ctx.db
+    .selectFrom("title_alias")
+    .select("alias")
+    .where("media_entity_id", "=", entityId)
+    .where("content_unit_id", "is", null)
+    .limit(5)
+    .execute();
+  const douban = await ctx.db
+    .selectFrom("external_identity")
+    .select("external_id")
+    .where("media_entity_id", "=", entityId)
+    .where("provider", "=", "douban")
+    .where("status", "=", "accepted")
+    .executeTakeFirst();
+  const keywords = [opts.keyword, e.title, e.original_title, ...aliases.map((a) => a.alias)]
+    .map((x) => x?.trim())
+    .filter((x): x is string => !!x && x.length >= 2);
+  return {
+    keywords: [...new Set(keywords)],
+    doubanId: douban?.external_id ?? null,
+    season: opts.season ?? null,
+    episode: opts.episode ?? null,
+  };
+}
+
+async function subtitleSites(ctx: ServerContext): Promise<SiteRow[]> {
+  return await ctx.db
+    .selectFrom("site")
+    .selectAll()
+    .where("kind", "=", "subtitle")
+    .where("enabled", "=", true)
+    .where("deleted_at", "is", null)
+    .orderBy("name")
+    .execute();
+}
+
+export async function searchOnline(
+  ctx: ServerContext,
+  rt: SourcesRuntime,
+  q: SearchQuery,
+): Promise<SubtitleSearchResponse> {
+  const sites = (await subtitleSites(ctx)).filter((s) => PROVIDERS[s.key]);
+  const results: SubtitleSearchResult[] = [];
+  const errors: SubtitleSearchResponse["errors"] = [];
+  await Promise.all(
+    sites.map(async (site) => {
+      const env = { rt, site };
+      try {
+        const entries = await PROVIDERS[site.key]!.search(env, q);
+        const seen = new Set<string>();
+        for (const e of entries) {
+          if (seen.has(e.ref)) continue;
+          seen.add(e.ref);
+          const r = toResult(env, e, q);
+          if (r) results.push(r);
+        }
+      } catch (e) {
+        errors.push({ siteName: site.name, reason: (e as Error).message.slice(0, 300) });
+      }
+    }),
+  );
+  results.sort((a, b) => b.score - a.score);
+  return { results: results.slice(0, 60), keywords: q.keywords, errors, noSites: sites.length === 0 };
+}
+
+export async function downloadOnline(
+  ctx: ServerContext,
+  rt: SourcesRuntime,
+  siteId: string,
+  ref: string,
+): Promise<{ name: string; base64: string; siteKey: string; pageUrl: string }> {
+  const site = (await subtitleSites(ctx)).find((s) => s.id === siteId);
+  if (!site || !PROVIDERS[site.key]) throw new SubtitleSiteError("字幕站不存在或已停用");
+  const env = { rt, site };
+  const f = await PROVIDERS[site.key]!.download(env, ref);
+  return { name: f.name, base64: b64encode(f.bytes), siteKey: site.key, pageUrl: abs(env, ref) };
+}

apps/server/src/modules/subtitles/online/parse.ts

@@ -0,0 +1,169 @@
+// 字幕站页面解析(纯函数,在沙箱任务 subtitle.page 里跑):SubHD / 字幕库的搜索页、条目页、字幕页、下载页 → 结构化数据。
+// 页面结构 2026-10-03 实测(样本在 .local/site-samples/subs/),对照 MP 插件 online_subtitles/providers/{subhd,zimuku}.py。
+
+export interface PageEntry {
+  /** 站内引用:SubHD 为 /a/<sid>,字幕库为 /detail/<id>.html */
+  ref: string;
+  title: string;
+  /** 简体 / 繁体 / 双语 / 英文 …(站点标注或从标题、说明里猜) */
+  languages: string[];
+  format: string | null;
+  downloads: number | null;
+  uploader: string | null;
+  publishedAt: string | null;
+}
+
+const strip = (s: string) =>
+  s
+    .replace(/<[^>]+>/g, " ")
+    .replace(/&nbsp;/g, " ")
+    .replace(/&amp;/g, "&")
+    .replace(/&lt;/g, "<")
+    .replace(/&gt;/g, ">")
+    .replace(/&#39;/g, "'")
+    .replace(/&quot;/g, '"')
+    .replace(/\s+/g, " ")
+    .trim();
+const href = (q: string) => `href=["']${q}["']`;
+
+/** 从标题 / 说明里猜语言 */
+export function guessLanguages(text: string): string[] {
+  const t = text.toLowerCase();
+  const out: string[] = [];
+  if (/双语|中英|简英|繁英|chs&eng|cht&eng|bilingual/.test(t)) out.push("双语");
+  if (/简体|简中|chs|gb|简繁|简日/.test(t)) out.push("简体");
+  if (/繁体|繁體|繁中|cht|big5|简繁/.test(t)) out.push("繁体");
+  if (/英文|英语|\beng?\b|english/.test(t) && !out.includes("双语")) out.push("英文");
+  return [...new Set(out)];
+}
+
+const formatOf = (text: string) => /\b(ass|ssa|srt|sup|vtt|sub|idx)\b/i.exec(text)?.[1]?.toUpperCase() ?? null;
+const num = (s: string | undefined) => {
+  if (!s) return null;
+  const m = /([\d.]+)\s*(万)?/.exec(s.replace(/,/g, ""));
+  return m ? Math.round(Number(m[1]) * (m[2] ? 10_000 : 1)) : null;
+};
+
+// ── SubHD ──
+
+/** 搜索页 / 条目页(/d/<豆瓣 ID>):结果区里的字幕(/a/<sid>),不含侧栏热门 */
+function subhdList(html: string): { entries: PageEntry[]; details: { id: string; title: string }[] } {
+  const hot = html.search(/<h\d[^>]*>\s*热门|class="[^"]*hot[^"]*"/);
+  const body = hot > 0 ? html.slice(0, hot) : html;
+  const entries: PageEntry[] = [];
+  const seen = new Set<string>();
+  for (const m of body.matchAll(new RegExp(`<a[^>]+${href("/a/([A-Za-z0-9_-]+)")}[^>]*>([\\s\\S]*?)</a>`, "g"))) {
+    const sid = m[1]!;
+    const title = strip(m[2]!);
+    if (seen.has(sid) || !title) continue;
+    seen.add(sid);
+    const around = strip(body.slice(Math.max(0, m.index! - 600), m.index! + 1200));
+    entries.push({
+      ref: `/a/${sid}`,
+      title,
+      languages: guessLanguages(`${title} ${around}`),
+      format: formatOf(`${title} ${around}`),
+      downloads: null,
+      uploader: /发布人\s*(\S+)/.exec(around)?.[1] ?? null,
+      publishedAt: /<time datetime="([^"]+)"/.exec(body.slice(m.index!, m.index! + 2500))?.[1] ?? null,
+    });
+  }
+  const details: { id: string; title: string }[] = [];
+  for (const m of body.matchAll(new RegExp(`<a[^>]+${href("/d/(\\d+)")}[^>]*>([\\s\\S]*?)</a>`, "g"))) {
+    const alt = /alt=["']([^"']+)["']/.exec(m[2]!)?.[1];
+    const title = strip(m[2]!) || alt || "";
+    if (title && !details.some((d) => d.id === m[1])) details.push({ id: m[1]!, title });
+  }
+  return { entries, details };
+}
+
+/** 字幕页(/a/<sid>):下载按钮 /down/<sid> 或 data-sid(要先 prepare-download) */
+function subhdSub(html: string): { down: string | null; dataSid: string | null } {
+  return {
+    down: new RegExp(`${href("([^\"']*/down/[A-Za-z0-9_-]+[^\"']*)")}`).exec(html)?.[1] ?? null,
+    dataSid: /\bdata-sid=["']([A-Za-z0-9_-]+)["']/.exec(html)?.[1] ?? null,
+  };
+}
+
+// ── 字幕库 ──
+
+/** 字幕库语言旗帜的说明(「简体中文字幕」「双语字幕」…)→ 统一叫法 */
+function langLabel(t: string): string {
+  if (/双语/.test(t)) return "双语";
+  if (/繁/.test(t)) return "繁体";
+  if (/简/.test(t)) return "简体";
+  if (/英/.test(t)) return "英文";
+  if (/日/.test(t)) return "日文";
+  if (/韩|韓/.test(t)) return "韩文";
+  return t.replace(/字幕$/, "");
+}
+
+function zimukuSearch(html: string): { subs: { path: string; title: string }[] } {
+  const subs: { path: string; title: string }[] = [];
+  for (const m of html.matchAll(/<a[^>]+href=["']([^"']*\/subs\/\d+\.html)["'][^>]*>([\s\S]*?)<\/a>/g)) {
+    // 链接可能写成 //zimuku.org/subs/…(别的域名):只取路径,拼到站点地址上
+    const path = /\/subs\/\d+\.html/.exec(m[1]!)![0];
+    const title =
+      strip(m[2]!) || /title=["']([^"']+)["']/.exec(m[0])?.[1] || /alt=["']([^"']+)["']/.exec(m[2]!)?.[1] || "";
+    const prev = subs.find((s) => s.path === path);
+    if (!prev) subs.push({ path, title });
+    else if (!prev.title && title) prev.title = title;
+  }
+  return { subs };
+}
+
+/** 条目页(/subs/<id>.html)的字幕表 */
+function zimukuSubs(html: string): { entries: PageEntry[] } {
+  const entries: PageEntry[] = [];
+  for (const row of html.matchAll(/<tr[^>]*>([\s\S]*?)<\/tr>/g)) {
+    const r = row[1]!;
+    const a = /<a href=["'](\/detail\/\d+\.html)["'][^>]*title=["']([^"']+)["']/.exec(r);
+    if (!a) continue;
+    const langs = [...r.matchAll(/<img[^>]+title=["']([^"']+字幕)["']/g)].map((m) => langLabel(m[1]!));
+    const cells = [...r.matchAll(/<td[^>]*>([\s\S]*?)<\/td>/g)].map((m) => m[1]!);
+    entries.push({
+      ref: a[1]!,
+      title: strip(a[2]!),
+      languages: langs.length ? [...new Set(langs)] : guessLanguages(a[2]!),
+      format: /<span class="label label-info">([^<]+)<\/span>/.exec(r)?.[1]?.trim() ?? formatOf(a[2]!),
+      downloads: num(strip(cells[4] ?? "")),
+      uploader: /<a href=["']\/u\/[^"']+["']>([^<]+)<\/a>/.exec(r)?.[1] ?? null,
+      publishedAt: strip(/glyphicon-time"><\/span>([^<]+)</.exec(r)?.[1] ?? "") || null,
+    });
+  }
+  return { entries };
+}
+
+/** 字幕页:下载入口(li.dlsub 里的链接) */
+function zimukuDetail(html: string): { dl: string | null } {
+  return {
+    dl: /<li[^>]+class=["'][^"']*dlsub[^"']*["'][^>]*>[\s\S]*?<a[^>]+href=["']([^"']+)["']/.exec(html)?.[1] ?? null,
+  };
+}
+
+/** 下载跳转页:各线路的 /download/ 链接 */
+function zimukuDl(html: string): { links: string[] } {
+  return { links: [...new Set([...html.matchAll(/href=["']([^"']*\/download\/[^"']+)["']/g)].map((m) => m[1]!))] };
+}
+
+/** 是否是字幕库(云锁)的验证码页;是则带回 BMP */
+function zimukuCaptcha(html: string): { bmpBase64: string | null } {
+  return {
+    bmpBase64:
+      /class=["']verifyimg["'][^>]*src=["']data:image\/bmp;base64,([^"']+)["']/.exec(html)?.[1] ??
+      /data:image\/bmp;base64,([^"']+)/.exec(html)?.[1] ??
+      null,
+  };
+}
+
+export const PAGE_PARSERS = {
+  "subhd.list": subhdList,
+  "subhd.sub": subhdSub,
+  "zimuku.search": zimukuSearch,
+  "zimuku.subs": zimukuSubs,
+  "zimuku.detail": zimukuDetail,
+  "zimuku.dl": zimukuDl,
+  "zimuku.captcha": zimukuCaptcha,
+} as const;
+export type PageKind = keyof typeof PAGE_PARSERS;
+export type PageResult<K extends PageKind> = ReturnType<(typeof PAGE_PARSERS)[K]>;

apps/server/src/modules/subtitles/service.ts

@@ -269,7 +269,7 @@ export async function applySubtitles(
             library_root_id: t.libraryRootId,
             relative_path: rel,
             language: it.language,
-            source: body.source ?? "upload",
+            source: it.source ?? body.source ?? "upload",
             source_name: it.name,
             source_ref: it.sourceRef ?? null,
             backup_path: backup,

apps/server/src/sandbox/archives.ts

@@ -0,0 +1,70 @@
+// 压缩包解压(沙箱内):zip 用 fflate,rar 用 node-unrar-js(WASM),7z 用 7z-wasm。
+// WASM 库要读自己包里的 .wasm 文件:prepareArchives() 在撤销权限前预热(载入内存),之后只在内存里解压。
+import SevenZip from "7z-wasm";
+import { unzipSync } from "fflate";
+import { createExtractorFromData } from "node-unrar-js";
+
+type Seven = Awaited<ReturnType<typeof SevenZip>>;
+let seven: Seven | null = null;
+let warmed = false;
+
+export async function prepareArchives(): Promise<void> {
+  if (warmed) return;
+  warmed = true;
+  try {
+    await createExtractorFromData({ data: new Uint8Array([0x52, 0x61, 0x72, 0x21]).buffer });
+  } catch {
+    // 只为载入 WASM,内容无效报错是预期的
+  }
+  seven = await SevenZip({ print: () => {}, printErr: () => {} });
+}
+
+export type AcceptEntry = (name: string, size: number) => boolean;
+
+export function archiveKindOf(name: string): "zip" | "rar" | "7z" | null {
+  const m = /\.(zip|rar|7z)$/i.exec(name);
+  return m ? (m[1]!.toLowerCase() as "zip" | "rar" | "7z") : null;
+}
+
+/** 解出通过 accept 的条目;压缩包损坏 / 加密抛错 */
+export async function extractArchive(
+  kind: "zip" | "rar" | "7z",
+  bytes: Uint8Array,
+  accept: AcceptEntry,
+): Promise<Record<string, Uint8Array>> {
+  if (kind === "zip")
+    return unzipSync(bytes, { filter: (e) => !e.name.endsWith("/") && accept(e.name, e.originalSize) });
+  if (kind === "rar") {
+    const ex = await createExtractorFromData({ data: bytes.slice().buffer });
+    const out: Record<string, Uint8Array> = {};
+    for (const f of ex.extract({ files: (h) => !h.flags.directory && accept(h.name, h.unpSize) }).files)
+      if (f.extraction) out[f.fileHeader.name] = f.extraction;
+    return out;
+  }
+  await prepareArchives();
+  const z = seven!;
+  const dir = `/x${crypto.randomUUID().slice(0, 8)}`;
+  z.FS.writeFile(`${dir}.7z`, bytes);
+  // callMain 出错时抛 ExitStatus 或静默不写文件:两种都当成损坏处理
+  z.callMain(["x", `${dir}.7z`, `-o${dir}`, "-y", "-p-"]);
+  try {
+    z.FS.stat(dir);
+  } catch {
+    throw new Error("7z 解压失败");
+  }
+  const out: Record<string, Uint8Array> = {};
+  const walk = (d: string, rel: string) => {
+    for (const n of z.FS.readdir(d) as string[]) {
+      if (n === "." || n === "..") continue;
+      const p = `${d}/${n}`;
+      const r = rel ? `${rel}/${n}` : n;
+      if (z.FS.isDir(z.FS.stat(p).mode)) walk(p, r);
+      else {
+        const data = z.FS.readFile(p) as Uint8Array;
+        if (accept(r, data.length)) out[r] = data;
+      }
+    }
+  };
+  walk(dir, "");
+  return out;
+}

apps/server/src/sandbox/captcha.ts

@@ -0,0 +1,233 @@
+// SubHD 下载验证码(SVG 文字):每个字符一条长 path,按 path 长度查表,长度撞车时看最高点 / 起笔 / 宽度区分
+// (移植自 MP 插件 captcha/subhd.py 的长度表与判别规则)
+const LENGTH_MAP: Record<number, string> = {
+  986: "I",
+  998: "1",
+  1068: "I",
+  1081: "1",
+  1082: "v",
+  1130: "Y",
+  1134: "Y",
+  1172: "v",
+  1224: "Y",
+  1274: "L",
+  1298: "V",
+  1311: "V",
+  1360: "i",
+  1380: "L",
+  1406: "V",
+  1473: "i",
+  1478: "T",
+  1491: "r",
+  1598: "N",
+  1601: "T",
+  1604: "X",
+  1610: "J",
+  1613: "x",
+  1614: "N",
+  1615: "N",
+  1616: "N",
+  1617: "N",
+  1618: "N",
+  1634: "k",
+  1637: "k",
+  1694: "t",
+  1706: "K",
+  1709: "K",
+  1731: "N",
+  1744: "J",
+  1754: "F",
+  1770: "k",
+  1835: "t",
+  1838: "u",
+  1840: "A",
+  1844: "A",
+  1848: "K",
+  1850: "Z",
+  1853: "Z",
+  1886: "h",
+  1900: "F",
+  1922: "H",
+  1928: "H",
+  1960: "P",
+  1991: "u",
+  1993: "A",
+  1996: "D",
+  2004: "Z",
+  2018: "w",
+  2035: "w",
+  2042: "7",
+  2043: "h",
+  2080: "j",
+  2082: "H",
+  2104: "R",
+  2107: "R",
+  2123: "P",
+  2140: "4",
+  2162: "D",
+  2164: "O",
+  2183: "w",
+  2198: "C",
+  2199: "C",
+  2200: "C",
+  2201: "C",
+  2202: "C",
+  2210: "f",
+  2212: "7",
+  2246: "E",
+  2253: "j",
+  2260: "o",
+  2272: "d",
+  2279: "M",
+  2282: "M",
+  2294: "U",
+  2301: "U",
+  2310: "W",
+  2318: "4",
+  2321: "M",
+  2332: "a",
+  2344: "O",
+  2345: "W",
+  2346: "W",
+  2366: "s",
+  2380: "b",
+  2381: "C",
+  2382: "0",
+  2394: "f",
+  2433: "E",
+  2448: "o",
+  2461: "d",
+  2464: "p",
+  2466: "M",
+  2485: "U",
+  2498: "c",
+  2501: "e",
+  2503: "W",
+  2512: "q",
+  2526: "a",
+  2546: "2",
+  2563: "s",
+  2578: "b",
+  2580: "0",
+  2606: "5",
+  2632: "6",
+  2669: "p",
+  2706: "c",
+  2709: "e",
+  2721: "q",
+  2758: "2",
+  2800: "9",
+  2823: "5",
+  2851: "6",
+  3033: "9",
+  3038: "S",
+  3054: "B",
+  3160: "g",
+  3244: "Q",
+  3254: "Q",
+  3266: "G",
+  3291: "S",
+  3308: "B",
+  3414: "8",
+  3423: "g",
+  3514: "Q",
+  3538: "G",
+  3663: "m",
+  3667: "m",
+  3698: "8",
+  3878: "3",
+  3968: "m",
+  4201: "3",
+};
+const nums = (p: string) => (p.match(/\d+(?:\.\d*)?/g) ?? []).map(Number);
+const minY = (p: string) => {
+  const v = nums(p).filter((_, i) => i % 2 === 1);
+  return v.length ? Math.min(...v) : 0;
+};
+const moveY = (p: string) => Number(/M(\d+(?:\.\d*)?)\s+(\d+(?:\.\d*)?)/.exec(p)?.[2] ?? 0);
+const width = (p: string) => {
+  const v = nums(p).filter((_, i) => i % 2 === 0);
+  return v.length ? Math.max(...v) - Math.min(...v) : 0;
+};
+function collision(len: number, p: string): string {
+  if (len === 986 || len === 1068) return minY(p) > 13 ? "I" : "l";
+  if (len === 1274 || len === 1380) return moveY(p) > 30 ? "y" : "L";
+  if (len === 1610 || len === 1744) return minY(p) > 19 ? "x" : "J";
+  if (len === 1615) return minY(p) > 18 ? "r" : "N";
+  if (len === 2198 || len === 2381) return minY(p) > 19 ? "n" : "C";
+  if (len === 2318) return width(p) > 30 ? "W" : "4";
+  if (len === 1598 || len === 1731) return minY(p) > 13 ? "X" : "N";
+  if (len === 1694 || len === 1835) return minY(p) > 22 ? "z" : "t";
+  if (len === 2279) return minY(p) > 13 ? "R" : "M";
+  return "";
+}
+export function solveSubhdCaptcha(svg: string): string {
+  const paths: [number, string][] = [];
+  for (const m of svg.matchAll(/d=["']([^"']+)["']/g)) {
+    const p = m[1]!;
+    if (p.length <= 500) continue;
+    paths.push([Number(/\d+(?:\.\d*)?/.exec(p)?.[0] ?? 0), p]);
+  }
+  return paths
+    .sort((a, b) => a[0] - b[0])
+    .map(([, p]) => collision(p.length, p) || LENGTH_MAP[p.length] || "")
+    .join("");
+}
+// 字幕库(云锁)BMP 数字验证码:100×27、5 位、每位 20 像素宽;按 9 个采样点比对 0-9 模板(移植自 MP 插件 captcha/zimuku.py)
+const W = 100,
+  H = 27,
+  CHAR_W = 20,
+  N = 5,
+  OFFSET = 54;
+const SAMPLES: [number, number][] = [
+  [10, 7],
+  [7, 8],
+  [12, 8],
+  [10, 13],
+  [7, 19],
+  [12, 19],
+  [10, 20],
+  [6, 13],
+  [14, 13],
+];
+const TEMPLATES: Record<string, number[]> = {
+  "0": [1, 1, 1, 1, 1, 1, 1, 1, 0],
+  "1": [0, 1, 0, 0, 0, 0, 1, 0, 0],
+  "2": [1, 0, 1, 0, 1, 0, 1, 0, 0],
+  "3": [1, 0, 1, 1, 0, 1, 1, 0, 0],
+  "4": [0, 0, 1, 0, 0, 1, 0, 0, 0],
+  "5": [1, 1, 0, 0, 0, 1, 1, 0, 0],
+  "6": [1, 0, 1, 1, 1, 1, 1, 1, 0],
+  "7": [1, 0, 1, 0, 0, 0, 0, 0, 0],
+  "8": [1, 1, 1, 1, 1, 1, 1, 0, 0],
+  "9": [1, 1, 1, 0, 1, 0, 1, 0, 0],
+};
+export function solveZimukuCaptcha(bmp: Uint8Array): string | null {
+  if (bmp.length < OFFSET || bmp[0] !== 0x42 || bmp[1] !== 0x4d) return null;
+  const dv = new DataView(bmp.buffer, bmp.byteOffset, bmp.byteLength);
+  if (dv.getInt32(18, true) !== W || dv.getInt32(22, true) !== H) return null;
+  const stride = (W * 3 + 3) & ~3;
+  const fg = (x: number, y: number) => {
+    const o = OFFSET + (H - 1 - y) * stride + x * 3;
+    if (o + 2 >= bmp.length || x < 0) return false;
+    return (bmp[o]! + bmp[o + 1]! + bmp[o + 2]!) / 3 < 70;
+  };
+  let out = "",
+    shift = 0;
+  for (let i = 0; i < N; i++) {
+    const f = SAMPLES.map(([px, py]) => (fg(i * CHAR_W + px - shift, py) ? 1 : 0));
+    let best = "?",
+      min = 99;
+    for (const [d, t] of Object.entries(TEMPLATES)) {
+      const diff = t.reduce((n, v, k) => n + (v !== f[k] ? 1 : 0), 0);
+      if (diff < min) {
+        min = diff;
+        best = d;
+      }
+    }
+    if (best === "1") shift += 1;
+    else if (best === "4") shift -= 1;
+    out += best;
+  }
+  return out.includes("?") ? null : out;
+}

apps/server/src/sandbox/client.ts

@@ -3,7 +3,7 @@
 // 起不来(权限 / 运行时不支持)时退回进程内执行并记日志:仍有超时,但卡死的同步代码无法收回(只在本机调试时出现)。
 import { basename } from "node:path";
 import { fileURLToPath } from "node:url";
-import { SANDBOX_JOBS, type SandboxInput, type SandboxJobName, type SandboxOutput } from "./jobs.ts";
+import { prepareJobs, SANDBOX_JOBS, type SandboxInput, type SandboxJobName, type SandboxOutput } from "./jobs.ts";
 
 export type SandboxErrorKind = "timeout" | "memory" | "crash" | "job" | "spawn";
 
@@ -144,6 +144,7 @@ async function runInProcess<J extends SandboxJobName>(
   input: SandboxInput<J>,
   timeoutMs: number,
 ): Promise<unknown> {
+  await prepareJobs();
   const fn = SANDBOX_JOBS[job] as (i: SandboxInput<J>) => unknown;
   let timer: ReturnType<typeof setTimeout> | undefined;
   try {

apps/server/src/sandbox/jobs.ts

@@ -3,8 +3,10 @@
 // 输入输出必须能 JSON 化(二进制用 base64)。新增任务:在这里加一项,调用方用 runSandbox(名字, 输入)。
 import { parseTorrent } from "@umt/adapters";
 import { decodeSubtitle, detectSubtitleLanguage, subtitleEpisode, subtitleExtOf } from "@umt/domain";
-import { unzipSync } from "fflate";
 import { buildPreview } from "../modules/file-manager/preview-build.ts";
+import { PAGE_PARSERS, type PageKind } from "../modules/subtitles/online/parse.ts";
+import { archiveKindOf, extractArchive, prepareArchives } from "./archives.ts";
+import { solveSubhdCaptcha, solveZimukuCaptcha } from "./captcha.ts";
 
 const MAX_SUBTITLE_BYTES = 10 * 1024 * 1024;
 const MAX_ARCHIVE_ENTRIES = 300;
@@ -30,8 +32,13 @@ export interface ParsedSubtitleFile {
   content: string;
 }
 
-/** 字幕 / zip:解压、解码转 UTF-8、识别语言与季集 */
-function subtitleParse(input: { files: { name: string; base64: string }[] }) {
+/** 子进程撤销权限前调用:预热要读文件的 WASM 库 */
+export async function prepareJobs(): Promise<void> {
+  await prepareArchives();
+}
+
+/** 字幕 / zip / rar / 7z:解压、解码转 UTF-8、识别语言与季集 */
+async function subtitleParse(input: { files: { name: string; base64: string }[] }) {
   const items: ParsedSubtitleFile[] = [];
   const skipped: { name: string; reason: string }[] = [];
   const one = (name: string, bytes: Uint8Array, archive: string | null) => {
@@ -61,17 +68,16 @@ function subtitleParse(input: { files: { name: string; base64: string }[] }) {
       skipped.push({ name: f.name, reason: "内容不是有效的 base64" });
       continue;
     }
-    if (/\.zip$/i.test(f.name)) {
+    const kind = archiveKindOf(f.name);
+    if (kind) {
       let entries: Record<string, Uint8Array>;
       try {
         let n = 0;
-        entries = unzipSync(bytes, {
-          filter: (e) =>
-            !e.name.endsWith("/") &&
-            !!subtitleExtOf(e.name) &&
-            e.originalSize <= MAX_SUBTITLE_BYTES &&
-            ++n <= MAX_ARCHIVE_ENTRIES,
-        });
+        entries = await extractArchive(
+          kind,
+          bytes,
+          (name, size) => !!subtitleExtOf(name) && size <= MAX_SUBTITLE_BYTES && ++n <= MAX_ARCHIVE_ENTRIES,
+        );
       } catch {
         skipped.push({ name: f.name, reason: "压缩包损坏或加密" });
         continue;
@@ -79,8 +85,6 @@ function subtitleParse(input: { files: { name: string; base64: string }[] }) {
       const names = Object.keys(entries);
       if (!names.length) skipped.push({ name: f.name, reason: "压缩包里没有字幕文件" });
       for (const n of names) one(n, entries[n]!, f.name);
-    } else if (/\.(rar|7z)$/i.test(f.name)) {
-      skipped.push({ name: f.name, reason: "暂不支持 rar / 7z，请解压后上传或改用 zip" });
     } else one(f.name, bytes, null);
   }
   return { items, skipped };
@@ -103,10 +107,26 @@ function selfTest(input: { mode: "ok" | "hang" | "memory" | "throw" | "io" }) {
   return { ok: true };
 }
 
+/** SubHD SVG / 字幕库 BMP 验证码 */
+function captchaSolve(input: { kind: "subhd"; svg: string } | { kind: "zimuku"; bmpBase64: string }) {
+  return {
+    code: input.kind === "subhd" ? solveSubhdCaptcha(input.svg) : solveZimukuCaptcha(b64decode(input.bmpBase64)),
+  };
+}
+
+/** 字幕站页面 → 结构化数据(网页同样不可信) */
+function subtitlePage(input: { kind: PageKind; html: string }) {
+  const fn = PAGE_PARSERS[input.kind] as (html: string) => unknown;
+  if (!fn) throw new Error(`没有这种页面解析：${input.kind}`);
+  return fn(input.html);
+}
+
 export const SANDBOX_JOBS = {
   "subtitle.parse": subtitleParse,
   "torrent.parse": torrentParse,
   "file.preview": buildPreview,
+  "captcha.solve": captchaSolve,
+  "subtitle.page": subtitlePage,
   "self.test": selfTest,
 } as const;
 

apps/server/src/sandbox/worker.ts

@@ -1,6 +1,6 @@
 // 沙箱子进程入口:先加载任务表,再撤销全部权限,然后从 stdin 读一条 {job, input},把 {ok, result | error} 写到 stdout。
 // 开发 / 测试由 deno run 直接跑本文件;编译二进制以 `<自身> --role sandbox` 拉起(main.ts 在加载其余模块前分派)。
-import { SANDBOX_JOBS } from "./jobs.ts";
+import { prepareJobs, SANDBOX_JOBS } from "./jobs.ts";
 
 const PERMISSIONS = ["read", "write", "net", "env", "run", "ffi", "sys"] as const;
 
@@ -10,6 +10,7 @@ async function writeAll(bytes: Uint8Array) {
 }
 
 export async function sandboxWorkerMain(): Promise<void> {
+  await prepareJobs();
   for (const name of PERMISSIONS) {
     try {
       Deno.permissions.revokeSync({ name } as Deno.PermissionDescriptor);

apps/server/test/sandbox/sandbox.test.ts

@@ -1,4 +1,4 @@
-import { buildFakeTorrent } from "@umt/testkit";
+import { buildFakeTorrent, RAR5_SAMPLE_BASE64, SEVENZIP_SUBTITLE_BASE64, ZIMUKU_CAPTCHA } from "@umt/testkit";
 import { zipSync } from "fflate";
 import { describe, expect, it } from "vitest";
 import { runSandbox, SandboxError } from "../../src/sandbox/client.ts";
@@ -48,4 +48,36 @@ describe("沙箱子进程", () => {
     expect(f.files.map((x) => x.path)).toEqual(["Show.S01E01.mkv"]);
     expect(f.hash).toBe(t.infoHash);
   });
+
+  it("rar / 7z 在沙箱里解压(撤销权限后 WASM 仍可用);非字幕条目跳过", async () => {
+    const r = await runSandbox("subtitle.parse", {
+      files: [
+        { name: "pack.7z", base64: SEVENZIP_SUBTITLE_BASE64 },
+        { name: "texts.rar", base64: RAR5_SAMPLE_BASE64 },
+      ],
+    });
+    expect(r.items).toEqual([
+      expect.objectContaining({ name: "s7/a.chs.srt", archive: "pack.7z", language: "zh-Hans" }),
+    ]);
+    expect(r.skipped).toEqual([{ name: "texts.rar", reason: "压缩包里没有字幕文件" }]);
+  });
+
+  it("字幕库验证码:真实样本识别正确", async () => {
+    expect(await runSandbox("captcha.solve", { kind: "zimuku", bmpBase64: ZIMUKU_CAPTCHA.bmpBase64 })).toEqual({
+      code: ZIMUKU_CAPTCHA.code,
+    });
+  });
+});
+
+describe("压缩包解压(进程内,直接测解压模块)", () => {
+  it("rar 列出全部条目", async () => {
+    const { extractArchive } = await import("../../src/sandbox/archives.ts");
+    const out = await extractArchive(
+      "rar",
+      Uint8Array.from(atob(RAR5_SAMPLE_BASE64), (c) => c.charCodeAt(0)),
+      () => true,
+    );
+    expect(Object.keys(out).sort()).toEqual(["stest1.txt", "stest2.txt"]);
+    await expect(extractArchive("rar", new Uint8Array([1, 2, 3]), () => true)).rejects.toThrow();
+  });
 });

apps/server/test/sources-identity/site-kinds.test.ts

@@ -129,7 +129,15 @@ describe.skipIf(!up)("站点类型:API、调度与 RSS", () => {
         .map((d) => d.key)
         .sort(),
     ).toEqual([...pub].sort());
-    expect(items.filter((d) => !pub.includes(d.key)).every((d) => d.kind === "pt")).toBe(true);
+    // 内置字幕站(SubHD、字幕库)也在定义列表里,类型 subtitle
+    const subs = ["subhd", "zimuku"];
+    expect(
+      items
+        .filter((d) => d.kind === "subtitle")
+        .map((d) => d.key)
+        .sort(),
+    ).toEqual(subs);
+    expect(items.filter((d) => !pub.includes(d.key) && !subs.includes(d.key)).every((d) => d.kind === "pt")).toBe(true);
     const nyaa = await json(
       await h.req("/sites", { body: { key: "nyaa", name: "Nyaa", adapter: "api", baseUrl: "https://nyaa.si/" } }),
     );
@@ -203,4 +211,23 @@ describe.skipIf(!up)("站点类型:API、调度与 RSS", () => {
       reason: "unsupported",
     });
   });
+
+  it("字幕站:只能从内置字幕站添加;不排快照、不进搜索、能力为空", async () => {
+    const ok = await json(
+      await h.req("/sites", {
+        body: { kind: "subtitle", key: "zimuku", name: "字幕库", adapter: "subtitle", baseUrl: "https://zmk.pw/" },
+      }),
+    );
+    expect(ok.status).toBe(200);
+    expect(ok.body.data).toMatchObject({
+      kind: "subtitle",
+      snapshotEnabled: false,
+      searchDefault: false,
+      capabilities: [],
+    });
+    const bad = await h.req("/sites", {
+      body: { kind: "subtitle", key: "random", name: "x", adapter: "subtitle", baseUrl: "https://x.example/" },
+    });
+    expect(bad.status).toBe(400);
+  });
 });

apps/server/test/subtitles/subtitle-online.test.ts

@@ -0,0 +1,122 @@
+import {
+  SUBHD_DETAIL_PAGE,
+  SUBHD_SUB_PAGE,
+  ZIMUKU_CAPTCHA,
+  ZIMUKU_CAPTCHA_PAGE,
+  ZIMUKU_DETAIL_PAGE,
+  ZIMUKU_DL_PAGE,
+  ZIMUKU_SEARCH_PAGE,
+  ZIMUKU_SUBS_PAGE,
+} from "@umt/testkit";
+import { zipSync } from "fflate";
+import { afterAll, beforeAll, describe, expect, it } from "vitest";
+import { downloadOnline, searchOnline } from "../../src/modules/subtitles/online/index.ts";
+import { createHarness, type Harness, pgAvailable } from "../sources-identity/harness.ts";
+
+// 在线字幕(docs/plan/subtitles.md 第 2 步):站点请求全部走注入的 fetch,页面是 2026-10-03 的真实截取;
+// 字幕库先回验证码页,沙箱识别后带 security_verify_img 提交,再开搜索页;SubHD 走 prepare-download → down → api/sub/down
+const pg = await pgAvailable();
+const enc = (s: string) => new TextEncoder().encode(s);
+
+describe.skipIf(!pg)("在线字幕:SubHD / 字幕库(注入 fetch)", () => {
+  let h: Harness;
+  const requests: string[] = [];
+  let verified = false;
+  const siteIds: Record<string, string> = {};
+  const html = (body: string, status = 200) =>
+    new Response(body, { status, headers: { "content-type": "text/html; charset=utf-8" } });
+  const zip = zipSync({
+    "The.Shawshank.Redemption.1994.chs&eng.ass": enc(
+      "[Script Info]\nDialogue: 0,0:00:01.00,0:00:02.00,Default,,0,0,0,,你好 Hello",
+    ),
+  });
+
+  beforeAll(async () => {
+    h = await createHarness("subonline", {
+      fetch: async (input, init) => {
+        const u = new URL(typeof input === "string" ? input : input instanceof URL ? input.href : input.url);
+        const method = init?.method ?? "GET";
+        requests.push(`${method} ${u.host}${u.pathname}${u.search}`);
+        if (u.host === "zmk.pw") {
+          if (u.searchParams.has("security_verify_img")) {
+            const want = [...ZIMUKU_CAPTCHA.code].map((c) => c.charCodeAt(0).toString(16)).join("");
+            verified = u.searchParams.get("security_verify_img") === want;
+            return html("ok", 302);
+          }
+          if (!verified) return html(ZIMUKU_CAPTCHA_PAGE, 404);
+          if (u.pathname === "/search") return html(ZIMUKU_SEARCH_PAGE);
+          if (/^\/subs\/\d+\.html$/.test(u.pathname)) return html(ZIMUKU_SUBS_PAGE);
+          if (u.pathname === "/detail/203404.html") return html(ZIMUKU_DETAIL_PAGE);
+          if (u.pathname === "/dld/203404.html") return html(ZIMUKU_DL_PAGE);
+          if (u.pathname.startsWith("/download/"))
+            return new Response(zip, {
+              headers: {
+                "content-type": "application/octet-stream",
+                "content-disposition": 'attachment; filename="sub.zip"',
+              },
+            });
+        }
+        if (u.host === "subhd.tv") {
+          if (u.pathname === "/d/1292052") return html(SUBHD_DETAIL_PAGE);
+          if (u.pathname.startsWith("/a/")) return html(SUBHD_SUB_PAGE);
+          if (u.pathname === "/api/sub/prepare-download") return Response.json({ success: true, url: "/down/Fp90EG" });
+          if (u.pathname === "/down/Fp90EG") return html("<html></html>");
+          if (u.pathname === "/api/sub/down")
+            return Response.json({ success: true, pass: true, url: "https://dl.subhd.example/x/1.srt" });
+        }
+        if (u.host === "dl.subhd.example")
+          return new Response(enc("1\n00:00:01,000 --> 00:00:02,000\n你好\n"), {
+            headers: { "content-disposition": 'attachment; filename="Shawshank.chs.srt"' },
+          });
+        return html("not found", 404);
+      },
+    });
+    for (const [key, name, url] of [
+      ["subhd", "SubHD", "https://subhd.tv/"],
+      ["zimuku", "字幕库", "https://zmk.pw/"],
+    ] as const) {
+      const r = await h.db
+        .insertInto("site")
+        .values({ key, name, kind: "subtitle", adapter: "subtitle", base_url: url } as never)
+        .returning("id")
+        .executeTakeFirstOrThrow();
+      siteIds[key] = r.id;
+    }
+  }, 120_000);
+  afterAll(async () => {
+    await h?.close();
+  });
+
+  // 字幕库样例是《葬送的芙莉莲》的搜索结果;SubHD 按豆瓣 ID 开条目页,不按标题过滤
+  const q = { keywords: ["葬送的芙莉莲"], doubanId: "1292052", season: null, episode: null };
+
+  it("搜索:SubHD 有豆瓣 ID 直接开条目页;字幕库先过验证码再搜;结果按语言与下载量排序", async () => {
+    const r = await searchOnline(h.ctx, h.rt, q);
+    expect(r.errors).toEqual([]);
+    expect(r.noSites).toBe(false);
+    expect(verified).toBe(true);
+    expect(requests).toContain("GET subhd.tv/d/1292052");
+    const zim = r.results.filter((x) => x.siteKey === "zimuku");
+    const sub = r.results.filter((x) => x.siteKey === "subhd");
+    expect(zim.length).toBeGreaterThan(0);
+    expect(sub.length).toBeGreaterThan(0);
+    // 双语排在前面
+    expect(r.results[0]!.languages).toContain("双语");
+    expect(zim.find((x) => x.ref === "/detail/203404.html")).toMatchObject({ siteName: "字幕库", downloads: 17000 });
+  });
+
+  it("下载:字幕库走下载页取文件;SubHD 走 prepare-download → api/sub/down", async () => {
+    const z = await downloadOnline(h.ctx, h.rt, siteIds.zimuku!, "/detail/203404.html");
+    expect(z).toMatchObject({ name: "sub.zip", siteKey: "zimuku", pageUrl: "https://zmk.pw/detail/203404.html" });
+    const s = await downloadOnline(h.ctx, h.rt, siteIds.subhd!, "/a/Fp90EG");
+    expect(s.name).toBe("Shawshank.chs.srt");
+    expect(requests).toContain("POST subhd.tv/api/sub/prepare-download");
+  });
+
+  it("字幕站停用:不搜;下载报错", async () => {
+    await h.db.updateTable("site").set({ enabled: false }).where("kind", "=", "subtitle").execute();
+    const r = await searchOnline(h.ctx, h.rt, q);
+    expect(r.noSites).toBe(true);
+    await expect(downloadOnline(h.ctx, h.rt, siteIds.subhd!, "/a/x")).rejects.toThrow(/已停用/);
+  });
+});

apps/web/app/api/subtitles.ts

@@ -2,7 +2,10 @@ import { useMutation, useQuery, useQueryClient } from "@tanstack/react-query";
 import {
   type SubtitleApplyRequest,
   SubtitleApplyResult,
+  type SubtitleFetchRequest,
   SubtitleParseResponse,
+  type SubtitleSearchRequest,
+  SubtitleSearchResponse,
   SubtitleTargetsResponse,
   SubtitleWritesResponse,
   SubtitleWriteView,
@@ -82,3 +85,18 @@ export function useUndoSubtitle(entityId: string) {
     },
   });
 }
+
+export function useSearchSubtitles() {
+  return useMutation({
+    mutationFn: async (body: SubtitleSearchRequest) =>
+      parseOrThrow(SubtitleSearchResponse, await api.post<unknown>("/subtitles/search", body), "/subtitles/search"),
+  });
+}
+
+/** 下载一条在线字幕并解析(同上传的预览,不写文件) */
+export function useFetchSubtitle() {
+  return useMutation({
+    mutationFn: async (body: SubtitleFetchRequest) =>
+      parseOrThrow(SubtitleParseResponse, await api.post<unknown>("/subtitles/fetch", body), "/subtitles/fetch"),
+  });
+}

apps/web/app/components/sites/site-add-dialog.tsx

@@ -6,7 +6,7 @@ import {
   type SiteListItem,
   type SiteLoginResult,
 } from "@umt/contracts";
-import { ArrowLeft, Check, Globe, LoaderCircle, Rss, Search, Server, ShieldCheck } from "lucide-react";
+import { ArrowLeft, Captions, Check, Globe, LoaderCircle, Rss, Search, Server, ShieldCheck } from "lucide-react";
 import { useMemo, useState } from "react";
 import { useNavigate } from "react-router";
 import { toast } from "sonner";
@@ -48,8 +48,8 @@ const PT_STEPS = ["选择类型", "选择站点", "填写凭证", "测试"] as c
 const FEED_STEPS = ["选择类型", "选择站点"] as const;
 const GROUP_ORDER = ["nexusphp", "mteam_api", "api", "generic_rss"] as const;
 /** 自定义 RSS 源与索引器入口暂时不开放(已有的 RSS 源站点照常显示) */
-const KIND_ORDER: SiteKind[] = ["pt", "public_bt"];
-const KIND_ICON = { pt: ShieldCheck, public_bt: Globe, rss_feed: Rss, torznab: Server } as const;
+const KIND_ORDER: SiteKind[] = ["pt", "public_bt", "subtitle"];
+const KIND_ICON = { pt: ShieldCheck, public_bt: Globe, rss_feed: Rss, torznab: Server, subtitle: Captions } as const;
 
 /** 馒头默认 API Key,其余默认 Cookie(验证码站点账密大概率失败) */
 function defaultKind(kinds: LoginKind[]): LoginKind {
@@ -105,13 +105,14 @@ export function SiteAddDialog({ onClose, initialKind }: { onClose: () => void; i
     })).filter((g) => g.items.length);
   }, [items, q]);
 
-  const publicDefs = items.filter((d) => d.kind === "public_bt");
+  // 公开 BT 站 / 字幕站:都是内置站,一键添加
+  const publicDefs = items.filter((d) => d.kind === kind);
   /** 内置公开 BT 站:一步建好(专用适配器、定时抓最新),进站点页 */
   const addPublic = async (d: SiteDefinitionView) => {
     try {
       setPhase("保存中…");
       const site = await create.mutateAsync({
-        kind: "public_bt",
+        kind: d.kind,
         key: d.key,
         adapter: d.adapter,
         name: d.name,
@@ -187,7 +188,9 @@ export function SiteAddDialog({ onClose, initialKind }: { onClose: () => void; i
         <LoadingState rows={2} />
       ) : null}
       <p className="mt-3 text-meta text-muted-foreground">
-        不需要账号；添加后默认走代理，定时抓取最新资源，也能在资源搜索里选用。
+        {kind === "subtitle"
+          ? "不需要账号；在作品页「补字幕」里搜索、下载。"
+          : "不需要账号；添加后默认走代理，定时抓取最新资源，也能在资源搜索里选用。"}
       </p>
     </div>
   );

apps/web/app/components/sites/site-card.tsx

@@ -336,7 +336,46 @@ function PublicSiteCard({ site, onDialog }: Omit<Props, "checks">) {
   );
 }
 
+/** 字幕站(SubHD、字幕库):出口与状态;搜索 / 下载在作品页「补字幕」里 */
+function SubtitleSiteCard({ site, checks, onDialog }: Props) {
+  const health = healthLine(site.lastHealth, site.lastHealthAt);
+  const busy = checks.isBusy(site.id, "health");
+  return (
+    <article className={cn("pg-slab flex min-w-0 flex-col rounded-xl p-4", !site.enabled && "opacity-75")}>
+      <SiteCardHeader site={site} />
+      <dl className="mt-3 grid grid-cols-2 gap-x-4 gap-y-2 rounded-lg border border-border/60 px-3 py-2.5 text-caption">
+        <div className="min-w-0">
+          <dt className="text-meta text-muted-foreground">出口</dt>
+          <dd className="text-foreground">{site.proxy ? "代理" : "直连"}</dd>
+        </div>
+        <div className="min-w-0">
+          <dt className="text-meta text-muted-foreground">状态</dt>
+          <dd className="truncate" title={healthMessage(site.lastHealth) ?? undefined}>
+            <StatusDot tone={health.tone} />
+            {health.text}
+          </dd>
+        </div>
+      </dl>
+      <p className="mt-2 text-meta text-muted-foreground">在作品页「补字幕」里搜索、下载。</p>
+      <div aria-hidden className="min-h-4 flex-1" />
+      <footer className="mt-auto flex flex-wrap items-center justify-end gap-2 border-t border-border/60 pt-3">
+        <Button
+          variant="outline"
+          size="sm"
+          disabled={busy || !site.enabled}
+          onClick={() => void checks.checkHealth(site)}
+        >
+          {busy ? <LoaderCircle className="animate-spin" /> : <Activity />}
+          检测健康
+        </Button>
+        <SiteMoreMenu site={site} onDialog={onDialog} />
+      </footer>
+    </article>
+  );
+}
+
 export function SiteCard({ site, checks, onDialog }: Props) {
+  if (site.kind === "subtitle") return <SubtitleSiteCard site={site} checks={checks} onDialog={onDialog} />;
   if (!hasLogin(site) && siteCan(site, "list")) return <PublicSiteCard site={site} onDialog={onDialog} />;
   if (!hasLogin(site)) return <FeedSiteCard site={site} onDialog={onDialog} />;
   return <AccountSiteCard site={site} checks={checks} onDialog={onDialog} />;
@@ -480,8 +519,8 @@ export function SiteTable({
   sites: SiteListItem[];
   checks: Checks;
   onDialog: (d: SiteDialog) => void;
-  /** pt = 带账号与流量列;public = 公开站(无账号),只列状态与资源更新 */
-  variant?: "pt" | "public";
+  /** pt = 带账号与流量列;public = 公开站(无账号),只列状态与资源更新;subtitle = 字幕站,只列出口与状态 */
+  variant?: "pt" | "public" | "subtitle";
 }) {
   const [params] = useSearchParams();
   const name: DataColumn<SiteListItem> = {
@@ -530,8 +569,22 @@ export function SiteTable({
       </div>
     ),
   };
+  const health: DataColumn<SiteListItem> = {
+    key: "health",
+    header: "状态",
+    hideBelow: "sm",
+    cell: (s) => {
+      const h = healthLine(s.lastHealth, s.lastHealthAt);
+      return (
+        <span className="whitespace-nowrap text-caption" title={healthMessage(s.lastHealth) ?? h.text}>
+          <StatusDot tone={h.tone} />
+          {h.text}
+        </span>
+      );
+    },
+  };
   const columns: DataColumn<SiteListItem>[] =
-    variant === "public"
+    variant === "subtitle"
       ? [
           name,
           {
@@ -540,94 +593,108 @@ export function SiteTable({
             hideBelow: "sm",
             cell: (s) => <span className="whitespace-nowrap text-muted-foreground">{s.proxy ? "代理" : "直连"}</span>,
           },
-          {
-            key: "search",
-            header: "资源搜索",
-            hideBelow: "sm",
-            cell: (s) => (
-              <span className="whitespace-nowrap text-muted-foreground">{s.searchDefault ? "默认参与" : "手动选"}</span>
-            ),
-          },
-          snapshot,
+          health,
           actions,
         ]
-      : [
-          name,
-          {
-            key: "account",
-            header: "账号",
-            hideBelow: "xl",
-            cell: (s) => (
-              <div className="flex flex-wrap items-center gap-1">
-                <AccountBadges site={s} />
-              </div>
-            ),
-          },
-          {
-            key: "login",
-            header: "上次登录",
-            hideBelow: "xl",
-            cell: (s) => (
-              <span className="whitespace-nowrap text-muted-foreground">
-                {relTime(primaryAccount(s.accounts)?.lastLoginAt)}
-              </span>
-            ),
-          },
-          {
-            key: "traffic",
-            header: "上传 / 下载",
-            hideBelow: "md",
-            cell: (s) => {
-              const u = dataAccount(s.accounts)?.userData;
-              return u ? (
-                <span className="whitespace-nowrap tabular-nums">
-                  {bytes(u.upload)} / {bytes(u.download)}
-                </span>
-              ) : (
-                "—"
-              );
+      : variant === "public"
+        ? [
+            name,
+            {
+              key: "proxy",
+              header: "出口",
+              hideBelow: "sm",
+              cell: (s) => <span className="whitespace-nowrap text-muted-foreground">{s.proxy ? "代理" : "直连"}</span>,
             },
-          },
-          {
-            key: "ratio",
-            header: "分享率",
-            align: "right",
-            cell: (s) => (
-              <span className="whitespace-nowrap tabular-nums">{ratioText(dataAccount(s.accounts)?.userData)}</span>
-            ),
-          },
-          {
-            key: "level",
-            header: "等级 · 做种",
-            hideBelow: "lg",
-            cell: (s) => {
-              const u = dataAccount(s.accounts)?.userData;
-              const t = u ? `${u.userLevel ?? "—"} · ${u.seeding}` : "—";
-              return (
-                <span className="block max-w-40 truncate whitespace-nowrap" title={t}>
-                  {t}
+            {
+              key: "search",
+              header: "资源搜索",
+              hideBelow: "sm",
+              cell: (s) => (
+                <span className="whitespace-nowrap text-muted-foreground">
+                  {s.searchDefault ? "默认参与" : "手动选"}
                 </span>
-              );
+              ),
             },
-          },
-          {
-            key: "health",
-            header: "状态",
-            hideBelow: "md",
-            cell: (s) => {
-              if (!hasLogin(s)) return <span className="text-muted-foreground">—</span>;
-              const h = healthLine(s.lastHealth, s.lastHealthAt);
-              return (
-                <span className="whitespace-nowrap text-caption" title={healthMessage(s.lastHealth) ?? h.text}>
-                  <StatusDot tone={h.tone} />
-                  {h.text}
+            snapshot,
+            actions,
+          ]
+        : [
+            name,
+            {
+              key: "account",
+              header: "账号",
+              hideBelow: "xl",
+              cell: (s) => (
+                <div className="flex flex-wrap items-center gap-1">
+                  <AccountBadges site={s} />
+                </div>
+              ),
+            },
+            {
+              key: "login",
+              header: "上次登录",
+              hideBelow: "xl",
+              cell: (s) => (
+                <span className="whitespace-nowrap text-muted-foreground">
+                  {relTime(primaryAccount(s.accounts)?.lastLoginAt)}
                 </span>
-              );
+              ),
             },
-          },
-          snapshot,
-          actions,
-        ];
+            {
+              key: "traffic",
+              header: "上传 / 下载",
+              hideBelow: "md",
+              cell: (s) => {
+                const u = dataAccount(s.accounts)?.userData;
+                return u ? (
+                  <span className="whitespace-nowrap tabular-nums">
+                    {bytes(u.upload)} / {bytes(u.download)}
+                  </span>
+                ) : (
+                  "—"
+                );
+              },
+            },
+            {
+              key: "ratio",
+              header: "分享率",
+              align: "right",
+              cell: (s) => (
+                <span className="whitespace-nowrap tabular-nums">{ratioText(dataAccount(s.accounts)?.userData)}</span>
+              ),
+            },
+            {
+              key: "level",
+              header: "等级 · 做种",
+              hideBelow: "lg",
+              cell: (s) => {
+                const u = dataAccount(s.accounts)?.userData;
+                const t = u ? `${u.userLevel ?? "—"} · ${u.seeding}` : "—";
+                return (
+                  <span className="block max-w-40 truncate whitespace-nowrap" title={t}>
+                    {t}
+                  </span>
+                );
+              },
+            },
+            {
+              key: "health",
+              header: "状态",
+              hideBelow: "md",
+              cell: (s) => {
+                if (!hasLogin(s)) return <span className="text-muted-foreground">—</span>;
+                const h = healthLine(s.lastHealth, s.lastHealthAt);
+                return (
+                  <span className="whitespace-nowrap text-caption" title={healthMessage(s.lastHealth) ?? h.text}>
+                    <StatusDot tone={h.tone} />
+                    {h.text}
+                  </span>
+                );
+              },
+            },
+            snapshot,
+            actions,
+          ];
   // 站点数量不多:分组表格不分页
   return (
     <DataTable

apps/web/app/components/sites/site-fields.ts

@@ -24,6 +24,7 @@ export const ADAPTER_LABEL: Record<SiteAdapter, string> = {
   mteam_api: "M-Team API",
   api: "站点 API",
   generic_rss: "通用 RSS",
+  subtitle: "字幕站",
 };
 
 /** 站点类型名(IA-2);旧数据缺 kind 时按 PT 站 */

apps/web/app/components/subtitles/subtitle-dialog.tsx

@@ -1,10 +1,18 @@
-import type { SubtitleLanguage, SubtitleParsedItem, SubtitleTarget, SubtitleWriteView } from "@umt/contracts";
-import { FileUp, LoaderCircle, Undo2 } from "lucide-react";
+import type {
+  SubtitleLanguage,
+  SubtitleParsedItem,
+  SubtitleSearchResult,
+  SubtitleTarget,
+  SubtitleWriteView,
+} from "@umt/contracts";
+import { Download, ExternalLink, FileUp, LoaderCircle, Search, Undo2 } from "lucide-react";
 import { useId, useMemo, useRef, useState } from "react";
 import { toast } from "sonner";
 import {
   useApplySubtitles,
+  useFetchSubtitle,
   useParseSubtitles,
+  useSearchSubtitles,
   useSubtitleTargets,
   useSubtitleWrites,
   useUndoSubtitle,
@@ -20,7 +28,9 @@ import {
   DialogHeader,
   DialogTitle,
 } from "~/components/ui/dialog.tsx";
+import { Input } from "~/components/ui/input.tsx";
 import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from "~/components/ui/select.tsx";
+import { Tabs, TabsList, TabsTrigger } from "~/components/ui/tabs.tsx";
 import { relTime } from "~/components/workbench/release-fields.ts";
 import { errorMessage } from "~/lib/api.ts";
 import { cn } from "~/lib/utils.ts";
@@ -74,6 +84,7 @@ export function SubtitleDialog({
   const [rows, setRows] = useState<Row[]>([]);
   const [skipped, setSkipped] = useState<{ name: string; reason: string }[]>([]);
   const [dragging, setDragging] = useState(false);
+  const [tab, setTab] = useState<"upload" | "online">("upload");
   const list = targets.data?.targets ?? [];
   const busy = parse.isPending || apply.isPending;
 
@@ -112,6 +123,8 @@ export function SubtitleDialog({
           language: x.language,
           content: x.content,
           targetLocationId: x.target,
+          ...(x.source ? { source: x.source } : {}),
+          ...(x.sourceRef ? { sourceRef: x.sourceRef } : {}),
         })),
       });
       if (r.written.length) toast.success(`已写入 ${r.written.length} 个字幕`);
@@ -141,41 +154,58 @@ export function SubtitleDialog({
           <EmptyState title="媒体库里还没有这部作品的视频" description="整理入库后才能补字幕。" />
         ) : (
           <div className="min-w-0 space-y-4">
-            <label
-              htmlFor={inputId}
-              onDragOver={(e) => {
-                e.preventDefault();
-                setDragging(true);
-              }}
-              onDragLeave={() => setDragging(false)}
-              onDrop={(e) => {
-                e.preventDefault();
-                setDragging(false);
-                void load([...e.dataTransfer.files]);
-              }}
-              className={cn(
-                "flex cursor-pointer flex-col items-center gap-1 rounded-lg border border-dashed border-border px-4 py-6 text-center transition-colors hover:border-foreground/40",
-                dragging && "border-foreground/60 bg-accent",
-              )}
-            >
-              {parse.isPending ? <LoaderCircle className="size-5 animate-spin" /> : <FileUp className="size-5" />}
-              <span className="text-caption text-foreground">选择或拖入字幕文件</span>
-              <span className="text-meta text-muted-foreground">
-                支持 srt / ass / ssa / vtt / sub / sbv 和 zip，可一次选多个
-              </span>
-              <input
-                ref={input}
-                id={inputId}
-                type="file"
-                multiple
-                accept={ACCEPT}
-                className="sr-only"
-                onChange={(e) => {
-                  void load([...(e.target.files ?? [])]);
-                  if (input.current) input.current.value = "";
+            <Tabs value={tab} onValueChange={(v) => setTab(v as "upload" | "online")}>
+              <TabsList>
+                <TabsTrigger value="upload">上传</TabsTrigger>
+                <TabsTrigger value="online">在线搜索</TabsTrigger>
+              </TabsList>
+            </Tabs>
+            {tab === "upload" ? (
+              <label
+                htmlFor={inputId}
+                onDragOver={(e) => {
+                  e.preventDefault();
+                  setDragging(true);
+                }}
+                onDragLeave={() => setDragging(false)}
+                onDrop={(e) => {
+                  e.preventDefault();
+                  setDragging(false);
+                  void load([...e.dataTransfer.files]);
+                }}
+                className={cn(
+                  "flex cursor-pointer flex-col items-center gap-1 rounded-lg border border-dashed border-border px-4 py-6 text-center transition-colors hover:border-foreground/40",
+                  dragging && "border-foreground/60 bg-accent",
+                )}
+              >
+                {parse.isPending ? <LoaderCircle className="size-5 animate-spin" /> : <FileUp className="size-5" />}
+                <span className="text-caption text-foreground">选择或拖入字幕文件</span>
+                <span className="text-meta text-muted-foreground">
+                  支持 srt / ass / ssa / vtt / sub / sbv 和 zip，可一次选多个
+                </span>
+                <input
+                  ref={input}
+                  id={inputId}
+                  type="file"
+                  multiple
+                  accept={ACCEPT}
+                  className="sr-only"
+                  onChange={(e) => {
+                    void load([...(e.target.files ?? [])]);
+                    if (input.current) input.current.value = "";
+                  }}
+                />
+              </label>
+            ) : (
+              <OnlinePanel
+                entityId={entityId}
+                seasons={[...new Set(list.map((t) => t.season).filter((x): x is number => x !== null))]}
+                onFetched={(items, skip) => {
+                  setRows((prev) => [...prev, ...items.map((it) => ({ ...it, target: it.targetLocationId ?? SKIP }))]);
+                  setSkipped((prev) => [...prev, ...skip]);
                 }}
               />
-            </label>
+            )}
 
             {rows.length ? (
               <ul className="space-y-2">
@@ -318,3 +348,145 @@ function History({
     </section>
   );
 }
+
+/** 在线搜索:已启用的字幕站(SubHD、字幕库);下载的字幕进上面同一个预览 */
+function OnlinePanel({
+  entityId,
+  seasons,
+  onFetched,
+}: {
+  entityId: string;
+  seasons: number[];
+  onFetched: (items: SubtitleParsedItem[], skipped: { name: string; reason: string }[]) => void;
+}) {
+  const search = useSearchSubtitles();
+  const fetch = useFetchSubtitle();
+  const [keyword, setKeyword] = useState("");
+  const [season, setSeason] = useState<string>(seasons.length === 1 ? String(seasons[0]) : "all");
+  const [episode, setEpisode] = useState("");
+  const [fetching, setFetching] = useState<string | null>(null);
+  const data = search.data;
+
+  const run = () =>
+    search.mutate(
+      {
+        entityId,
+        ...(keyword.trim() ? { keyword: keyword.trim() } : {}),
+        ...(season !== "all" ? { season: Number(season) } : {}),
+        ...(episode.trim() && /^\d+$/.test(episode.trim()) ? { episode: Number(episode.trim()) } : {}),
+      },
+      { onError: (err) => toast.error(errorMessage(err, "搜索失败")) },
+    );
+  const take = async (r: SubtitleSearchResult) => {
+    setFetching(`${r.siteId}${r.ref}`);
+    try {
+      const p = await fetch.mutateAsync({ entityId, siteId: r.siteId, ref: r.ref, title: r.title });
+      if (!p.items.length) toast.error("这个字幕包里没有能用的字幕文件");
+      else toast.success(`已下载 ${p.items.length} 个字幕，检查下面的配对后写入`);
+      onFetched(p.items, p.skipped);
+    } catch (err) {
+      toast.error(errorMessage(err, "下载失败"));
+    } finally {
+      setFetching(null);
+    }
+  };
+
+  return (
+    <div className="min-w-0 space-y-3">
+      <form
+        className="flex flex-wrap items-center gap-2"
+        onSubmit={(e) => {
+          e.preventDefault();
+          run();
+        }}
+      >
+        <Input
+          value={keyword}
+          onChange={(e) => setKeyword(e.target.value)}
+          placeholder="默认按作品名搜，可另填关键词"
+          aria-label="关键词"
+          className="min-w-48 flex-1"
+        />
+        {seasons.length ? (
+          <>
+            <Select value={season} onValueChange={setSeason}>
+              <SelectTrigger aria-label="季" className="w-28">
+                <SelectValue />
+              </SelectTrigger>
+              <SelectContent position="popper">
+                <SelectItem value="all">全部季</SelectItem>
+                {seasons.map((n) => (
+                  <SelectItem key={n} value={String(n)}>
+                    第 {n} 季
+                  </SelectItem>
+                ))}
+              </SelectContent>
+            </Select>
+            <Input
+              value={episode}
+              onChange={(e) => setEpisode(e.target.value)}
+              placeholder="集（可空）"
+              aria-label="集"
+              inputMode="numeric"
+              className="w-24"
+            />
+          </>
+        ) : null}
+        <Button type="submit" disabled={search.isPending}>
+          {search.isPending ? <LoaderCircle className="animate-spin" /> : <Search />}
+          搜索
+        </Button>
+      </form>
+
+      {data?.noSites ? (
+        <p className="text-caption text-muted-foreground">
+          还没有启用的字幕站。到「站点 → 添加站点 → 字幕站」添加 SubHD 或字幕库。
+        </p>
+      ) : null}
+      {data?.errors.length ? (
+        <p className="text-meta text-warning">{data.errors.map((e) => `${e.siteName}：${e.reason}`).join("；")}</p>
+      ) : null}
+      {data && !data.noSites && !data.results.length ? (
+        <p className="text-caption text-muted-foreground">
+          没搜到（用了：{data.keywords.join("、") || "无"}）。可以换个关键词，或去掉季集限制再试。
+        </p>
+      ) : null}
+      {data?.results.length ? (
+        <ul className="max-h-80 space-y-1.5 overflow-y-auto">
+          {data.results.map((r) => {
+            const key = `${r.siteId}${r.ref}`;
+            return (
+              <li key={key} className="flex min-w-0 items-center gap-2 rounded-lg border border-border/70 px-3 py-2">
+                <div className="min-w-0 flex-1">
+                  <p className="truncate text-caption text-foreground" title={r.title}>
+                    {r.title}
+                  </p>
+                  <p className="truncate text-meta text-muted-foreground">
+                    {[
+                      r.siteName,
+                      r.languages.join(" / ") || null,
+                      r.format,
+                      r.downloads !== null ? `下载 ${r.downloads}` : null,
+                      r.publishedAt ? (r.publishedAt.includes("T") ? relTime(r.publishedAt) : r.publishedAt) : null,
+                    ]
+                      .filter(Boolean)
+                      .join(" · ")}
+                  </p>
+                </div>
+                <Button asChild variant="ghost" size="icon-sm" title="在字幕站打开">
+                  <a href={r.pageUrl} target="_blank" rel="noreferrer" aria-label="在字幕站打开">
+                    <ExternalLink />
+                  </a>
+                </Button>
+                <Button size="sm" variant="outline" disabled={fetching !== null} onClick={() => void take(r)}>
+                  {fetching === key ? <LoaderCircle className="animate-spin" /> : <Download />}
+                  下载
+                </Button>
+              </li>
+            );
+          })}
+        </ul>
+      ) : null}
+    </div>
+  );
+}

apps/web/app/routes/console/sites.tsx

@@ -21,11 +21,12 @@ const VIEW_KEY = "umt.sites.view";
 type View = "cards" | "list";
 type Filter = "all" | "enabled" | "attention";
 /** 分组顺序;公开 BT 站的表只列它用得到的列 */
-const GROUPS: { kind: SiteKind; title: string; variant: "pt" | "public" }[] = [
+const GROUPS: { kind: SiteKind; title: string; variant: "pt" | "public" | "subtitle" }[] = [
   { kind: "pt", title: "PT 站", variant: "pt" },
   { kind: "public_bt", title: "公开 BT 站", variant: "public" },
   { kind: "rss_feed", title: "RSS 源", variant: "public" },
   { kind: "torznab", title: "索引器", variant: "public" },
+  { kind: "subtitle", title: "字幕站", variant: "subtitle" },
 ];
 
 const FILTERS: { value: Filter; label: string }[] = [