feat(subtitles): 字幕库验证码由智能体视觉接管;字幕站原站视图下载当前字幕

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

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

@@ -1,9 +1,16 @@
 import { Api, type SubtitleApplyRequest, type SubtitleFetchRequest, type SubtitleSource } from "@umt/contracts";
 import { ApiException } from "../../http/errors.ts";
 import { route } from "../agent/http.ts";
+import { human } from "../browser/routes.ts";
 import { sourcesIdentityModule } from "../sources-identity/index.ts";
 import type { ServerModule } from "../types.ts";
-import { downloadOnline, SubtitleSiteError, searchOnline, subtitleSearchQuery } from "./online/index.ts";
+import {
+  downloadFromSession,
+  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 */
@@ -32,10 +39,12 @@ export function createSubtitlesModule(): ServerModule {
       route(
         app,
         Api.subtitles.fetch,
-        ({ body }) =>
+        ({ body, c }) =>
           siteErr(async () => {
             const b = body as SubtitleFetchRequest;
-            const f = await downloadOnline(ctx, rt(), b.siteId, b.ref);
+            const f = b.sessionId
+              ? await downloadFromSession(ctx, rt(), b.sessionId, human(c))
+              : 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 {

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

@@ -6,6 +6,8 @@ import type { SiteRow } from "@umt/db";
 import { subtitleEpisode } from "@umt/domain";
 import { runSandbox } from "../../../sandbox/client.ts";
 import { b64encode } from "../../../sandbox/jobs.ts";
+import { browserServiceFor } from "../../browser/index.ts";
+import type { Actor } from "../../browser/service.ts";
 import type { SourcesRuntime } from "../../sources-identity/runtime.ts";
 import type { ServerContext } from "../../types.ts";
 import type { PageEntry, PageKind, PageResult } from "./parse.ts";
@@ -22,6 +24,7 @@ export class SubtitleSiteError extends Error {
 const jars = new Map<string, Map<string, string>>();
 
 interface Env {
+  ctx: ServerContext;
   rt: SourcesRuntime;
   site: SiteRow;
 }
@@ -206,22 +209,53 @@ async function subhdDownload(env: Env, ref: string): Promise<{ name: string; byt
 
 // ── 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) */
+/** 云锁验证码页上的字样(浏览器里 innerText 能读到) */
+const ZIMUKU_CAPTCHA_MARKER = "网站访问认证页面";
+
+/**
+ * 打开页面;遇到云锁验证码页:先在沙箱里识别、提交验证后重开(最多 3 次);
+ * 还没过 → 智能体视觉在浏览器里接管(读图输码),通过后把浏览器里的站点 Cookie 并进 Cookie 罐再开一次。
+ */
 async function zimukuPage(env: Env, url: string, referer?: string): Promise<string> {
-  for (let i = 0; i < 3; i++) {
+  /** 开一次页面:正常页返回 HTML,验证码页返回 BMP */
+  const open = async (): Promise<{ html: string } | { bmp: string }> => {
     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 ${res.status}`);
-      return html;
-    }
-    const { code } = await runSandbox("captcha.solve", { kind: "zimuku", bmpBase64: cap.bmpBase64 });
+    const bmp = (await page("zimuku.captcha", html)).bmpBase64;
+    if (bmp) return { bmp };
+    if (!res.ok) throw new SubtitleSiteError(`${env.site.name} 返回 HTTP ${res.status}`);
+    return { html };
+  };
+  for (let i = 0; i < 3; i++) {
+    const o = await open();
+    if ("html" in o) return o.html;
+    const { code } = await runSandbox("captcha.solve", { kind: "zimuku", bmpBase64: o.bmp });
     if (!code) break;
     const hex = [...code].map((c) => c.charCodeAt(0).toString(16)).join("");
     await (await request(env, `${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");
+  const browser = browserServiceFor(env.ctx);
+  const r = browser
+    ? await browser.solvePageCaptcha({
+        siteId: env.site.id,
+        url,
+        marker: ZIMUKU_CAPTCHA_MARKER,
+        goal: "这是字幕库的网站防火墙验证页：截图看清验证码图片里的字符，在下面的输入框里输入，再点「点击继续访问网站」，等页面跳转。只做这些；看不清就刷新一次图片再试，仍不行就停下说明需要人工。",
+      })
+    : { status: "unavailable" as const, reason: "浏览器模块未启用" };
+  if (r.status === "passed") {
+    const jar = jars.get(env.site.id) ?? new Map<string, string>();
+    jars.set(env.site.id, jar);
+    for (const c of r.cookies) jar.set(c.name, c.value);
+    const o = await open();
+    if ("html" in o) return o.html;
+    throw new SubtitleSiteError("字幕库的验证码在浏览器里过了，但验证结果带不回站点请求：稍后再试");
+  }
+  if (r.status === "needs_human")
+    throw new SubtitleSiteError(
+      "字幕库的验证码代码识别和智能体都没过：已开待处理事项，去站点的原站视图输一次验证码后再搜",
+    );
+  throw new SubtitleSiteError(`字幕库的验证码没能自动通过（${r.reason}）：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[]> {
@@ -326,7 +360,7 @@ export async function searchOnline(
   const errors: SubtitleSearchResponse["errors"] = [];
   await Promise.all(
     sites.map(async (site) => {
-      const env = { rt, site };
+      const env = { ctx, rt, site };
       try {
         const entries = await PROVIDERS[site.key]!.search(env, q);
         const seen = new Set<string>();
@@ -350,10 +384,40 @@ export async function downloadOnline(
   rt: SourcesRuntime,
   siteId: string,
   ref: string,
+  cookies: { name: string; value: 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 env = { ctx, rt, site };
+  if (cookies.length) {
+    const jar = jars.get(site.id) ?? new Map<string, string>();
+    jars.set(site.id, jar);
+    for (const c of cookies) jar.set(c.name, c.value);
+  }
   const f = await PROVIDERS[site.key]!.download(env, ref);
   return { name: f.name, base64: b64encode(f.bytes), siteKey: site.key, pageUrl: abs(env, ref) };
 }
+
+/** 字幕详情页地址 → 条目 ref(SubHD /a/<sid>、字幕库 /detail/<id>.html);不是详情页为 null */
+const DETAIL_PAGE: Record<string, RegExp> = {
+  subhd: /^\/a\/[A-Za-z0-9_-]+$/,
+  zimuku: /^\/detail\/\d+\.html$/,
+};
+
+/** 原站视图里人工打开的字幕页:取当前页地址与浏览器里的 Cookie,按同一流程下载 */
+export async function downloadFromSession(
+  ctx: ServerContext,
+  rt: SourcesRuntime,
+  sessionId: string,
+  actor: Actor,
+): Promise<{ name: string; base64: string; siteKey: string; pageUrl: string }> {
+  const browser = browserServiceFor(ctx);
+  if (!browser) throw new SubtitleSiteError("浏览器模块未启用");
+  const p = await browser.currentPage(sessionId, actor);
+  const site = (await subtitleSites(ctx)).find((s) => s.id === p.siteId);
+  if (!site || !PROVIDERS[site.key]) throw new SubtitleSiteError("这个会话不是已启用的字幕站");
+  const url = new URL(p.url);
+  if (url.hostname !== new URL(site.base_url).hostname || !DETAIL_PAGE[site.key]?.test(url.pathname))
+    throw new SubtitleSiteError("先在原站视图里打开一条字幕的详情页，再点下载");
+  return downloadOnline(ctx, rt, site.id, url.pathname, p.cookies);
+}