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/browser/service.ts

@@ -217,6 +217,12 @@ export type ChallengeVision = (req: {
 
 const visions = new WeakMap<object, ChallengeVision>();
 
+/** 页面验证码接管结果:passed 带回站点 Cookie;needs_human 已开待处理事项;unavailable 没有接管条件(自动化关 / 没有浏览器能力) */
+export type PageCaptchaOutcome =
+  | { status: "passed"; by: "browser" | "vision"; cookies: { name: string; value: string }[] }
+  | { status: "needs_human"; reason: string; decisionId: string | null }
+  | { status: "unavailable"; reason: string };
+
 export function registerChallengeVision(ctx: object, fn: ChallengeVision): void {
   visions.set(ctx, fn);
 }
@@ -2012,7 +2018,7 @@ export class BrowserService {
    */
   private async cfSession(
     site: SiteRow,
-    accountId: string,
+    accountId: string | null,
     url: string,
   ): Promise<{ id: string } | { busy: BrowserSessionRow }> {
     const live = await this.liveSessionFor(site.id, accountId);
@@ -2194,6 +2200,111 @@ export class BrowserService {
     }
   }
 
+  /**
+   * 站点自己的页面验证码(字幕库的云锁)代码识别没过时,由智能体视觉接管(docs/plan/subtitles.md):
+   * 匿名会话打开 url → 已不是验证码页就直接算过 → 否则派视觉子流程(系统发起,读图、输入、提交,≤ 6 步,最多 1 次)
+   * → 按页面核实 marker 消失 → 读回站点 Cookie,交调用方并入它自己的 Cookie 罐后重试原请求。
+   * 没过:会话转等待,待处理事项写明去原站视图手动过一次;人过完交还后,下一次站点请求再走这里时就直接读回 Cookie。
+   */
+  async solvePageCaptcha(req: {
+    siteId: string;
+    url: string;
+    marker: string;
+    goal: string;
+  }): Promise<PageCaptchaOutcome> {
+    const site = await this.liveSite(req.siteId);
+    const vision = visions.get(this.ctx);
+    const auto = await this.ctx.settings.get("automation").catch(() => null);
+    if (!vision || !auto?.enabled || auto.readOnlyReconcile)
+      return { status: "unavailable", reason: "自动化未开启，智能体不接管验证码" };
+    const profileKey = profileKeyFor(site.id, null);
+    return this.cfLock.run(profileKey, async () => {
+      const got = await this.cfSession(site, null, req.url);
+      if ("busy" in got)
+        return {
+          status: "needs_human" as const,
+          reason: "已在等你在原站视图里过验证码",
+          decisionId: (obj(got.busy.recovery_state) as RecoveryState).waitingDecisionId ?? null,
+        };
+      const id = got.id;
+      let page = await this.readPage(id).catch(() => null);
+      if (!page) return { status: "unavailable" as const, reason: "浏览器服务不支持读取页面内容" };
+      let by: "browser" | "vision" = "browser";
+      if (page.text.includes(req.marker)) {
+        const r = await vision({ sessionId: id, goal: req.goal, textAbsent: req.marker, maxSteps: 6 }).catch((e) => ({
+          status: "failed",
+          reason: String((e as Error)?.message ?? e),
+          childRunId: null,
+        }));
+        this.ctx.logger.info(
+          { component: "browser", sessionId: id, siteId: site.id, status: r.status },
+          "页面验证码视觉接管结束",
+        );
+        page = await this.readPage(id).catch(() => null);
+        if (!page || page.text.includes(req.marker)) return this.captchaNeedsHuman(id, site, r.reason);
+        by = "vision";
+      }
+      const cookies = await this.siteCookies(profileKey, site.base_url);
+      return { status: "passed" as const, by, cookies };
+    });
+  }
+
+  /** profile 里作用于站点主机、未过期的 Cookie(只取名和值) */
+  private async siteCookies(profileKey: string, siteUrl: string): Promise<{ name: string; value: string }[]> {
+    const prow = await this.providerRow().catch(() => null);
+    const provider = prow ? this.providerFor(prow) : null;
+    const host = new URL(siteUrl).hostname;
+    const now = this.now().getTime();
+    return ((await provider?.getCookies?.(profileKey, [siteUrl]).catch(() => [])) ?? [])
+      .filter((c) => {
+        const d = c.domain.replace(/^\./, "");
+        return (
+          (host === d || host.endsWith(`.${d}`)) &&
+          !(c.expires !== undefined && c.expires > 0 && c.expires * 1000 <= now)
+        );
+      })
+      .map((c) => ({ name: c.name, value: c.value }));
+  }
+
+  /** 原站视图当前页的地址与站点 Cookie:字幕站从人工打开的字幕页直接下载(带上人在浏览器里过的验证) */
+  async currentPage(
+    id: string,
+    actor: Actor,
+  ): Promise<{ siteId: string; url: string; cookies: { name: string; value: string }[] }> {
+    const row = await this.sessionRow(id);
+    const page = await this.run(
+      id,
+      actor,
+      (p, ref) => (p.pageContent ? p.pageContent(ref, { currentVisible: true }) : Promise.resolve(null)),
+      {
+        write: false,
+      },
+    );
+    if (!page?.url) throw new BrowserServiceError("precondition", "无法读取当前页面，请重新连接浏览器");
+    const site = await this.liveSite(row.site_id);
+    return { siteId: site.id, url: page.url, cookies: await this.siteCookies(row.profile_key, site.base_url) };
+  }
+
+  private async captchaNeedsHuman(id: string, site: SiteRow, reason: string): Promise<PageCaptchaOutcome> {
+    const text = {
+      title: `${site.name}需要你在原站视图里过一次验证码`,
+      summary: `站点要求输入验证码，代码识别和智能体都没能通过（${reason.slice(0, 200) || "未说明原因"}）。在站点详情的原站视图接管会话，输完验证码后交还，再搜索一次即可。`,
+    };
+    let row = await this.sessionRow(id);
+    const rs = obj(row.recovery_state) as RecoveryState;
+    if (row.controller === "automation" || !rs.waitingDecisionId)
+      row = await this.enterWaiting(row, "challenge_unsolved", text);
+    else
+      await this.ctx.db
+        .updateTable("decision")
+        .set({ title: text.title, summary: text.summary })
+        .where("id", "=", rs.waitingDecisionId)
+        .execute();
+    const decisionId = (obj(row.recovery_state) as RecoveryState).waitingDecisionId ?? null;
+    this.ctx.logger.info({ component: "browser", sessionId: id, siteId: site.id }, "页面验证码转人工");
+    return { status: "needs_human", reason, decisionId };
+  }
+
   /** 人过完验证交还后:站点在等人过验证、会话是那一个 → 读回 cf_clearance / UA,通知站点请求继续 */
   private async afterHumanChallenge(id: string): Promise<void> {
     const row = await this.sessionRow(id);