Merge branch 'hh2'

# Conflicts:
#	apps/server/src/modules/browser/service.ts
#	apps/web/app/components/browser/desktop-view.tsx
#	apps/web/app/components/browser/site-origin-view.tsx
#	packages/contracts/src/browser.ts

apps/server/src/modules/browser/__tests__/browser-e2e.test.ts

@@ -96,7 +96,10 @@ describe.skipIf(!up)("e2e:登录失效 → 人工接管 → 交还 → Agent 重
     const fresh = await invoke("browser.snapshot", { siteId: h.siteId });
     expect(fresh).toMatchObject({ controller: "automation", loginRequired: false, waitingReason: null });
     expect(fresh.snapshot.text).toContain("种子列表");
-    expect((await h.req(`/browser/sessions/${sid}`)).body.data.needsSnapshot).toBe(false);
+    const afterAgent = (await h.req(`/browser/sessions/${sid}`)).body.data;
+    expect(afterAgent.needsSnapshot).toBe(false);
+    // 原站视图据此显示「Agent 正在操作」;人工动作不刷新它
+    expect(Date.now() - Date.parse(afterAgent.lastAutomationAt)).toBeLessThan(60_000);
     const dl = await invoke("browser.download", { sessionId: sid, target: { ref: "e11" }, timeoutMs: 2_000 });
     expect(dl).toMatchObject({
       sourceReleaseId: h.releaseId,

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

@@ -161,6 +161,8 @@ export interface RecoveryState {
   /** 打开 / 重连时浏览器服务的实例标识;变了 = 浏览器服务重启过 */
   instanceId?: string | null;
   reconnect?: BrowserReconnectState;
+  /** 最近一次自动化(Agent 能力)动作成功的时间;原站视图据此区分「Agent 正在操作」与「无人操作」 */
+  lastAutomationAt?: string;
 }
 
 /** 同一浏览器服务的实例标识多久查一次 */
@@ -674,6 +676,7 @@ export class BrowserService {
       credential: rs.credential ?? null,
       needsReconnect: rs.needsReconnect === true,
       reconnect: rs.reconnect ?? null,
+      lastAutomationAt: rs.lastAutomationAt ?? null,
     };
   }
 
@@ -1263,7 +1266,10 @@ export class BrowserService {
         throw fromBrowserError(e2);
       }
     }
-    if (opts.persist !== false) await this.persist(id, loaded.provider, opts.patch?.(out) ?? {});
+    const automation: RecoveryState = actor.kind === "automation" ? { lastAutomationAt: this.now().toISOString() } : {};
+    if (opts.persist !== false)
+      await this.persist(id, loaded.provider, { ...(opts.patch?.(out) ?? {}), ...automation });
+    else if (automation.lastAutomationAt) await this.patchRecovery(await this.sessionRow(id), automation);
     this.maybeSyncCookies(row);
     return out;
   }

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

@@ -5,7 +5,7 @@ import { errorBody } from "../../../http/errors.ts";
 import type { ModuleApp } from "../../types.ts";
 import { route } from "../http.ts";
 import type { SourcesRuntime } from "../runtime.ts";
-import { detailScreenshotUrl } from "../service/detail-page.ts";
+import { detailScreenshotRequest } from "../service/detail-page.ts";
 import { fetchPublicImage, imageResponse } from "../service/image-proxy.ts";
 import { listSiteReleases, releaseDetail } from "../service/workbench.ts";
 
@@ -53,9 +53,13 @@ export function workbenchRoutes(app: ModuleApp, rt: () => SourcesRuntime) {
       return c.json(errorBody("bad_request", "截图地址无效"), 400);
     }
     const r = rt();
-    const url = await detailScreenshotUrl(r, rid, index);
-    if (!url) return c.json(errorBody("not_found", "截图不存在或详情缓存已过期"), 404);
-    const got = await fetchPublicImage(r.overrides.fetch ?? fetch, url, { maxBytes: SHOT_MAX_BYTES, label: "截图" });
+    const req = await detailScreenshotRequest(r, rid, index);
+    if (!req) return c.json(errorBody("not_found", "截图不存在或详情缓存已过期"), 404);
+    const got = await fetchPublicImage(r.overrides.fetch ?? fetch, req.url, {
+      maxBytes: SHOT_MAX_BYTES,
+      label: "截图",
+      ...(req.referer ? { referer: req.referer } : {}),
+    });
     if (!got.ok) return c.json(errorBody("upstream_unavailable", got.message), 502);
     return imageResponse(got, 86_400);
   });

apps/server/src/modules/sources-identity/service/detail-page.ts

@@ -64,6 +64,38 @@ export async function detailScreenshotUrl(rt: SourcesRuntime, srId: string, inde
   return (await loadStored(rt, srId))?.screenshots[index] ?? null;
 }
 
+/** 根域(去 www.、取最后两级):pic.hhanclub.net 与 hhanclub.net 视为同站 */
+function rootHost(u: string): string | null {
+  try {
+    const host = new URL(u).hostname.toLowerCase().replace(/^www\./, "");
+    return /^[\d.]+$/.test(host) || host.includes(":") ? host : host.split(".").slice(-2).join(".");
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * 截图代理请求:原始地址 + 站点自家图床(同根域,如 pic.hhanclub.net / proxy-cover.hhanclub.net)要带的 Referer。
+ * 站点 CDN 常做防盗链,只认本站 Referer;外部图床不带 Referer。任何情况下都不带 Cookie。
+ */
+export async function detailScreenshotRequest(
+  rt: SourcesRuntime,
+  srId: string,
+  index: number,
+): Promise<{ url: string; referer?: string } | null> {
+  const url = await detailScreenshotUrl(rt, srId, index);
+  if (!url) return null;
+  const site = await rt.ctx.db
+    .selectFrom("source_release as r")
+    .innerJoin("site as s", "s.id", "r.site_id")
+    .select(["s.base_url"])
+    .where("r.id", "=", srId)
+    .executeTakeFirst();
+  const siteRoot = site ? rootHost(site.base_url) : null;
+  if (siteRoot && rootHost(url) === siteRoot) return { url, referer: `${new URL(site!.base_url).origin}/` };
+  return { url };
+}
+
 export async function loadDetailPage(rt: SourcesRuntime, srId: string): Promise<SiteReleasePage | null> {
   const p = await loadStored(rt, srId);
   if (!p) return null;

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

@@ -7,6 +7,7 @@ import {
   type FeedFetchState,
   type FeedItem,
   getSiteDefinition,
+  normalizeSiteFeedItem,
 } from "@umt/adapters";
 import type { FeedHealth } from "@umt/contracts";
 import { type FeedSourceRow, type SiteRow, sql } from "@umt/db";
@@ -157,6 +158,9 @@ export async function ingestFeed(
 
   const url = await secrets.resolve(feed.url_secret_ref);
   const result = await rt.rss.fetchFeed(url, stateFromRow(feed), { ignoreBackoff: !!opts.manual });
+  // 挂站点的 Feed:每次真正发出的 RSS 请求也记进站点请求计数(熔断 / 退避跳过的不算)
+  if (site && result.kind !== "skipped") rt.countSiteRequest(site.id);
+  if (site) await renormalizeFeedReleases(rt, site).catch(() => {});
   const now = rt.now();
   const auth = isAuthFailure(result);
   const health: FeedHealth = auth ? "failing" : healthOf(result);
@@ -175,7 +179,7 @@ export async function ingestFeed(
     if (result.kind === "ok") {
       const applied = applyFeedItems(window, result.feed.items);
       window = applied.cursor;
-      const counts = await writeItems(rt, trx, feed, site?.id ?? null, applied.newItems, applied.updatedItems);
+      const counts = await writeItems(rt, trx, feed, site ?? null, applied.newItems, applied.updatedItems);
       const unchanged = result.feed.items
         .filter((i) => !applied.newItems.includes(i) && !applied.updatedItems.includes(i))
         .map((i) => i.guid);
@@ -312,16 +316,36 @@ function observationRaw(item: FeedItem) {
   };
 }
 
+/** 规范化后的 RSS 条目字段(写进 source_release.raw.rss,v 为规范化版本) */
+export const RSS_NORMALIZE_VERSION = 1;
+
+function feedFields(siteKey: string, item: Pick<FeedItem, "title" | "size" | "categories">) {
+  const n = normalizeSiteFeedItem(siteKey, item.title);
+  // 分类:标题前缀 → RSS <category>(可能是名称)
+  const catName = n.category.name ?? item.categories[0] ?? null;
+  return {
+    title: n.title,
+    subtitle: n.subtitle,
+    categoryRaw: n.category.id ?? catName,
+    categoryName: catName,
+    mediaType: n.mediaType,
+    // enclosure length 是精确字节数,标题尾部的大小是四舍五入后的显示值
+    sizeBytes: item.size > 0 ? item.size : n.sizeBytes,
+    uploader: n.uploader,
+  };
+}
+
 async function writeItems(
   rt: SourcesRuntime,
   trx: Exec,
   feed: FeedSourceRow,
-  siteId: string | null,
+  site: Pick<SiteRow, "id" | "key"> | null,
   newItems: FeedItem[],
   updatedItems: FeedItem[],
 ): Promise<{ newReleases: number }> {
   let newReleases = 0;
   const updated = new Set(updatedItems);
+  const siteId = site?.id ?? null;
   for (const item of [...newItems, ...updatedItems]) {
     const raw = observationRaw(item);
     const obs = (
@@ -337,7 +361,7 @@ async function writeItems(
         returning id, source_release_id, (select content_digest from prev) as prev_digest`.execute(trx)
     ).rows[0]!;
 
-    if (!siteId) {
+    if (!site || !siteId) {
       // 无站点的通用 Feed:只留观察,无法按站点资源聚合
       await trx
         .updateTable("observation")
@@ -349,20 +373,36 @@ async function writeItems(
     const srid =
       extractSiteReleaseId(item) ?? `guid:${createHash("sha1").update(item.guid).digest("hex").slice(0, 20)}`;
     const contentChanged = updated.has(item) || (obs.prev_digest !== null && obs.prev_digest !== item.rawDigest);
-    const sizeBytes = item.size > 0 ? item.size : null;
+    const f = feedFields(site.key, item);
+    const rssRaw = {
+      link: item.link,
+      categories: item.categories,
+      rawTitle: item.title,
+      categoryName: f.categoryName,
+      mediaType: f.mediaType,
+      ...(f.uploader ? { uploader: f.uploader } : {}),
+      v: RSS_NORMALIZE_VERSION,
+    };
+    // 与列表快照按站点资源 ID 合并:快照见过(raw._site)的行,标题 / 副标题 / 分类 / 做种 / 促销 / H&R 以快照为准,
+    // RSS 只补空缺;快照没见过的行,内容变化时用 RSS 的新值
     const rel = (
       await sql<{ id: string; inserted: boolean; prev_title: string | null; resolution_locked: boolean }>`
         with prev as (select title from source_release where site_id = ${siteId} and site_release_id = ${srid})
-        insert into source_release (site_id, site_release_id, title, category_raw, size_bytes, source_published_at, raw)
-        values (${siteId}, ${srid}, ${item.title}, ${item.categories[0] ?? null}, ${sizeBytes}, ${item.publishedAt ?? null},
-                ${jsonb({ rss: { link: item.link, categories: item.categories } })}::jsonb)
+        insert into source_release (site_id, site_release_id, title, subtitle, category_raw, size_bytes, source_published_at, raw)
+        values (${siteId}, ${srid}, ${f.title}, ${f.subtitle}, ${f.categoryRaw}, ${f.sizeBytes}, ${item.publishedAt ?? null},
+                ${jsonb({ rss: rssRaw })}::jsonb)
         on conflict (site_id, site_release_id) do update set
           last_seen_at = now(),
-          title = case when ${contentChanged} then excluded.title else source_release.title end,
+          title = case when source_release.raw ? '_site' then source_release.title
+                       when ${contentChanged} then excluded.title else source_release.title end,
+          subtitle = case when source_release.raw ? '_site' then coalesce(source_release.subtitle, excluded.subtitle)
+                          when ${contentChanged} then coalesce(excluded.subtitle, source_release.subtitle)
+                          else coalesce(source_release.subtitle, excluded.subtitle) end,
           size_bytes = case when ${contentChanged} then coalesce(excluded.size_bytes, source_release.size_bytes)
                             else coalesce(source_release.size_bytes, excluded.size_bytes) end,
           source_published_at = coalesce(source_release.source_published_at, excluded.source_published_at),
-          category_raw = coalesce(source_release.category_raw, excluded.category_raw)
+          category_raw = coalesce(source_release.category_raw, excluded.category_raw),
+          raw = source_release.raw || excluded.raw
         returning id, (xmax = 0) as inserted, (select title from prev) as prev_title, resolution_locked`.execute(trx)
     ).rows[0]!;
     await trx
@@ -379,7 +419,7 @@ async function writeItems(
         { sourceReleaseId: rel.id, observationId: obs.id, siteId, isNew: true },
         { idempotencyKey: `release.observed:${rel.id}` },
       );
-    } else if (contentChanged && rel.prev_title !== item.title && !rel.resolution_locked) {
+    } else if (contentChanged && rel.prev_title !== f.title && !rel.resolution_locked) {
       // 标题修正:按新标题重新识别(证据版本变化)
       await rt.ctx.outbox.emit(
         trx,
@@ -391,3 +431,50 @@ async function writeItems(
   }
   return { newReleases };
 }
+
+/**
+ * 旧 RSS 行补规范化:规范化上线前写进来的站点 RSS 行(raw.rss 没有 v,标题还是 `[分类]…[大小]` 整串),
+ * 在该站 Feed 下次抓取时按原标题重算一次(每次最多 500 行,不请求站点)。快照见过的行不动。
+ * 标题变了且未锁定的,重新发 release.observed 让身份解析按干净标题再识别。
+ */
+export async function renormalizeFeedReleases(rt: SourcesRuntime, site: Pick<SiteRow, "id" | "key">): Promise<number> {
+  const rows = await rt.ctx.db
+    .selectFrom("source_release")
+    .select(["id", "title", "subtitle", "category_raw", "size_bytes", "raw", "resolution_locked"])
+    .where("site_id", "=", site.id)
+    .where(sql<boolean>`raw ? 'rss'`)
+    .where(sql<boolean>`not (raw ? '_site')`)
+    .where(sql<boolean>`coalesce((raw->'rss'->>'v')::int, 0) < ${RSS_NORMALIZE_VERSION}`)
+    .limit(500)
+    .execute();
+  if (!rows.length) return 0;
+  await rt.ctx.db.transaction().execute(async (trx) => {
+    for (const r of rows) {
+      const rss = ((r.raw ?? {}) as { rss?: Record<string, unknown> }).rss ?? {};
+      const rawTitle = typeof rss.rawTitle === "string" ? rss.rawTitle : r.title;
+      const categories = Array.isArray(rss.categories) ? (rss.categories as string[]) : [];
+      const f = feedFields(site.key, { title: rawTitle, size: Number(r.size_bytes ?? 0), categories });
+      const nextRss = {
+        ...rss,
+        rawTitle,
+        categoryName: f.categoryName,
+        mediaType: f.mediaType,
+        ...(f.uploader ? { uploader: f.uploader } : {}),
+        v: RSS_NORMALIZE_VERSION,
+      };
+      await sql`update source_release set title = ${f.title}, subtitle = coalesce(subtitle, ${f.subtitle}),
+          category_raw = coalesce(${f.categoryRaw}, category_raw), size_bytes = coalesce(size_bytes, ${f.sizeBytes}),
+          raw = raw || jsonb_build_object('rss', ${JSON.stringify(nextRss)}::jsonb)
+        where id = ${r.id}`.execute(trx);
+      if (f.title !== r.title && !r.resolution_locked) {
+        await rt.ctx.outbox.emit(
+          trx,
+          "release.observed",
+          { sourceReleaseId: r.id, observationId: null, siteId: site.id, isNew: false },
+          { idempotencyKey: `release.observed:${r.id}:rss-norm-${RSS_NORMALIZE_VERSION}` },
+        );
+      }
+    }
+  });
+  return rows.length;
+}