feat(scrape): 任意目录就地刮削,按路径识别作品,每次刮削一条记录(SCR-2)

- POST /files/scrape 改为对目录 / 视频文件刮削:已登记用作品,否则按路径识别(本地 + TMDB),
  认不准返回候选,选定后再提交;未登记直接用 TMDB 字段,addToLibrary 顺手建作品
- 作品详情重新刮削、入库后自动刮削改为对作品目录刮削,同一实现
- 迁移 0058_scrape_record 替代 media_scrape;GET /files/scrape-records
- 与 MP ScrapingChain 的文件布局差分(7 个场景,3 处登记偏离)

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

apps/server/src/modules/acquisition-organize/__tests__/scrape.test.ts

@@ -3,11 +3,11 @@ import { join, relative } from "node:path";
 import { createTempTree, type TempTree } from "@umt/testkit";
 import { afterAll, beforeAll, describe, expect, it } from "vitest";
 import { executeRecord, planImport } from "../services/organize.ts";
-import { runScrape, SCRAPE_TASK } from "../services/scrape.ts";
+import { runScrapeRecord, SCRAPE_TASK, scrapeEntity } from "../services/scrape.ts";
 import { createHarness, type Harness, pgAvailable, seedBase, seedEntity } from "./harness.ts";
 
-// SCR-1 刮削:入库成功后自动排 scrapeMedia → 写 NFO 与图片(临时库根);只补缺失 / 全部覆盖;图片失败不阻塞 NFO;
-// 只登记库根不写;作品详情「重新刮削」接口与刮削状态。
+// 刮削(SCR-1 / SCR-2):入库成功后自动对作品目录排 scrapeMedia → 写 NFO 与图片(临时库根);只补缺失 / 全部覆盖;
+// 图片失败不阻塞 NFO;只登记库根不写;作品详情「重新刮削」接口与刮削状态(最近一条刮削记录)。
 
 const hasPg = await pgAvailable();
 const SIZE = 200_000;
@@ -75,13 +75,26 @@ describe.skipIf(!hasPg)("刮削:NFO 与图片写进媒体库目录", () => {
     return plan.recordId!;
   }
 
-  /** 按引擎记录的入队项跑刮削(假引擎不执行) */
+  /** 按引擎记录的入队项跑刮削(假引擎不执行):该作品最近一条刮削记录 */
   async function runQueued(entity: string) {
-    const q = h.engine.enqueued.filter(
-      (x) => x.task === SCRAPE_TASK && (x.input as { mediaEntityId: string }).mediaEntityId === entity,
-    );
-    expect(q.length).toBeGreaterThan(0);
-    return runScrape(h.d, q[q.length - 1]!.input as Parameters<typeof runScrape>[1]);
+    const recs = await h.db
+      .selectFrom("scrape_record")
+      .select(["id"])
+      .where("media_entity_id", "=", entity)
+      .where("status", "=", "queued")
+      .orderBy("requested_at", "desc")
+      .execute();
+    expect(recs.length).toBeGreaterThan(0);
+    const q = h.engine.enqueued.filter((x) => x.task === SCRAPE_TASK);
+    expect(q.some((x) => (x.input as { recordId: string }).recordId === recs[0]!.id)).toBe(true);
+    return runScrapeRecord(h.d, recs[0]!.id);
+  }
+
+  /** 作品详情「重新刮削」的服务端做法:对作品目录排刮削并执行 */
+  async function rescrape(entity: string, mode?: "missing" | "overwrite") {
+    const q = await scrapeEntity(h.d, entity, { trigger: "manual", mode });
+    expect(q.length).toBe(1);
+    return runScrapeRecord(h.d, q[0]!.recordId);
   }
 
   let series = "";
@@ -165,8 +178,18 @@ describe.skipIf(!hasPg)("刮削:NFO 与图片写进媒体库目录", () => {
     const rec = await organize(series, "Severance.S01E01.1080p.WEB-DL.H264-GRP.mkv");
     await organize(series, "Severance.S01E02.1080p.WEB-DL.H264-GRP.mkv");
     const q = h.engine.enqueued.filter((x) => x.task === SCRAPE_TASK);
-    expect(q[0]!.key).toBe(`${SCRAPE_TASK}:record:${rec}`);
-    expect(q[0]!.input).toMatchObject({ mediaEntityId: series, trigger: "auto", mode: "missing", recordId: rec });
+    expect(q[0]!.key).toBe(`${SCRAPE_TASK}:record:${rec}:0`);
+    const first = await h.db
+      .selectFrom("scrape_record")
+      .selectAll()
+      .where("id", "=", (q[0]!.input as { recordId: string }).recordId)
+      .executeTakeFirstOrThrow();
+    expect(first).toMatchObject({
+      media_entity_id: series,
+      trigger: "auto",
+      mode: "missing",
+      organization_record_id: rec,
+    });
 
     const r = await runQueued(series);
     expect(r.status).toBe("partial"); // 背景图下载失败
@@ -213,23 +236,23 @@ describe.skipIf(!hasPg)("刮削:NFO 与图片写进媒体库目录", () => {
     });
     const audit = await h.db
       .selectFrom("audit_event")
-      .select(["action", "target_id"])
+      .select(["action", "target_id", "target_type"])
       .where("action", "=", "organize.scrape")
       .execute();
-    expect(audit.some((a) => a.target_id === series)).toBe(true);
+    expect(audit.some((a) => a.target_type === "scrape_record" && a.target_id === st.body.data.scrape.id)).toBe(true);
   }, 60_000);
 
   it("只补缺失:已有文件(包括别人写的)不动;全部覆盖:改写", async () => {
     const files = await tree(lib.root);
     const show = join(lib.root, files.find((p) => p.endsWith("tvshow.nfo"))!);
     await writeFile(show, "<tvshow>别的工具写的</tvshow>");
-    const again = await runScrape(h.d, { mediaEntityId: series, trigger: "manual", mode: "missing" });
+    const again = await rescrape(series, "missing");
     expect(await readFile(show, "utf8")).toBe("<tvshow>别的工具写的</tvshow>");
     expect(again.files.find((x) => x.role === "tvshow_nfo")).toMatchObject({
       action: "skipped",
       reason: "scrape_exists",
     });
-    const over = await runScrape(h.d, { mediaEntityId: series, trigger: "manual", mode: "overwrite" });
+    const over = await rescrape(series, "overwrite");
     expect(over.written).toBeGreaterThan(0);
     expect(await readFile(show, "utf8")).toContain("<title>人生切割术</title>");
   }, 60_000);
@@ -276,24 +299,23 @@ describe.skipIf(!hasPg)("刮削:NFO 与图片写进媒体库目录", () => {
     expect(nfo).toContain("<runtime>169</runtime>");
   }, 60_000);
 
-  it("写入失败记原因;只登记库根不写文件;没有入库文件的作品 409", async () => {
+  it("写入失败记原因;只登记库根不写文件;没有入库文件的作品 412", async () => {
     const files = await tree(lib.root);
     const video = files.find((p) => p.includes("星际穿越") && p.endsWith(".mkv"))!;
     const dirAbs = join(lib.root, video.slice(0, video.lastIndexOf("/")));
     if (process.getuid?.() !== 0) {
       await chmod(dirAbs, 0o555);
-      const r = await runScrape(h.d, { mediaEntityId: movie, trigger: "manual", mode: "overwrite" });
+      const r = await rescrape(movie, "overwrite");
       await chmod(dirAbs, 0o755);
       expect(r.files.some((x) => x.reason === "scrape_write_failed")).toBe(true);
     }
     await h.db.updateTable("library_root").set({ management_mode: "register_only" }).execute();
-    const ro = await runScrape(h.d, { mediaEntityId: movie, trigger: "manual", mode: "overwrite" });
+    const ro = await rescrape(movie, "overwrite");
     expect(ro).toMatchObject({ status: "failed", reason: "scrape_root_read_only", written: 0 });
     await h.db.updateTable("library_root").set({ management_mode: "managed" }).execute();
 
     const empty = await seedEntity(h.db, { kind: "movie", title: "没入库", year: 2020 });
     expect((await h.api("POST", `/library/${empty}/scrape`, {})).status).toBe(412);
-    const none = await runScrape(h.d, { mediaEntityId: empty, trigger: "manual" });
-    expect(none).toMatchObject({ status: "failed", reason: "scrape_no_metadata" });
+    expect(await scrapeEntity(h.d, empty, { trigger: "manual" })).toEqual([]);
   }, 60_000);
 });

apps/server/src/modules/acquisition-organize/handlers.ts

@@ -5,7 +5,7 @@ import { type AoDeps, automationBlock, ensureTaskRow, markTask } from "./service
 import { cleanupDownloadCache, submitDownload } from "./services/downloads.ts";
 import { executeRecord, planImport } from "./services/organize.ts";
 import { reconcileLibrary } from "./services/reconcile.ts";
-import { runScrape, SCRAPE_TASK } from "./services/scrape.ts";
+import { runScrapeRecord, SCRAPE_TASK } from "./services/scrape.ts";
 import { errMessage } from "./util.ts";
 
 // 持久任务处理器。输入只放业务 id;业务状态写业务表;副作用经 withAttemptToken(在 services 里)。
@@ -188,22 +188,16 @@ export function registerAoTasks(registry: TaskRegistry, d: AoDeps) {
         );
       },
     }),
-    // SCR-1 刮削:写 NFO 与图片到媒体库目录(文件道);结果写 media_scrape,不重试(失败的文件在作品详情里重新刮削)
+    // 刮削(SCR-1 / SCR-2):按刮削记录在目录里就地写 NFO 与图片(文件道);结果写回 scrape_record,不重试(失败的重新刮削)
     defineTask({
       name: SCRAPE_TASK,
-      version: 1,
+      version: 2,
       lane: "file",
-      inputSchema: z.object({
-        mediaEntityId: id,
-        trigger: z.enum(["auto", "manual"]),
-        mode: z.enum(["missing", "overwrite"]).optional(),
-        recordId: id.optional(),
-        actor: z.object({ type: z.string(), id: z.string().nullable() }).optional(),
-      }),
+      inputSchema: z.object({ recordId: id }),
       retryPolicy: { maximumAttempts: 1 },
       handler: async (t) =>
         t.step.run("scrape", async () => {
-          const r = await runScrape(d, { ...t.input, runId: t.runId }, t.abortSignal);
+          const r = await runScrapeRecord(d, t.input.recordId, t.abortSignal);
           return { status: r.status, written: r.written, skipped: r.skipped, failed: r.failed, reason: r.reason };
         }),
     }),

apps/server/src/modules/acquisition-organize/ports.ts

@@ -6,6 +6,8 @@ import {
   type DownloaderClient,
   diskUsage,
   type EmbyClient,
+  fetchRichDetail,
+  fetchRichSeason,
   ffprobe,
   getSiteAdapter,
   getSiteDefinition,
@@ -13,6 +15,8 @@ import {
   qbAuthConfig,
   type SecretResolver,
   SiteError,
+  type TmdbRichDetail,
+  type TmdbRichSeasonDetail,
   type TorrentSource,
   type VerifyMethod,
 } from "@umt/adapters";
@@ -88,6 +92,12 @@ export interface AoPorts {
    * 失败抛错(由刮削记为该图片失败,不影响 NFO)。
    */
   scrapeImage(tmdbPath: string): Promise<{ content: Uint8Array; mime: string }>;
+  /**
+   * SCR-2 刮削未登记的目录:直接取 TMDB 富详情 / 季详情(按设置的元数据语言链,经 TMDB 缓存)。
+   * TMDB 没配置返回 null;条目不存在或上游失败抛错。
+   */
+  scrapeTmdbDetail(type: "movie" | "tv", id: string): Promise<TmdbRichDetail | null>;
+  scrapeTmdbSeason(tvId: string, season: number): Promise<TmdbRichSeasonDetail | null>;
   options: AoOptions;
 }
 
@@ -197,6 +207,16 @@ export function createPorts(
       });
       return images.bytes("original", tmdbPath);
     },
+    async scrapeTmdbDetail(type, id) {
+      const client = await rt().tmdb();
+      if (!client) return null;
+      return fetchRichDetail(client, type, id, await metadataLanguages(ctx), await metadataTextOptions(ctx));
+    },
+    async scrapeTmdbSeason(tvId, season) {
+      const client = await rt().tmdb();
+      if (!client) return null;
+      return fetchRichSeason(client, tvId, season, await metadataLanguages(ctx), await metadataTextOptions(ctx));
+    },
     diskUsage: (p) => diskUsage(p),
     probe: (p) => ffprobe(p),
     async mediaServer(inst) {

apps/server/src/modules/acquisition-organize/routes/library.ts

@@ -27,7 +27,7 @@ import { sql } from "@umt/db";
 import { ApiException } from "../../../http/errors.ts";
 import type { ModuleApp } from "../../types.ts";
 import { type AoDeps, CANDIDATE_ARTIFACT_TITLE } from "../services/common.ts";
-import { enqueueScrape, loadScrapeState } from "../services/scrape.ts";
+import { latestEntityScrape, scrapeEntity } from "../services/scrape.ts";
 import { notFound, toApi, toApiList } from "../util.ts";
 import { mount, paginate } from "./helpers.ts";
 import { registerLibraryVersionRoutes } from "./library-versions.ts";
@@ -407,21 +407,21 @@ export function registerLibraryRoutes(app: ModuleApp, d: AoDeps) {
           createdAt: new Date(r.created_at).toISOString(),
           finishedAt: r.finished_at ? new Date(r.finished_at).toISOString() : null,
         })),
-        scrape: await loadScrapeState(d, id),
+        scrape: await latestEntityScrape(d, id),
       };
       return detail;
     },
     { tag: TAG },
   );
 
-  // SCR-1 刮削:作品详情「重新刮削」;与入库后自动刮削、文件管理「刮削」同一任务 scrapeMedia
+  // 作品详情「重新刮削」「覆盖重写」:对作品的目录刮削(SCR-2:与文件管理「刮削」、入库后自动刮削同一实现)
   mount(
     app,
     {
       method: "post",
       path: "/library/{id}/scrape",
       scope: "organize.execute",
-      summary: "刮削：把作品的 NFO 与图片写进媒体库目录（后台任务 scrapeMedia；覆盖方式缺省按设置）",
+      summary: "刮削：对作品目录就地写 NFO 与图片（后台任务 scrapeMedia；覆盖方式缺省按设置）",
       params: IdParam,
       body: LibraryScrapeRequest,
       response: LibraryScrapeResult,
@@ -429,22 +429,20 @@ export function registerLibraryRoutes(app: ModuleApp, d: AoDeps) {
     async (c, { params, body }) => {
       const e = await db().selectFrom("media_entity").select(["id"]).where("id", "=", params.id).executeTakeFirst();
       if (!e) notFound("作品");
-      const has = await db()
-        .selectFrom("media_asset")
-        .select(["id"])
-        .where("media_entity_id", "=", params.id)
-        .where("status", "=", "active")
-        .executeTakeFirst();
-      if (!has) throw new ApiException("precondition_failed", "作品没有已入库文件。先整理入库，再刮削");
       const actor = c.get("actor");
       const req = (body ?? {}) as LibraryScrapeRequest;
-      const r = await enqueueScrape(d, {
-        mediaEntityId: params.id,
+      const recs = await scrapeEntity(d, params.id, {
         trigger: "manual",
         mode: req.mode,
         actor: { type: actor?.type === "api_key" ? "api_key" : "user", id: actor?.id ?? null },
       });
-      return { mediaEntityId: params.id, ...r };
+      if (!recs.length) throw new ApiException("precondition_failed", "作品没有已入库文件。到文件管理里对目录刮削");
+      return {
+        mediaEntityId: params.id,
+        taskId: recs[0]?.taskId ?? null,
+        deduplicated: recs.every((r) => r.deduplicated),
+        recordIds: recs.map((r) => r.recordId),
+      };
     },
     { tag: TAG, status: 202 },
   );

apps/server/src/modules/acquisition-organize/services/scrape-input.ts

@@ -1,4 +1,6 @@
+import type { TmdbRichDetail, TmdbRichSeasonDetail } from "@umt/adapters";
 import type { NfoEpisode, NfoIds, NfoPerson, NfoSeason, NfoWork } from "@umt/domain";
+import { richEntityFields } from "../../sources-identity/service/metadata-rich.ts";
 
 // SCR-1:本地作品 / 季 / 集的最终字段(media_entity.metadata、content_unit.metadata,字段级 {value, …})→ NFO 输入。
 // 纯函数,不查库;差分测试(apps/server/test/scrape/nfo-mp-diff.test.ts)直接调用。
@@ -134,3 +136,58 @@ export function imagePathsOf(entityMeta: unknown) {
 }
 export const unitPosterPath = (m: unknown) => str(val(m, "posterPath"));
 export const unitStillPath = (m: unknown) => str(val(m, "stillPath"));
+
+// ── SCR-2:未登记目录直接用 TMDB 字段(与刷新元数据同一套映射,形状同 media_entity.metadata / content_unit.metadata) ──
+
+/** TMDB 富详情 → 作品(NFO 输入);字段映射同刷新元数据(richEntityFields) */
+export function tmdbEntity(r: TmdbRichDetail): { entity: EntityForNfo; ids: IdentityMap } {
+  return {
+    entity: {
+      kind: r.entityType === "movie" ? "movie" : "series",
+      title: r.title,
+      original_title: r.originalTitle,
+      year: r.year,
+      metadata: richEntityFields(r),
+    },
+    ids: {
+      tmdb: r.externalId,
+      ...(r.externalIds.imdb ? { imdb: r.externalIds.imdb } : {}),
+      ...(r.externalIds.tvdb ? { tvdb: r.externalIds.tvdb } : {}),
+    },
+  };
+}
+
+const fv = (value: unknown) => ({ value });
+
+/** TMDB 季详情 → 季与各集(NFO 输入) */
+export function tmdbSeasonUnits(sd: TmdbRichSeasonDetail): { season: UnitForNfo; episodes: Map<number, UnitForNfo> } {
+  const season: UnitForNfo = {
+    season_number: sd.seasonNumber,
+    episode_number: null,
+    title: sd.name,
+    air_date: sd.airDate,
+    metadata: {
+      name: fv(sd.name),
+      overview: fv(sd.overview),
+      airDate: fv(sd.airDate),
+      posterPath: fv(sd.posterPath),
+    },
+  };
+  const episodes = new Map<number, UnitForNfo>();
+  for (const e of sd.episodes)
+    episodes.set(e.episodeNumber, {
+      season_number: e.seasonNumber,
+      episode_number: e.episodeNumber,
+      title: e.name,
+      air_date: e.airDate,
+      metadata: {
+        name: fv(e.name),
+        overview: fv(e.overview),
+        airDate: fv(e.airDate?.slice(0, 10) ?? null),
+        runtime: fv(e.runtime),
+        stillPath: fv(e.stillPath),
+        voteAverage: fv(e.voteAverage),
+      },
+    });
+  return { season, episodes };
+}

apps/server/src/modules/file-manager/identify.ts

@@ -11,7 +11,14 @@ import {
   type OrganizationStatus,
 } from "@umt/contracts";
 import { sql } from "@umt/db";
-import { type IdentityCandidate, type ParsedRelease, parseReleasePath, queryTitles, rankCandidates } from "@umt/domain";
+import {
+  type IdentityCandidate,
+  type ParsedRelease,
+  parseReleasePath,
+  queryTitles,
+  rankCandidates,
+  resolveIdentity,
+} from "@umt/domain";
 import { localIdentityCandidate } from "@umt/mapping";
 import type { Principal } from "../../http/auth.ts";
 import { EXTERNAL_INVOKE_TIMEOUT_MS, sharedInvokeLimiter } from "../agent/invoke-limits.ts";
@@ -177,13 +184,24 @@ export async function runIdentifyRemote(
   return payload;
 }
 
-async function identifyCore(
+/** 候选键:本地作品 id 或 `provider:externalId`(与 FileIdentifyCandidate 对应) */
+export const candidateKey = (c: { mediaEntityId: string | null; provider: string; externalId: string | null }) =>
+  c.mediaEntityId ?? `${c.provider}:${c.externalId}`;
+
+export async function identifyCore(
   d: FmDeps,
   body: FilesIdentifyRequest,
   opts: CoreOpts,
-): Promise<{ result: FilesIdentifyResult; remotePending: number; pendingPaths: Set<string> }> {
+): Promise<{
+  result: FilesIdentifyResult;
+  remotePending: number;
+  pendingPaths: Set<string>;
+  /** SCR-2:每个视频按 resolveIdentity 能否自动确认(root\0path → 候选键;不能为 null) */
+  autoPicks: Map<string, string | null>;
+}> {
   const db = d.ctx.db;
   const pendingPaths = new Set<string>();
+  const autoPicks = new Map<string, string | null>();
   const { videos, truncated } = await collectVideos(d, body);
   // G2:最小体积(请求/向导 > settings.organize.minSizeBytes,缺省 50 MiB;设置读不到时不过滤)
   const orgSettings = (await d.ctx.settings.get("organize").catch(() => null)) as { minSizeBytes?: unknown } | null;
@@ -411,6 +429,14 @@ async function identifyCore(
         ),
       };
     });
+    const res = resolveIdentity(parsed, all);
+    const pick = res.status === "auto" && res.best ? res.best.candidate : null;
+    autoPicks.set(
+      `${v.root}\u0000${v.rel}`,
+      pick && (!kindHint || (kindHint === "movie") === (pick.entityType === "movie"))
+        ? (pick.localEntityId ?? `${pick.provider}:${pick.externalId}`)
+        : null,
+    );
     const best = candidates[0];
     let contentUnitId: string | null = null;
     let contentUnit: string | null = null;
@@ -454,7 +480,7 @@ async function identifyCore(
   }
   if (videos.length === 0 && body.items.length > 0 && !truncated)
     throw new FilesError("bad_request", "所选项中没有视频文件");
-  return { result: { items, truncated, excludedBySize }, remotePending: remoteQueries, pendingPaths };
+  return { result: { items, truncated, excludedBySize }, remotePending: remoteQueries, pendingPaths, autoPicks };
 }
 
 /** TMDB 图片 CDN 缩略图(w154);已是完整 URL 的原样返回 */

apps/server/src/modules/file-manager/probe.ts

@@ -8,21 +8,17 @@ import type {
   FilesProbeResult,
   FilesScrapeBatchRequest,
   FilesScrapeBatchResult,
-  FilesScrapeRequest,
-  FilesScrapeResult,
 } from "@umt/contracts";
 import { sql } from "@umt/db";
 import type { Principal } from "../../http/auth.ts";
-import { enqueueScrape } from "../acquisition-organize/services/scrape.ts";
 import { FilesError, type FmDeps, jsonb, resolveIn } from "./common.ts";
 import { audit } from "./ledger.ts";
 import { kindOfName, mediaSummary } from "./list.ts";
 import { entitiesUnder } from "./records.ts";
 
 // probe:ffprobe 摘要;已登记资产写回 media_asset.media_info.probe,未登记只缓存 24h(DATA_DIR/cache/probe)。
-// scrape:只对已登记作品,等价 POST /media-entities/{id}/refresh(入队 refreshMetadata,字段锁优先、不动路径);
-// writeNfo / images(SCR-1)另排 scrapeMedia,把 NFO 与图片写进媒体库目录(与作品详情「重新刮削」同一任务)。
-// scrape-batch write=true:文件管理「刮削」,按作品去重各排一次 scrapeMedia(不刷新元数据)。
+// scrape-batch:文件管理「刷新元数据」,按作品去重各排一次 refreshMetadata(字段锁优先、不动路径、不写文件)。
+// 刮削(写 NFO 与图片)见 scrape.ts。
 
 const CACHE_TTL_MS = 24 * 3600_000;
 
@@ -104,62 +100,8 @@ export async function probeFiles(d: FmDeps, body: FilesProbeRequest): Promise<Fi
   return { items };
 }
 
-export async function scrape(
-  d: FmDeps,
-  principal: Principal | undefined,
-  body: FilesScrapeRequest,
-): Promise<FilesScrapeResult> {
-  const db = d.ctx.db;
-  const e = await db
-    .selectFrom("media_entity")
-    .select(["id", "status", "field_locks"])
-    .where("id", "=", body.mediaEntityId)
-    .executeTakeFirst();
-  if (!e) throw new FilesError("not_found", "作品不存在");
-  const registered = await db
-    .selectFrom("media_asset")
-    .select(["id"])
-    .where("media_entity_id", "=", e.id)
-    .where("status", "=", "active")
-    .executeTakeFirst();
-  if (!registered) throw new FilesError("bad_request", "作品没有已入库文件。先识别并整理入库，再刮削");
-  const { runId } = await d.ctx.engine.enqueue(
-    "refreshMetadata",
-    { mediaEntityId: e.id },
-    { idempotencyKey: `refreshMetadata:files:${body.idempotencyKey}`, traceId: randomUUID() },
-  );
-  const overrides = await db
-    .selectFrom("field_override")
-    .select(["field"])
-    .where("media_entity_id", "=", e.id)
-    .where("locked", "=", true)
-    .execute();
-  const lockedFields = [...new Set([...(e.field_locks ?? []), ...overrides.map((o) => o.field)])].sort();
-  const scrapeTaskId =
-    body.writeNfo || body.images
-      ? (
-          await enqueueScrape(d.ao, {
-            mediaEntityId: e.id,
-            trigger: "manual",
-            actor: { type: principal?.kind === "api_key" ? "api_key" : "user", id: principal?.id ?? null },
-          })
-        ).taskId
-      : null;
-  const result: FilesScrapeResult = {
-    mediaEntityId: e.id,
-    taskId: runId,
-    lockedFields,
-    nfo: body.writeNfo ? "queued" : "not_requested",
-    images: body.images ? "queued" : "not_requested",
-    scrapeTaskId,
-  };
-  await audit(d, principal, "scrape", { ...result });
-  return result;
-}
-
 /**
  * G5 批量刷新元数据:文件/目录归到其下已登记资产的作品,与显式作品合并去重,逐个入队 refreshMetadata。
- * 只对有 active 资产的作品提交;其余进 skipped(前端汇总"已提交 n 个作品,锁定字段不变")。
  */
 export async function scrapeBatch(
   d: FmDeps,
@@ -183,18 +125,6 @@ export async function scrapeBatch(
   const ents = list.length
     ? await db.selectFrom("media_entity").select(["id", "title", "field_locks"]).where("id", "in", list).execute()
     : [];
-  const active = new Set(
-    list.length
-      ? (
-          await db
-            .selectFrom("media_asset")
-            .select(["media_entity_id"])
-            .where("media_entity_id", "in", list)
-            .where("status", "=", "active")
-            .execute()
-        ).map((a) => a.media_entity_id)
-      : [],
-  );
   const locks = list.length
     ? await db
         .selectFrom("field_override")
@@ -210,34 +140,16 @@ export async function scrapeBatch(
       skipped.push({ ref: id, reason: "not_found" });
       continue;
     }
-    if (!active.has(id)) {
-      skipped.push({ ref: id, reason: "not_registered" });
-      continue;
-    }
-    const runId = body.write
-      ? (
-          await enqueueScrape(d.ao, {
-            mediaEntityId: id,
-            trigger: "manual",
-            actor: { type: principal?.kind === "api_key" ? "api_key" : "user", id: principal?.id ?? null },
-          })
-        ).taskId
-      : (
-          await d.ctx.engine.enqueue(
-            "refreshMetadata",
-            { mediaEntityId: id },
-            { idempotencyKey: `refreshMetadata:files-batch:${body.idempotencyKey}:${id}`, traceId: randomUUID() },
-          )
-        ).runId;
+    const { runId } = await d.ctx.engine.enqueue(
+      "refreshMetadata",
+      { mediaEntityId: id },
+      { idempotencyKey: `refreshMetadata:files-batch:${body.idempotencyKey}:${id}`, traceId: randomUUID() },
+    );
     const lockedFields = [
       ...new Set([...(e.field_locks ?? []), ...locks.filter((l) => l.media_entity_id === id).map((l) => l.field)]),
     ].sort();
     submitted.push({ mediaEntityId: id, title: e.title, taskId: runId, lockedFields });
   }
-  await audit(d, principal, "scrape_batch", {
-    write: body.write,
-    submitted: submitted.map((s) => s.mediaEntityId),
-    skipped,
-  });
+  await audit(d, principal, "scrape_batch", { submitted: submitted.map((s) => s.mediaEntityId), skipped });
   return { submitted, skipped };
 }

apps/server/src/modules/file-manager/routes.ts

@@ -15,8 +15,9 @@ import { listDir, listRoots, statEntry } from "./list.ts";
 import { mkdirIn, renameEntry } from "./mutate.ts";
 import { organizeFiles } from "./organize.ts";
 import { previewFile } from "./preview.ts";
-import { probeFiles, scrape, scrapeBatch } from "./probe.ts";
+import { probeFiles, scrapeBatch } from "./probe.ts";
 import { findRecords } from "./records.ts";
+import { scrapeFiles, scrapeRecords } from "./scrape.ts";
 import { suggestName } from "./suggest-name.ts";
 import { transfer } from "./transfer.ts";
 import { deleteFiles, listTrash, purgeTrash, restoreTrash } from "./trash.ts";
@@ -110,7 +111,11 @@ export function registerFileRoutes(app: ModuleApp, d: FmDeps) {
   mount(app, F.identify, async (c, { body }) =>
     identify(d, c.get("principal"), body, { wait: /^(1|true)$/i.test(c.req.query("wait") ?? "") }),
   );
-  mount(app, F.scrape, async (c, { body }) => idem(d, c, "scrape", body, () => scrape(d, c.get("principal"), body)));
+  // SCR-2:任意目录 / 视频文件就地刮削(识别 → 写 NFO 与图片);认不准的项返回候选
+  mount(app, F.scrape, async (c, { body }) =>
+    idem(d, c, "scrape", body, () => scrapeFiles(d, c.get("principal"), body)),
+  );
+  mount(app, F.scrapeRecords, async (_c, { query }) => scrapeRecords(d, query.ids));
   mount(app, F.probe, async (_c, { body }) => probeFiles(d, body));
   mount(app, F.records, async (_c, { query }) => findRecords(d, query));
   mount(app, F.download, async (c, { query }) => downloadFile(d, c.get("principal"), query), true);

apps/server/src/modules/file-manager/scrape.ts

@@ -0,0 +1,211 @@
+import type {
+  FileIdentifyCandidate,
+  FilesScrapeItem,
+  FilesScrapeNeedsChoice,
+  FilesScrapeRequest,
+  FilesScrapeResult,
+  ScrapeMatch,
+} from "@umt/contracts";
+import type { Principal } from "../../http/auth.ts";
+import {
+  entityScrapeMatch,
+  loadScrapeRecords,
+  queueScrape,
+  scrapeSettings,
+  waitScrapeRecords,
+} from "../acquisition-organize/services/scrape.ts";
+import { actorOfP, FilesError, type FmDeps, resolveIn } from "./common.ts";
+import { candidateKey, identifyCore } from "./identify.ts";
+import { audit } from "./ledger.ts";
+import { kindOfName } from "./list.ts";
+import { entitiesUnder } from "./records.ts";
+
+// SCR-2 文件管理「刮削」(照 MP 文件管理的刮削):任意目录 / 视频文件 → 识别作品 → 就地写 NFO 与图片。
+// 识别顺序:请求指定(选定的候选)> 目录里已登记文件所属的唯一作品 > 按路径识别(本地召回 + TMDB,resolveIdentity
+// 能自动确认且各视频一致时自动用)。认不准的项进 needsChoice(前 5 候选),选定后带 mediaEntityId / external 再提交。
+// 未登记也能刮削(直接用 TMDB 字段);addToLibrary 时顺手把 TMDB 条目加入媒体库。
+// 每项一条刮削记录,排 scrapeMedia;请求内最多等 transferWaitMs,没刮完的 pending=true,前端按 id 轮询。
+
+const RECOGNIZE_TIMEOUT_MS = 15_000;
+
+type Actor = { type: string; id: string | null };
+const actorOf = (p: Principal | undefined): Actor => ({
+  type: p?.kind === "api_key" ? "api_key" : "user",
+  id: p?.id ?? null,
+});
+
+/** 路径识别:各视频都能自动确认且是同一作品时返回它;否则返回候选(按出现次数与等级合并,前 5) */
+async function recognize(
+  d: FmDeps,
+  principal: Principal | undefined,
+  it: FilesScrapeItem,
+): Promise<{ pick: FileIdentifyCandidate | null; choice: Omit<FilesScrapeNeedsChoice, "root" | "path"> }> {
+  const r = await identifyCore(
+    d,
+    { items: [{ root: it.root, path: it.path }], minSizeBytes: 0 },
+    { mode: "remote", actorId: actorOfP(principal), signal: AbortSignal.timeout(RECOGNIZE_TIMEOUT_MS) },
+  );
+  const items = r.result.items;
+  const picks = items.map((i) => r.autoPicks.get(`${i.root}\u0000${i.path}`) ?? null);
+  const agreed = picks.length > 0 && picks.every((p) => p && p === picks[0]) ? picks[0] : null;
+  const LV = { high: 3, medium: 2, low: 1 } as const;
+  const merged = new Map<string, { c: FileIdentifyCandidate; n: number }>();
+  for (const i of items)
+    for (const c of i.candidates) {
+      const k = candidateKey(c);
+      const cur = merged.get(k);
+      if (!cur) merged.set(k, { c, n: 1 });
+      else {
+        cur.n += 1;
+        if (LV[c.level] > LV[cur.c.level]) cur.c = { ...c, posterUrl: c.posterUrl ?? cur.c.posterUrl };
+      }
+    }
+  const ranked = [...merged.values()].sort((a, b) => b.n - a.n || LV[b.c.level] - LV[a.c.level]).map((x) => x.c);
+  const pick = agreed ? (ranked.find((c) => candidateKey(c) === agreed) ?? null) : null;
+  const first = items[0]?.parsed;
+  return {
+    pick,
+    choice: {
+      title: first?.title ?? null,
+      candidates: ranked.slice(0, 5),
+      warnings: [...new Set(items.flatMap((i) => i.warnings))].slice(0, 5),
+    },
+  };
+}
+
+/** TMDB 条目 → 识别结果:已有本地作品(已接受的 TMDB 映射)就用它;addToLibrary 时建作品 */
+async function externalMatch(
+  d: FmDeps,
+  ext: { externalId: string; kind: "movie" | "tv"; title?: string; year?: number | null },
+  source: ScrapeMatch["source"],
+  addToLibrary: boolean,
+  actor: Actor,
+  entities: FilesScrapeResult["entities"],
+): Promise<ScrapeMatch> {
+  const local = await d.ctx.db
+    .selectFrom("external_identity")
+    .select(["media_entity_id"])
+    .where("provider", "=", "tmdb")
+    .where("entity_type", "=", ext.kind)
+    .where("external_id", "=", ext.externalId)
+    .where("status", "=", "accepted")
+    .where("content_unit_id", "is", null)
+    .executeTakeFirst();
+  let entityId = local?.media_entity_id ?? null;
+  if (!entityId && addToLibrary) {
+    entityId = await d.ao.ports.entityFromTmdb({ type: ext.kind, id: ext.externalId }, actor);
+    const e = await d.ctx.db.selectFrom("media_entity").select(["title"]).where("id", "=", entityId).executeTakeFirst();
+    entities.push({ mediaEntityId: entityId, title: e?.title ?? ext.title ?? ext.externalId, created: true });
+  }
+  if (entityId) {
+    const m = await entityScrapeMatch(d.ao, entityId, source);
+    if (m) return { ...m, provider: "tmdb", externalId: ext.externalId };
+  }
+  return {
+    mediaEntityId: null,
+    provider: "tmdb",
+    externalId: ext.externalId,
+    kind: ext.kind,
+    title: ext.title ?? `TMDB ${ext.externalId}`,
+    year: ext.year ?? null,
+    source,
+  };
+}
+
+export async function scrapeFiles(
+  d: FmDeps,
+  principal: Principal | undefined,
+  body: FilesScrapeRequest,
+): Promise<FilesScrapeResult> {
+  const db = d.ctx.db;
+  const actor = actorOf(principal);
+  const mode = body.mode ?? (await scrapeSettings(d.ao)).scrapeMode;
+  const recordIds: string[] = [];
+  const needsChoice: FilesScrapeResult["needsChoice"] = [];
+  const skipped: FilesScrapeResult["skipped"] = [];
+  const entities: FilesScrapeResult["entities"] = [];
+  for (const it of body.items) {
+    if (it.mediaEntityId && it.external) throw new FilesError("bad_request", "mediaEntityId 与 external 只能给一个");
+    const r = await resolveIn(d, it.root, it.path);
+    const ref = { root: it.root, path: r.rel };
+    if (r.def.kind === "library" && !r.def.modifiable) {
+      skipped.push({ ...ref, reason: "scrape_root_read_only" });
+      continue;
+    }
+    const isDir = !!r.lst?.isDirectory();
+    if (!isDir && !(r.lst?.isFile() && kindOfName(r.abs) === "video")) {
+      skipped.push({ ...ref, reason: "scrape_not_video" });
+      continue;
+    }
+    let match: ScrapeMatch | null = null;
+    if (it.mediaEntityId) {
+      match = await entityScrapeMatch(d.ao, it.mediaEntityId, "chosen");
+      if (!match) throw new FilesError("not_found", `作品不存在：${it.mediaEntityId}`);
+    } else if (it.external) {
+      match = await externalMatch(d, it.external, "chosen", body.addToLibrary, actor, entities);
+    } else {
+      // 目录里的文件已登记到唯一作品:直接用(重新刮削已入库的)
+      const owners = r.def.libraryRootId ? await entitiesUnder(db, r.def.libraryRootId, r.rel, isDir) : [];
+      if (owners.length === 1) match = await entityScrapeMatch(d.ao, owners[0]!, "library");
+      if (!match) {
+        let rec: Awaited<ReturnType<typeof recognize>>;
+        try {
+          rec = await recognize(d, principal, { ...it, path: r.rel });
+        } catch (e) {
+          if (e instanceof FilesError && e.reason === "bad_request") {
+            skipped.push({ ...ref, reason: "scrape_no_videos" });
+            continue;
+          }
+          throw e;
+        }
+        const c = rec.pick;
+        if (c?.mediaEntityId) match = await entityScrapeMatch(d.ao, c.mediaEntityId, "auto");
+        else if (c && c.provider === "tmdb" && c.externalId)
+          match = await externalMatch(
+            d,
+            { externalId: c.externalId, kind: c.kind === "movie" ? "movie" : "tv", title: c.title, year: c.year },
+            "auto",
+            body.addToLibrary,
+            actor,
+            entities,
+          );
+        if (!match) {
+          needsChoice.push({ ...ref, ...rec.choice });
+          continue;
+        }
+      }
+    }
+    const q = await queueScrape(d.ao, {
+      rootId: r.def.id,
+      rootPath: r.def.hostPath,
+      libraryRootId: r.def.libraryRootId,
+      path: r.rel,
+      match,
+      mode,
+      trigger: "manual",
+      actor,
+    });
+    recordIds.push(q.recordId);
+  }
+  const records = recordIds.length ? await waitScrapeRecords(d.ao, recordIds, d.options.transferWaitMs) : [];
+  await audit(d, principal, "scrape", {
+    items: body.items.map((i) => `${i.root}:${i.path}`).slice(0, 100),
+    mode,
+    recordIds,
+    needsChoice: needsChoice.map((n) => `${n.root}:${n.path}`),
+    skipped,
+  });
+  return {
+    records,
+    needsChoice,
+    skipped,
+    pending: records.some((r) => r.status === "queued" || r.status === "running"),
+    entities,
+  };
+}
+
+export async function scrapeRecords(d: FmDeps, idsParam: string) {
+  const ids = [...new Set(idsParam.split(",").map((s) => s.trim()))].filter((s) => /^[0-9a-f-]{36}$/i.test(s));
+  if (!ids.length || ids.length > 100) throw new FilesError("bad_request", "ids 需要 1–100 个刮削记录 id");
+  return { records: await loadScrapeRecords(d.ao, ids) };
+}