0.1.14 上线前更改

2026-01-11 12:35:53 +08:00
parent be71849aa5
commit 725a60d3aa
44 changed files with 3427 additions and 310 deletions
@@ -0,0 +1,384 @@
+import supabaseAdmin from "@/lib/supabase/admin";
+import { applyMindmapOps, ensureMindmapUids, type MindmapOp, type MindmapTreeNode, type NodeRef } from "@/lib/mindmap/mindmapOps";
+import { readMindmapLocal, writeMindmapLocal } from "@/lib/mindmap/mindmapLocalStore";
+
+type SupabaseRouteClient = {
+  from: (table: string) => any;
+};
+
+export type OnlyOfficeSupabaseClient = SupabaseRouteClient;
+
+export type OnlyOfficeToolContext = {
+  userId: string;
+  documentId?: string;
+  // 来自前端（@ 选择/上传）的附件列表，优先使用（避免额外查询）
+  attachments?: Array<{ id: string; title: string; fileUrl: string; mimeType?: string | null }>;
+};
+
+const isRecord = (v: unknown): v is Record<string, unknown> =>
+  typeof v === "object" && v !== null && !Array.isArray(v);
+
+type ResolvedAttachment = { id: string; title: string; fileUrl: string; mimeType?: string | null };
+
+const resolveAttachment = (ctx: OnlyOfficeToolContext, ref: string): ResolvedAttachment | null => {
+  const s = String(ref || "").trim();
+  if (!s) return null;
+  const list = Array.isArray(ctx.attachments) ? ctx.attachments : [];
+  const byId = list.find((a) => String(a.id) === s);
+  if (byId) return { id: String(byId.id), title: String(byId.title || byId.id), fileUrl: String(byId.fileUrl || ""), mimeType: byId.mimeType ?? null };
+  const exactTitle = list.find((a) => String(a.title) === s);
+  if (exactTitle)
+    return { id: String(exactTitle.id), title: String(exactTitle.title || exactTitle.id), fileUrl: String(exactTitle.fileUrl || ""), mimeType: exactTitle.mimeType ?? null };
+  const fuzzy = list.find((a) => String(a.title || "").includes(s) || String(a.fileUrl || "").includes(s));
+  if (fuzzy) return { id: String(fuzzy.id), title: String(fuzzy.title || fuzzy.id), fileUrl: String(fuzzy.fileUrl || ""), mimeType: fuzzy.mimeType ?? null };
+  return null;
+};
+
+const parseStoragePath = (fileUrl: string) => {
+  try {
+    const url = new URL(fileUrl);
+    const segments = url.pathname.split("/").filter(Boolean);
+    const objectIdx = segments.findIndex((seg) => seg === "object");
+    if (objectIdx === -1 || objectIdx + 2 >= segments.length) return null;
+    if (segments[objectIdx + 1] === "public") {
+      const bucket = segments[objectIdx + 2];
+      const path = segments.slice(objectIdx + 3).join("/");
+      return { bucket, path };
+    }
+    if (segments[objectIdx + 1] === "sign") {
+      const bucket = segments[objectIdx + 2];
+      const path = segments.slice(objectIdx + 3).join("/");
+      return { bucket, path };
+    }
+    return null;
+  } catch {
+    return null;
+  }
+};
+
+const guessExt = (fileNameOrUrl: string) => {
+  const s = String(fileNameOrUrl || "").trim();
+  if (!s) return "";
+  const cleaned = s.split("?")[0].split("#")[0];
+  const parts = cleaned.split(".");
+  if (parts.length < 2) return "";
+  return String(parts[parts.length - 1] || "").toLowerCase();
+};
+
+const isPdf = (mimeType: string, fileName: string) => {
+  const m = String(mimeType || "").toLowerCase();
+  if (m.includes("pdf")) return true;
+  return guessExt(fileName) === "pdf";
+};
+
+const signForDownload = async (row: Record<string, unknown>) => {
+  const bucket = typeof row.bucket === "string" ? row.bucket : "";
+  const storagePath = typeof row.storage_path === "string" ? row.storage_path : "";
+  const fileUrl = typeof row.file_url === "string" ? row.file_url : "";
+  const fileName = typeof row.file_name === "string" ? row.file_name : undefined;
+
+  if (bucket && storagePath) {
+    const { data, error } = await supabaseAdmin.storage.from(bucket).createSignedUrl(storagePath, 60 * 60, { download: fileName });
+    if (error || !data?.signedUrl) throw new Error(error?.message ?? "生成签名 URL 失败");
+    return data.signedUrl;
+  }
+
+  const parsed = fileUrl ? parseStoragePath(fileUrl) : null;
+  if (!parsed) return fileUrl;
+
+  const { data, error } = await supabaseAdmin.storage.from(parsed.bucket).createSignedUrl(parsed.path, 60 * 60, { download: fileName });
+  if (error || !data?.signedUrl) throw new Error(error?.message ?? "生成签名 URL 失败");
+  return data.signedUrl;
+};
+
+const downloadBytes = async (url: string) => {
+  const resp = await fetch(url);
+  if (!resp.ok) throw new Error(`下载失败：${resp.status} ${resp.statusText}`);
+  return await resp.arrayBuffer();
+};
+
+const mineruParseContentList = async (args: { bytes: ArrayBuffer; filename: string; mimeType: string }) => {
+  const endpoint = (process.env.MINERU_ENDPOINT || "").trim().replace(/\/$/, "");
+  if (!endpoint) throw new Error("缺少 MINERU_ENDPOINT，无法解析 PDF 结构");
+
+  const form = new FormData();
+  form.append("files", new Blob([args.bytes], { type: args.mimeType || "application/octet-stream" }), args.filename || "file.pdf");
+  form.append("output_dir", "./output");
+  form.append("lang_list", "ch");
+  form.append("backend", "pipeline");
+  form.append("parse_method", "auto");
+  form.append("return_md", "false");
+  form.append("return_middle_json", "false");
+  form.append("return_model_output", "false");
+  form.append("return_content_list", "true");
+  form.append("return_images", "false");
+  form.append("response_format_zip", "false");
+
+  const resp = await fetch(`${endpoint}/file_parse`, { method: "POST", body: form });
+  if (!resp.ok) {
+    const text = await resp.text().catch(() => "");
+    throw new Error(`MinerU 解析失败：${resp.status} ${resp.statusText}${text ? `，${text.slice(0, 300)}` : ""}`);
+  }
+  const payload = (await resp.json().catch(() => null)) as unknown;
+  if (!isRecord(payload) || !isRecord(payload.results)) throw new Error("MinerU 返回格式不正确");
+  const keys = Object.keys(payload.results);
+  if (!keys.length) throw new Error("MinerU 返回为空");
+  const first = payload.results[keys[0]];
+  if (!isRecord(first)) throw new Error("MinerU 返回格式不正确（results.*）");
+  const raw = first.content_list;
+  if (typeof raw !== "string" || !raw.trim()) return [];
+  const parsed = JSON.parse(raw) as unknown;
+  if (!Array.isArray(parsed)) return [];
+  return parsed as Array<Record<string, unknown>>;
+};
+
+type ExtractedOutlineItem = { level: number; title: string; page?: number };
+
+const outlineFromContentList = (list: Array<Record<string, unknown>>) => {
+  const out: ExtractedOutlineItem[] = [];
+  const seen = new Set<string>();
+  for (const item of list) {
+    const levelRaw = item.text_level;
+    const textRaw = item.text;
+    const pageIdxRaw = item.page_idx;
+    if (typeof levelRaw !== "number" || !Number.isFinite(levelRaw)) continue;
+    const level = Math.max(1, Math.min(6, Math.floor(levelRaw)));
+    const title = String(textRaw ?? "").replace(/\s+/g, " ").trim();
+    if (!title) continue;
+    const pageIdx = typeof pageIdxRaw === "number" && Number.isFinite(pageIdxRaw) ? Math.max(0, Math.floor(pageIdxRaw)) : null;
+    const page = pageIdx === null ? undefined : pageIdx + 1;
+    const key = `${level}|${page ?? ""}|${title}`;
+    if (seen.has(key)) continue;
+    seen.add(key);
+    out.push({ level, title, page });
+  }
+  return out;
+};
+
+const findNodeByUid = (root: MindmapTreeNode, uid: string): MindmapTreeNode | null => {
+  const target = String(uid || "");
+  if (!target) return null;
+  const walk = (n: MindmapTreeNode): MindmapTreeNode | null => {
+    if (String(n?.data?.uid || "") === target) return n;
+    const children = Array.isArray(n.children) ? n.children : [];
+    for (const c of children) {
+      const hit = walk(c);
+      if (hit) return hit;
+    }
+    return null;
+  };
+  return walk(root);
+};
+
+const createUid = () => {
+  const anyCrypto = globalThis.crypto as unknown as { randomUUID?: () => string } | undefined;
+  if (anyCrypto?.randomUUID) return anyCrypto.randomUUID();
+  return Math.random().toString(36).slice(2);
+};
+
+const buildOutlineOps = (args: { parentUid: string; items: ExtractedOutlineItem[]; attachment: { assetId: string; fileUrl: string; title: string; mimeType?: string | null } }) => {
+  const ops: MindmapOp[] = [];
+  const parents: Record<number, string> = { 0: args.parentUid };
+
+  for (const item of args.items) {
+    const level = Math.max(1, Math.min(6, Math.floor(item.level)));
+    const parent = parents[level - 1] || args.parentUid;
+    const uid = createUid();
+    const page = item.page;
+
+    const refs: NodeRef[] = [
+      {
+        kind: "pdf",
+        assetId: args.attachment.assetId,
+        fileUrl: args.attachment.fileUrl,
+        page,
+        title: args.attachment.title,
+        snippet: page ? `第 ${page} 页` : undefined,
+      },
+    ];
+
+    const hyperlink = (() => {
+      const base = String(args.attachment.fileUrl || "").trim();
+      if (!base) return undefined;
+      if (page && isPdf(String(args.attachment.mimeType ?? ""), base)) return `${base}#page=${page}`;
+      return base;
+    })();
+
+    ops.push({
+      op: "addChild",
+      parentUid: parent,
+      node: {
+        uid,
+        text: item.title,
+        hyperlink,
+        refs,
+      },
+    });
+
+    parents[level] = uid;
+    for (let d = level + 1; d <= 10; d += 1) delete parents[d];
+  }
+
+  return ops;
+};
+
+export const createOnlyOfficeServerTools = (args: {
+  supabase: OnlyOfficeSupabaseClient;
+  ctx: OnlyOfficeToolContext;
+  allowedToolIds: Set<string>;
+}) => {
+  const asset_extract_outline = async (toolArgs: Record<string, unknown>) => {
+    const assetId = String(toolArgs.assetId ?? "").trim();
+    const attachmentRef = String(toolArgs.attachmentRef ?? "").trim();
+    const resolved = attachmentRef ? resolveAttachment(args.ctx, attachmentRef) : null;
+    const targetAssetId = assetId || resolved?.id || "";
+    if (!targetAssetId) throw new Error("缺少 assetId / attachmentRef");
+
+    const { data, error } = await args.supabase
+      .from("media_assets")
+      .select("id,file_name,file_url,mime_type,ocr_text,ocr_status,storage_path,bucket,updated_at,deleted_at,purged_at")
+      .eq("id", targetAssetId)
+      .is("deleted_at", null)
+      .limit(1)
+      .maybeSingle();
+    if (error) throw new Error(isRecord(error) && typeof error.message === "string" ? error.message : "读取附件失败");
+    const row = (data && typeof data === "object" ? (data as Record<string, unknown>) : null) ?? null;
+    if (!row) return { ok: true, found: false };
+
+    const fileName = String(row.file_name ?? "");
+    const fileUrl = String(row.file_url ?? "");
+    const mimeType = String(row.mime_type ?? "");
+    const ocrText = String(row.ocr_text ?? "");
+    const ocrStatus = String(row.ocr_status ?? "");
+    const textPreview = ocrText.trim().slice(0, 3000);
+
+    if (!isPdf(mimeType, fileName || fileUrl)) {
+      return {
+        ok: true,
+        found: true,
+        supported: false,
+        note: "当前仅优先支持 PDF（M3：PDF→大纲→导图）。Word/PPT 将在后续阶段通过 ONLYOFFICE /converter 或其它解析策略补齐。",
+        asset: { id: String(row.id ?? ""), fileName, fileUrl, mimeType, ocrStatus, updatedAt: row.updated_at ?? null },
+        hasOcrText: Boolean(ocrText.trim()),
+        textPreview,
+      };
+    }
+
+    // M3：PDF 优先使用 MinerU 的 content_list（含 page_idx + text_level），用于做可跳转大纲
+    let outline: ExtractedOutlineItem[] = [];
+    let strategy: "mineru_content_list" | "fallback_ocr_text" = "mineru_content_list";
+
+    try {
+      const signed = await signForDownload(row);
+      const bytes = await downloadBytes(signed);
+      const list = await mineruParseContentList({ bytes, filename: fileName || "file.pdf", mimeType: mimeType || "application/pdf" });
+      outline = outlineFromContentList(list);
+    } catch (e) {
+      // 兜底：返回 OCR 文本的一部分，至少让模型能“基于文本生成大纲”（但没有页码保障）
+      strategy = "fallback_ocr_text";
+      outline = [];
+      if (textPreview) outline.push({ level: 1, title: "（未解析到结构化标题：请结合 textPreview 自行生成提纲）" });
+    }
+
+    return {
+      ok: true,
+      found: true,
+      supported: true,
+      strategy,
+      asset: {
+        id: String(row.id ?? ""),
+        fileName,
+        fileUrl,
+        mimeType,
+        ocrStatus,
+        updatedAt: row.updated_at ?? null,
+      },
+      hasOcrText: Boolean(ocrText.trim()),
+      textPreview,
+      outline,
+      note:
+        outline.length > 0
+          ? `已提取大纲条目：${outline.length}。`
+          : strategy === "fallback_ocr_text"
+            ? "未能从 MinerU 获取结构化标题；请确认 MINERU_ENDPOINT 正常，或稍后重试。"
+            : "未提取到标题（该 PDF 可能没有明显标题结构）。",
+    };
+  };
+
+  const asset_to_mindmap = async (toolArgs: Record<string, unknown>) => {
+    const mindmapId = String(toolArgs.mindmapId ?? "").trim();
+    if (!mindmapId) throw new Error("缺少 mindmapId");
+    const parentUidArg = String(toolArgs.parentUid ?? "").trim();
+    const maxItemsRaw = Number(toolArgs.maxItems ?? 120);
+    const maxItems = Number.isFinite(maxItemsRaw) ? Math.max(10, Math.min(600, Math.floor(maxItemsRaw))) : 120;
+
+    const documentId = String(args.ctx.documentId ?? "").trim();
+    if (!documentId) throw new Error("缺少 documentId 上下文（OnlyOffice 工具需要落盘到指定文档）");
+
+    const outlineResult = (await asset_extract_outline(toolArgs)) as unknown;
+    if (!isRecord(outlineResult) || outlineResult.ok !== true) throw new Error("提取大纲失败");
+    if (outlineResult.found !== true) throw new Error("未找到附件");
+    if (outlineResult.supported !== true) throw new Error(String(outlineResult.note ?? "附件类型暂不支持"));
+
+    const outline = Array.isArray(outlineResult.outline) ? (outlineResult.outline as unknown[]) : [];
+    const items = outline
+      .map((x) =>
+        isRecord(x)
+          ? {
+              level: Number(x.level ?? 1),
+              title: String(x.title ?? ""),
+              ...(typeof x.page === "number" ? { page: x.page } : {}),
+            }
+          : null,
+      )
+      .filter((x): x is ExtractedOutlineItem => x !== null && x.title.trim().length > 0)
+      .slice(0, maxItems);
+    if (!items.length) throw new Error("未提取到可用大纲条目");
+
+    const asset = isRecord(outlineResult.asset) ? outlineResult.asset : null;
+    const attachment = {
+      assetId: String(asset?.id ?? ""),
+      title: String(asset?.fileName ?? "附件"),
+      fileUrl: String(asset?.fileUrl ?? ""),
+      mimeType: isRecord(asset) ? (asset.mimeType as string | null | undefined) : null,
+    };
+    if (!attachment.assetId) throw new Error("缺少附件 id");
+
+    const { data: baseRaw } = await readMindmapLocal(documentId, mindmapId);
+    const base = (baseRaw && typeof baseRaw === "object" ? (baseRaw as MindmapTreeNode) : null) ?? { data: { text: "中心主题" }, children: [] };
+    ensureMindmapUids(base);
+
+    const rootUid = String(base?.data?.uid || "");
+    const parentUid = parentUidArg || rootUid;
+    if (!parentUid) throw new Error("无法确定 parentUid（mindmap 根节点缺少 uid）");
+    if (!findNodeByUid(base, parentUid)) throw new Error("未找到 parentUid 对应节点");
+
+    const ops = buildOutlineOps({ parentUid, items, attachment });
+    const { data: nextData, applied, errors } = applyMindmapOps(base, ops);
+    await writeMindmapLocal(documentId, mindmapId, nextData, "OnlyOffice 生成导图");
+
+    return {
+      ok: true,
+      mindmapId,
+      parentUid,
+      applied,
+      errors,
+      ops,
+      meta: {
+        assetId: attachment.assetId,
+        fileName: attachment.title,
+        items: items.length,
+        strategy: String(outlineResult.strategy ?? ""),
+      },
+    };
+  };
+
+  const run = async (toolId: string, toolArgs: Record<string, unknown>) => {
+    if (!args.allowedToolIds.has(toolId)) throw new Error(`工具未被允许：${toolId}`);
+    if (toolId === "asset_extract_outline") return await asset_extract_outline(toolArgs);
+    if (toolId === "asset_to_mindmap") return await asset_to_mindmap(toolArgs);
+    throw new Error(`未知工具：${toolId}`);
+  };
+
+  return { run };
+};
@@ -239,6 +239,77 @@ export const builtinTools: AiAgentTool[] = [
    requiresConfirmation: true,
    isWriteTool: true,
  },
+  {
+    id: "asset_extract_outline",
+    displayName: "提取附件大纲（PDF 优先）",
+    modelDescription:
+      "从附件（当前优先 PDF）提取结构化大纲（含层级与页码）。用于“文档驱动导图/可跳转引用”（不写入）。",
+    inputSchemaText: `{ "assetId?": "string", "attachmentRef?": "string (可用附件 id/title/url 片段)" }`,
+    source: "builtin",
+    requiresConfirmation: false,
+    isWriteTool: false,
+  },
+  {
+    id: "asset_to_mindmap",
+    displayName: "附件生成思维导图（写入）",
+    modelDescription:
+      "把附件大纲转为思维导图节点并落盘（PDF 页码会写入 refs，并尽量生成可跳转 hyperlink）。这是写工具，执行前必须确认。",
+    inputSchemaText:
+      `{ "mindmapId": "string", "parentUid?": "string (默认根节点)", "assetId?": "string", "attachmentRef?": "string", "maxItems?": "number (10~600, 默认120)", "reason?": "string" }`,
+    source: "builtin",
+    requiresConfirmation: true,
+    isWriteTool: true,
+  },
+  {
+    id: "oo_get_selection",
+    displayName: "读取选区（OnlyOffice）",
+    modelDescription:
+      "读取 OnlyOffice 当前选区（用于增/删/改/查的基础能力：查/改前先读）。注意：该工具在客户端插件内执行。",
+    inputSchemaText: `{ "format?": "\"text\"|\"html\" (默认 text)" }`,
+    source: "builtin",
+    requiresConfirmation: false,
+    isWriteTool: false,
+  },
+  {
+    id: "oo_replace_selection",
+    displayName: "替换选区（OnlyOffice）",
+    modelDescription:
+      "用文本/HTML 替换 OnlyOffice 当前选区（基础改/删：删=传空字符串）。注意：该工具在客户端插件内执行。",
+    inputSchemaText: `{ "text": "string", "format?": "\"text\"|\"html\" (默认 text)", "reason?": "string" }`,
+    source: "builtin",
+    requiresConfirmation: true,
+    isWriteTool: true,
+  },
+  {
+    id: "oo_insert_text",
+    displayName: "插入文本（OnlyOffice）",
+    modelDescription:
+      "在 OnlyOffice 光标/选区位置插入文本（若存在选区通常会覆盖选区）。注意：该工具在客户端插件内执行。",
+    inputSchemaText: `{ "text": "string", "reason?": "string" }`,
+    source: "builtin",
+    requiresConfirmation: true,
+    isWriteTool: true,
+  },
+  {
+    id: "oo_insert_html",
+    displayName: "插入 HTML（OnlyOffice）",
+    modelDescription:
+      "在 OnlyOffice 光标/选区位置插入 HTML（用于保留简单格式）。注意：该工具在客户端插件内执行。",
+    inputSchemaText: `{ "html": "string", "reason?": "string" }`,
+    source: "builtin",
+    requiresConfirmation: true,
+    isWriteTool: true,
+  },
+  {
+    id: "oo_insert_image",
+    displayName: "插入图片（OnlyOffice）",
+    modelDescription:
+      "在 OnlyOffice 光标/选区位置插入图片（imageRef 可为附件 id 或 URL）。注意：该工具在客户端插件内执行。",
+    inputSchemaText: `{ "imageRef": "string (attachmentId 或 URL)", "width?": "number", "height?": "number", "reason?": "string" }`,
+    source: "builtin",
+    requiresConfirmation: true,
+    isWriteTool: true,
+  },
 ];

 export const builtinToolSets: AiAgentToolSet[] = [
@@ -262,6 +333,27 @@ export const builtinToolSets: AiAgentToolSet[] = [
    displayName: "媒体读取（OCR/元信息）",
    toolIds: ["image_read"],
  },
+  {
+    id: "toolset.onlyoffice_read",
+    displayName: "OnlyOffice/附件结构化读取（M3）",
+    toolIds: ["asset_extract_outline"],
+  },
+  {
+    id: "toolset.onlyoffice_write",
+    displayName: "OnlyOffice→思维导图（写入，M3）",
+    toolIds: ["asset_to_mindmap"],
+  },
+  {
+    id: "toolset.onlyoffice_editor",
+    displayName: "OnlyOffice 编辑器（选区读写）",
+    toolIds: [
+      "oo_get_selection",
+      "oo_replace_selection",
+      "oo_insert_text",
+      "oo_insert_html",
+      "oo_insert_image",
+    ],
+  },
  {
    id: "toolset.slash_write",
    displayName: "斜杠命令（写入）",