fix: 文档去重防止 API 30MB 限制报错

src/__tests__/document-dedup.test.ts

@@ -0,0 +1,91 @@
+/**
+ * Document Deduplication Tests
+ *
+ * Tests for deduplicateDocuments() which prevents duplicate PDF documents
+ * from exceeding the Anthropic API 30MB message size limit.
+ *
+ * Bug: In a thread where multiple messages quote/reply-to the same PDF file,
+ * each message independently downloads that PDF. Without dedup, the same 4.4MB
+ * PDF could be included 8+ times, pushing the payload to 36MB and triggering
+ * "message size exceeds 30.000MB limit" error.
+ */
+import { describe, it, expect } from 'vitest';
+import { deduplicateDocuments } from '../feishu/event-handler.js';
+import type { DocumentAttachment } from '../claude/types.js';
+
+// ============================================================
+// Helpers
+// ============================================================
+
+function makePdf(fileName: string, sizeBytes: number): DocumentAttachment {
+  // base64 encoded size is ~4/3 of raw, but we use string length as proxy
+  return {
+    data: 'A'.repeat(sizeBytes),
+    mediaType: 'application/pdf',
+    fileName,
+  };
+}
+
+// ============================================================
+// Tests
+// ============================================================
+
+describe('deduplicateDocuments', () => {
+  it('removes duplicate documents by fileName', () => {
+    const pdf1 = makePdf('resume.pdf', 1000);
+    const pdf2 = makePdf('resume.pdf', 1000); // same name
+    const pdf3 = makePdf('report.pdf', 2000);
+
+    const result = deduplicateDocuments([pdf1, pdf2, pdf3]);
+
+    expect(result).toHaveLength(2);
+    expect(result[0].fileName).toBe('resume.pdf');
+    expect(result[1].fileName).toBe('report.pdf');
+  });
+
+  it('preserves order — first occurrence wins', () => {
+    const current = makePdf('吴亮.pdf', 5000);
+    const history = makePdf('吴亮.pdf', 5000);
+
+    // Current message doc comes first (higher priority)
+    const result = deduplicateDocuments([current, history]);
+
+    expect(result).toHaveLength(1);
+    expect(result[0]).toBe(current); // same reference
+  });
+
+  it('enforces total size limit', () => {
+    const big1 = makePdf('big1.pdf', 12 * 1024 * 1024); // 12MB
+    const big2 = makePdf('big2.pdf', 12 * 1024 * 1024); // 12MB — would exceed 20MB total
+    const small = makePdf('small.pdf', 100);
+
+    const result = deduplicateDocuments([big1, big2, small]);
+
+    expect(result).toHaveLength(2);
+    expect(result[0].fileName).toBe('big1.pdf');
+    expect(result[1].fileName).toBe('small.pdf'); // big2 skipped, small fits
+  });
+
+  it('returns empty array for empty input', () => {
+    expect(deduplicateDocuments([])).toEqual([]);
+  });
+
+  it('returns single doc unchanged', () => {
+    const pdf = makePdf('only.pdf', 500);
+    const result = deduplicateDocuments([pdf]);
+    expect(result).toHaveLength(1);
+    expect(result[0]).toBe(pdf);
+  });
+
+  it('handles the exact bug scenario: same PDF from quoted parent + history', () => {
+    // This is the exact scenario that caused the 36MB error:
+    // Same "吴亮.pdf" (4.4MB) downloaded from both quoted parent and history
+    const fromQuotedParent = makePdf('吴亮 .pdf', 4_425_740);
+    const fromHistory = makePdf('吴亮 .pdf', 4_425_740);
+
+    const result = deduplicateDocuments([fromQuotedParent, fromHistory]);
+
+    expect(result).toHaveLength(1);
+    expect(result[0]).toBe(fromQuotedParent);
+  });
+});

src/feishu/event-handler.ts

@@ -1238,6 +1238,41 @@ function isTextFile(fileName: string): boolean {
 /** 最多从历史消息中下载的文件数 */
 const MAX_HISTORY_FILES = 3;
 
+/**
+ * 文档 payload 大小上限（base64 字节数）。
+ * Anthropic API 的 message size 上限为 30MB，预留 10MB 给 system prompt + 对话历史 + 图片等。
+ */
+const MAX_TOTAL_DOCUMENT_BYTES = 20 * 1024 * 1024;
+
+/**
+ * 按 fileName 去重 + 总大小截断，防止重复文档撑爆 API 30MB 限制。
+ * 优先保留靠前的文档（当前消息 > 历史消息）。
+ */
+export function deduplicateDocuments(docs: DocumentAttachment[]): DocumentAttachment[] {
+  const seen = new Set<string>();
+  const result: DocumentAttachment[] = [];
+  let totalBytes = 0;
+  for (const doc of docs) {
+    const key = doc.fileName;
+    if (seen.has(key)) {
+      logger.info({ fileName: doc.fileName }, 'Skipping duplicate document');
+      continue;
+    }
+    const docBytes = doc.data.length; // base64 string length ≈ bytes
+    if (totalBytes + docBytes > MAX_TOTAL_DOCUMENT_BYTES) {
+      logger.warn({ fileName: doc.fileName, totalBytes, docBytes, limit: MAX_TOTAL_DOCUMENT_BYTES }, 'Document payload size limit reached, skipping');
+      continue;
+    }
+    seen.add(key);
+    result.push(doc);
+    totalBytes += docBytes;
+  }
+  if (result.length < docs.length) {
+    logger.info({ original: docs.length, deduplicated: result.length, totalBytes }, 'Documents deduplicated');
+  }
+  return result;
+}
+
 /**
  * 从历史消息中下载文件附件（PDF → DocumentAttachment, 文本类 → 嵌入文本）
  *
@@ -1256,7 +1291,15 @@ async function downloadHistoryFiles(
   }
   if (refs.length === 0) return { documents: [], fileTexts: [] };
 
-  const toDownload = refs.slice(-MAX_HISTORY_FILES);
+  // 按 fileKey 去重（同一文件可能在多条历史消息中出现，如话题内引用同一文件）
+  const seenKeys = new Set<string>();
+  const uniqueRefs = refs.filter(ref => {
+    if (seenKeys.has(ref.fileKey)) return false;
+    seenKeys.add(ref.fileKey);
+    return true;
+  });
+
+  const toDownload = uniqueRefs.slice(-MAX_HISTORY_FILES);
   const MAX_PDF_SIZE = 30 * 1024 * 1024;
   const MAX_TEXT_SIZE = 1 * 1024 * 1024;
 
@@ -1839,9 +1882,10 @@ export async function executeClaudeTask(
     if (history.images && history.images.length > 0) {
       images = [...(history.images), ...(images ?? [])];
     }
-    // 合并历史消息中的文档（PDF）
+    // 合并历史消息中的文档（PDF），按 fileName 去重 + 大小截断
     if (history.documents && history.documents.length > 0) {
-      documents = [...(history.documents), ...(documents ?? [])];
+      // 当前消息的文档优先（放前面），历史文档补充
+      documents = deduplicateDocuments([...(documents ?? []), ...(history.documents)]);
     }
     // 合并历史消息中的文本文件内容到 prompt
     if (history.fileTexts && history.fileTexts.length > 0) {
@@ -2380,9 +2424,9 @@ export async function executeDirectTask(
     if (history.images && history.images.length > 0) {
       images = [...(history.images), ...(images ?? [])];
     }
-    // 合并历史消息中的文档（PDF）
+    // 合并历史消息中的文档（PDF），按 fileName 去重 + 大小截断
     if (history.documents && history.documents.length > 0) {
-      documents = [...(history.documents), ...(documents ?? [])];
+      documents = deduplicateDocuments([...(documents ?? []), ...(history.documents)]);
     }
     // 合并历史消息中的文本文件内容到 prompt
     if (history.fileTexts && history.fileTexts.length > 0) {