From 81e7634ecd948d9a007a0d97d9a5c9cb316beaac Mon Sep 17 00:00:00 2001
From: Tony Coder <407243179@qq.com>
Date: Mon, 17 Aug 2026 10:57:23 +0000
Subject: [PATCH] fix(indexing): stop prefixing markdown chunks with the string
 "undefined"

markdownChunker recurses one header level at a time and prepends each
section's header to every chunk it produces. When a section has no
header, findHeader returns undefined and the template literal
interpolated it as the literal text "undefined", so headerless docs
were indexed with up to five bogus "undefined" lines at the top of
every chunk.

Only prepend the header when the section actually has one.
---
 core/indexing/chunk/markdown.ts        |  4 +-
 core/indexing/chunk/markdown.vitest.ts | 77 ++++++++++++++++++++++++++
 2 files changed, 80 insertions(+), 1 deletion(-)
 create mode 100644 core/indexing/chunk/markdown.vitest.ts

diff --git a/core/indexing/chunk/markdown.ts b/core/indexing/chunk/markdown.ts
index 5a0df6776d8..b3b3765fdf9 100644
--- a/core/indexing/chunk/markdown.ts
+++ b/core/indexing/chunk/markdown.ts
@@ -133,7 +133,9 @@ export async function* markdownChunker(
       hLevel + 1,
     )) {
       yield {
-        content: `${section.header}\n${chunk.content}`,
+        content: section.header
+          ? `${section.header}\n${chunk.content}`
+          : chunk.content,
         startLine: section.startLine + chunk.startLine,
         endLine: section.startLine + chunk.endLine,
         otherMetadata: {
diff --git a/core/indexing/chunk/markdown.vitest.ts b/core/indexing/chunk/markdown.vitest.ts
new file mode 100644
index 00000000000..e7eb791b8f7
--- /dev/null
+++ b/core/indexing/chunk/markdown.vitest.ts
@@ -0,0 +1,77 @@
+import { beforeEach, describe, expect, it, vi } from "vitest";
+
+vi.mock("../../llm/countTokens", () => ({
+  countTokens: (content: string) => content.length,
+  countTokensAsync: async (content: string) => content.length,
+}));
+
+import { ChunkWithoutID } from "../../index";
+
+import { markdownChunker } from "./markdown";
+
+async function collectChunks(
+  content: string,
+  maxChunkSize: number,
+  hLevel: number,
+): Promise<ChunkWithoutID[]> {
+  const chunks: ChunkWithoutID[] = [];
+  for await (const chunk of markdownChunker(content, maxChunkSize, hLevel)) {
+    chunks.push(chunk);
+  }
+  return chunks;
+}
+
+describe("markdownChunker", () => {
+  beforeEach(() => {
+    vi.clearAllMocks();
+  });
+
+  it("should not prefix chunks with 'undefined' when a section has no header", async () => {
+    const content = Array.from(
+      { length: 10 },
+      (_, i) => `Line ${i} of a markdown document that has no headers at all.`,
+    ).join("\n");
+
+    const chunks = await collectChunks(content, 100, 0);
+
+    expect(chunks.length).toBeGreaterThan(0);
+    for (const chunk of chunks) {
+      expect(chunk.content).not.toContain("undefined");
+    }
+  });
+
+  it("should not prefix headerless sub-sections with 'undefined'", async () => {
+    const content = [
+      "## Installation",
+      ...Array.from(
+        { length: 10 },
+        (_, i) => `Installation step ${i} with a bit of extra detail.`,
+      ),
+    ].join("\n");
+
+    const chunks = await collectChunks(content, 120, 1);
+
+    expect(chunks.length).toBeGreaterThan(0);
+    for (const chunk of chunks) {
+      expect(chunk.content).not.toContain("undefined");
+      expect(chunk.content.startsWith("## Installation\n")).toBe(true);
+    }
+  });
+
+  it("should still prepend the section header to each chunk", async () => {
+    const content = [
+      "## Installation",
+      ...Array.from(
+        { length: 10 },
+        (_, i) => `Installation step ${i} with a bit of extra detail.`,
+      ),
+    ].join("\n");
+
+    const chunks = await collectChunks(content, 120, 1);
+
+    expect(chunks.length).toBeGreaterThan(1);
+    for (const chunk of chunks) {
+      expect(chunk.otherMetadata?.title).toBe("Installation");
+    }
+  });
+});