From 51f5909c08d8ad0466e6a6b2f566f46b043617db Mon Sep 17 00:00:00 2001
From: Dipesh Babu <dipeshmahato@outlook.com>
Date: Fri, 31 Jul 2026 15:02:41 -0400
Subject: [PATCH 1/2] Preserve DOCX manual line breaks

---
 sdk/typescript/src/knowledge-base.ts           |  1 +
 sdk/typescript/tests-ts/knowledge-base.test.ts | 11 +++++++----
 2 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/sdk/typescript/src/knowledge-base.ts b/sdk/typescript/src/knowledge-base.ts
index ad82a3c5..6aaaeeea 100644
--- a/sdk/typescript/src/knowledge-base.ts
+++ b/sdk/typescript/src/knowledge-base.ts
@@ -193,6 +193,7 @@ function extractDocx(path: string, bytes: Uint8Array): string {
     return decodeXml(
       xml
         .replace(/<\/(?:\w+:)?p\s*>/gu, "\n")
+        .replace(/<(?:\w+:)?br\b[^>]*\/>/gu, "\n")
         .replace(/<(?:\w+:)?tab\b[^>]*\/>/gu, "\t")
         .replace(/<[^>]+>/gu, ""),
     );
diff --git a/sdk/typescript/tests-ts/knowledge-base.test.ts b/sdk/typescript/tests-ts/knowledge-base.test.ts
index 79ca83c4..33b69fdb 100644
--- a/sdk/typescript/tests-ts/knowledge-base.test.ts
+++ b/sdk/typescript/tests-ts/knowledge-base.test.ts
@@ -41,10 +41,10 @@ async function extractedDocuments(path: string): Promise<string[]> {
   );
 }
 
-function docx(text: string): Uint8Array {
+function docx(text: string, secondLine?: string): Uint8Array {
   return zipSync({
     "word/document.xml": strToU8(
-      `<?xml version="1.0"?><w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p><w:r><w:t>${text}</w:t></w:r></w:p></w:body></w:document>`,
+      `<?xml version="1.0"?><w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p><w:r><w:t>${text}</w:t></w:r>${secondLine === undefined ? "" : `<w:br/><w:r><w:t>${secondLine}</w:t></w:r>`}</w:p></w:body></w:document>`,
     ),
   });
 }
@@ -111,14 +111,17 @@ describe("scan knowledge bases", () => {
       join(root, "architecture.pdf"),
       pdf("Payment service boundary"),
     );
-    await writeFile(join(root, "threat-model.docx"), docx("SSRF &amp; IDOR"));
+    await writeFile(
+      join(root, "threat-model.docx"),
+      docx("SSRF &amp; IDOR", "Review authentication"),
+    );
 
     const knowledgeBase = await prepareKnowledgeBase([root]);
     temporaryDirectories.push(knowledgeBase.path);
     const documents = await extractedDocuments(knowledgeBase.path);
 
     expect(documents).toContain("Payment service boundary");
-    expect(documents).toContain("SSRF & IDOR\n");
+    expect(documents).toContain("SSRF & IDOR\nReview authentication\n");
   });
 
   test("cleans up documents and rediscovers directory contents on later runs", async () => {

From 3fe6cf95c67f897e111089b8d0886639fa9f7d60 Mon Sep 17 00:00:00 2001
From: Dipesh Babu <dipeshmahato@outlook.com>
Date: Sat, 1 Aug 2026 10:27:00 -0400
Subject: [PATCH 2/2] Handle paired DOCX break elements

---
 sdk/typescript/src/knowledge-base.ts           |  2 +-
 sdk/typescript/tests-ts/knowledge-base.test.ts | 13 +++++++++++--
 2 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/sdk/typescript/src/knowledge-base.ts b/sdk/typescript/src/knowledge-base.ts
index 6aaaeeea..e150d937 100644
--- a/sdk/typescript/src/knowledge-base.ts
+++ b/sdk/typescript/src/knowledge-base.ts
@@ -193,7 +193,7 @@ function extractDocx(path: string, bytes: Uint8Array): string {
     return decodeXml(
       xml
         .replace(/<\/(?:\w+:)?p\s*>/gu, "\n")
-        .replace(/<(?:\w+:)?br\b[^>]*\/>/gu, "\n")
+        .replace(/<(?:\w+:)?br\b[^>]*>/gu, "\n")
         .replace(/<(?:\w+:)?tab\b[^>]*\/>/gu, "\t")
         .replace(/<[^>]+>/gu, ""),
     );
diff --git a/sdk/typescript/tests-ts/knowledge-base.test.ts b/sdk/typescript/tests-ts/knowledge-base.test.ts
index 33b69fdb..8404c5c9 100644
--- a/sdk/typescript/tests-ts/knowledge-base.test.ts
+++ b/sdk/typescript/tests-ts/knowledge-base.test.ts
@@ -41,10 +41,14 @@ async function extractedDocuments(path: string): Promise<string[]> {
   );
 }
 
-function docx(text: string, secondLine?: string): Uint8Array {
+function docx(
+  text: string,
+  secondLine?: string,
+  breakElement = "<w:br/>",
+): Uint8Array {
   return zipSync({
     "word/document.xml": strToU8(
-      `<?xml version="1.0"?><w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p><w:r><w:t>${text}</w:t></w:r>${secondLine === undefined ? "" : `<w:br/><w:r><w:t>${secondLine}</w:t></w:r>`}</w:p></w:body></w:document>`,
+      `<?xml version="1.0"?><w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p><w:r><w:t>${text}</w:t></w:r>${secondLine === undefined ? "" : `${breakElement}<w:r><w:t>${secondLine}</w:t></w:r>`}</w:p></w:body></w:document>`,
     ),
   });
 }
@@ -115,6 +119,10 @@ describe("scan knowledge bases", () => {
       join(root, "threat-model.docx"),
       docx("SSRF &amp; IDOR", "Review authentication"),
     );
+    await writeFile(
+      join(root, "paired-break.docx"),
+      docx("Authorization", "Review permissions", "<w:br></w:br>"),
+    );
 
     const knowledgeBase = await prepareKnowledgeBase([root]);
     temporaryDirectories.push(knowledgeBase.path);
@@ -122,6 +130,7 @@ describe("scan knowledge bases", () => {
 
     expect(documents).toContain("Payment service boundary");
     expect(documents).toContain("SSRF & IDOR\nReview authentication\n");
+    expect(documents).toContain("Authorization\nReview permissions\n");
   });
 
   test("cleans up documents and rediscovers directory contents on later runs", async () => {