diff --git a/client/src/components/conversation/MarkdownContent.tsx b/client/src/components/conversation/MarkdownContent.tsx
index f9986633..9cf8df0c 100644
--- a/client/src/components/conversation/MarkdownContent.tsx
+++ b/client/src/components/conversation/MarkdownContent.tsx
@@ -84,7 +84,12 @@ const HR_RE = /^\s*(-{3,}|\*{3,}|_{3,})\s*$/;
 const UL_RE = /^(\s*)([-*+])\s+(.*)$/;
 const OL_RE = /^(\s*)(\d+)\.\s+(.*)$/;
 const QUOTE_RE = /^\s*>\s?(.*)$/;
-const TABLE_DIVIDER_RE = /^\s*\|?\s*:?-{3,}:?\s*(\|\s*:?-{3,}:?\s*)+\|?\s*$/;
+const TABLE_DIVIDER_RE = /^\s*\|?\s*:?-+:?\s*(\|\s*:?-+:?\s*)+\|?\s*$/;
+
+/** A GFM table starts at a line holding a pipe that is followed by a divider. */
+function isTableStart(lines: string[], i: number): boolean {
+  return lines[i]!.includes("|") && i + 1 < lines.length && TABLE_DIVIDER_RE.test(lines[i + 1]!);
+}
 
 function splitTableRow(line: string): string[] {
   // Trim leading/trailing pipes, then split, respecting escaped pipes.
@@ -171,7 +176,7 @@ function parseBlocks(src: string): Block[] {
     }
 
     // Table: header line followed by an alignment divider
-    if (line.includes("|") && i + 1 < lines.length && TABLE_DIVIDER_RE.test(lines[i + 1]!)) {
+    if (isTableStart(lines, i)) {
       const header = splitTableRow(line);
       const aligns = parseAlignments(lines[i + 1]!);
       i += 2;
@@ -227,7 +232,8 @@ function parseBlocks(src: string): Block[] {
       continue;
     }
 
-    // Paragraph: collect until a blank line or the start of another block
+    // Paragraph: collect until a blank line or the start of another block —
+    // a table included, which (as in GFM) may follow prose with no blank line.
     const paraLines: string[] = [line];
     i++;
     while (i < lines.length) {
@@ -239,7 +245,8 @@ function parseBlocks(src: string): Block[] {
         HR_RE.test(nl) ||
         UL_RE.test(nl) ||
         OL_RE.test(nl) ||
-        QUOTE_RE.test(nl)
+        QUOTE_RE.test(nl) ||
+        isTableStart(lines, i)
       ) {
         break;
       }
@@ -300,7 +307,10 @@ function UrlLink({ url }: { url: string }) {
 
 /** A markdown link's file target as a path: `%20` decoded, `#L12` → `:12`. */
 function decodeHref(href: string): string {
-  let p = href.replace(/^file:\/\//i, "").replace(/#L(\d+)(?:-L?\d+)?$/, ":$1").replace(/#.*$/, "");
+  let p = href
+    .replace(/^file:\/\//i, "")
+    .replace(/#L(\d+)(?:-L?\d+)?$/, ":$1")
+    .replace(/#.*$/, "");
   try {
     p = decodeURIComponent(p);
   } catch {
@@ -383,13 +393,7 @@ function renderInline(text: string, baseKey = ""): React.ReactNode[] {
           {codeM[1]}
         </code>
       );
-      push(
-        matchPath(codeM[1]!) === codeM[1] ? (
-          <PathLink path={codeM[1]!}>{code}</PathLink>
-        ) : (
-          code
-        )
-      );
+      push(matchPath(codeM[1]!) === codeM[1] ? <PathLink path={codeM[1]!}>{code}</PathLink> : code);
       i += codeM[0].length;
       continue;
     }
@@ -440,9 +444,7 @@ function renderInline(text: string, baseKey = ""): React.ReactNode[] {
         (!SCHEME_RE.test(href) && !href.startsWith("#") && !href.startsWith("//"))
       ) {
         push(
-          <PathLink path={decodeHref(href)}>
-            {renderInline(linkM[1]!, `${baseKey}-l${n}`)}
-          </PathLink>
+          <PathLink path={decodeHref(href)}>{renderInline(linkM[1]!, `${baseKey}-l${n}`)}</PathLink>
         );
         i += linkM[0].length;
         continue;
diff --git a/client/src/components/conversation/__tests__/MarkdownContent.test.tsx b/client/src/components/conversation/__tests__/MarkdownContent.test.tsx
index 78c95dc7..50bb2114 100644
--- a/client/src/components/conversation/__tests__/MarkdownContent.test.tsx
+++ b/client/src/components/conversation/__tests__/MarkdownContent.test.tsx
@@ -153,4 +153,18 @@ describe("<MarkdownContent />", () => {
     expect(container.firstChild).not.toBeNull();
     expect(container.querySelectorAll("p, ul, ol, blockquote, pre, hr").length).toBe(0);
   });
+
+  it("renders a table that follows prose with no blank line (issue #83)", () => {
+    const text = 'Cách kiểm tra:\n| Token xoá | Số chỗ |\n|---|---|\n| `width="N"` | 29 |';
+    const { container } = render(<MarkdownContent text={text} />);
+    expect(container.querySelector("p")?.textContent).toBe("Cách kiểm tra:");
+    expect(container.querySelectorAll("th").length).toBe(2);
+    expect(container.querySelectorAll("tbody tr").length).toBe(1);
+    expect(container.textContent).not.toContain("|---|");
+  });
+
+  it("accepts a GFM divider with single dashes", () => {
+    const { container } = render(<MarkdownContent text={"| a | b |\n|-|:-:|\n| 1 | 2 |"} />);
+    expect(container.querySelectorAll("th").length).toBe(2);
+  });
 });