diff --git a/.github/aicodingflow-tests/test_validate_pr_body.py b/.github/aicodingflow-tests/test_validate_pr_body.py
index abe209e..7235607 100644
--- a/.github/aicodingflow-tests/test_validate_pr_body.py
+++ b/.github/aicodingflow-tests/test_validate_pr_body.py
@@ -49,6 +49,57 @@ def test_html_comment_inside_fenced_code_is_literal_content(self) -> None:
         body = "## Summary\nChange.\n## Testing\n```html\n<!-- test fixture -->\n```\n## Risk\nLow.\n## How to revert\nRevert."
         validator.validate_body(body)
 
+    def test_accepts_plain_atx_h2(self) -> None:
+        body = "## Summary\nChange.\n## Testing\nPassed.\n## Risk\nLow.\n## How to revert\nRevert."
+        validator.validate_body(body)
+        self.assertEqual(
+            sorted(validator.section_contents(body)),
+            ["how to revert", "risk", "summary", "testing"],
+        )
+
+    def test_accepts_closing_hashes(self) -> None:
+        body = "## Summary ##\nChange.\n## Testing ##\nPassed.\n## Risk ##\nLow.\n## How to revert ##\nRevert."
+        validator.validate_body(body)
+        self.assertNotIn("summary ##", validator.section_contents(body))
+
+    def test_accepts_up_to_three_leading_spaces(self) -> None:
+        for indent in ("", " ", "  ", "   "):
+            with self.subTest(indent=len(indent)):
+                body = (
+                    f"{indent}## Summary\nChange.\n"
+                    f"{indent}## Testing\nPassed.\n"
+                    f"{indent}## Risk\nLow.\n"
+                    f"{indent}## How to revert\nRevert."
+                )
+                validator.validate_body(body)
+
+    def test_rejects_four_space_indented_heading(self) -> None:
+        body = "    ## Summary\nChange.\n## Testing\nPassed.\n## Risk\nLow.\n## How to revert\nRevert."
+        with self.assertRaisesRegex(SystemExit, "Summary"):
+            validator.validate_body(body)
+
+    def test_rejects_third_level_heading(self) -> None:
+        body = "### Summary\nChange.\n## Testing\nPassed.\n## Risk\nLow.\n## How to revert\nRevert."
+        with self.assertRaisesRegex(SystemExit, "Summary"):
+            validator.validate_body(body)
+
+    def test_rejects_heading_without_space(self) -> None:
+        body = "##Summary\nChange.\n## Testing\nPassed.\n## Risk\nLow.\n## How to revert\nRevert."
+        with self.assertRaisesRegex(SystemExit, "Summary"):
+            validator.validate_body(body)
+
+    def test_hash_inside_heading_is_not_a_closing_sequence(self) -> None:
+        self.assertEqual(validator.normalize_heading("C#"), "c#")
+        self.assertEqual(validator.normalize_heading("Summary ##"), "summary")
+        self.assertEqual(validator.normalize_heading("Summary ## extra"), "summary ## extra")
+        self.assertIsNone(validator.normalize_heading(""))
+
+    def test_empty_h2_does_not_fill_previous_section(self) -> None:
+        body = "## Summary\n##   \n## Testing\nPassed.\n## Risk\nLow.\n## How to revert\nRevert."
+        self.assertEqual(validator.section_contents(body)["summary"], "")
+        with self.assertRaisesRegex(SystemExit, "Summary"):
+            validator.validate_body(body)
+
 
 if __name__ == "__main__":
     unittest.main()
diff --git a/.github/scripts/validate_pr_body.py b/.github/scripts/validate_pr_body.py
index 3558d0c..ad1a6af 100644
--- a/.github/scripts/validate_pr_body.py
+++ b/.github/scripts/validate_pr_body.py
@@ -9,10 +9,20 @@
 
 
 REQUIRED_SECTIONS = ("Summary", "Testing", "Risk", "How to revert")
-HEADING = re.compile(r"^##[ \t]+(.+?)[ \t]*$", re.MULTILINE)
+# CommonMark ATX heading: up to three leading spaces, then `##` plus space/tab.
+HEADING = re.compile(r"^ {0,3}##[ \t]+(.+?)[ \t]*$")
+# A closing sequence is only stripped when whitespace-separated and at line end,
+# so `## C#` keeps its hash while `## Summary ##` normalises to `summary`.
+CLOSING_HASHES = re.compile(r"[ \t]+#+$")
 FENCE_OPEN = re.compile(r"^ {0,3}(`{3,}|~{3,})([^\r\n]*)$")
 
 
+def normalize_heading(text: str) -> str | None:
+    """Return the comparable section name for an ATX H2 heading text."""
+    name = CLOSING_HASHES.sub("", text.strip()).strip().lower()
+    return name or None
+
+
 def strip_html_comments(line: str, in_comment: bool) -> tuple[str, bool]:
     visible = []
     offset = 0
@@ -57,9 +67,10 @@ def parse_sections(body: str) -> tuple[dict[str, str], dict[str, int]]:
             continue
         heading = HEADING.fullmatch(line)
         if heading:
-            current_section = heading.group(1).strip().lower()
-            sections[current_section] = []
-            headings[current_section] = line_number
+            current_section = normalize_heading(heading.group(1))
+            if current_section is not None:
+                sections[current_section] = []
+                headings[current_section] = line_number
         elif current_section is not None:
             sections[current_section].append(line)
     return {name: "\n".join(lines).strip() for name, lines in sections.items()}, headings