fix(paper2code): preserve fenced code in paper sections

CHANGELOG.md

@@ -5,6 +5,12 @@ Released versions are named in the entry title; the version on PyPI is
 [deepcode-hku](https://pypi.org/project/deepcode-hku/). 中文版见
 [CHANGELOG_ZH.md](CHANGELOG_ZH.md).
 
+## 2026-10-31 · Preserve code examples in paper sections
+
+- **Keep fenced code in its real paper section.** Python comments and other
+  hash-prefixed lines inside backtick or tilde code blocks no longer create
+  false headings or break the paper context passed to the workflow.
+
 ## 2026-10-29 · Lighter compaction, checked Goal completion, and folder browsing on the Web
 
 - **Clear old tool output first.** With `agents.defaults.microcompact` on,

CHANGELOG_ZH.md

@@ -4,6 +4,11 @@ DeepCode 的主要变化，最新的在前，每天一条。正式版本号写
 PyPI 上的包名是 [deepcode-hku](https://pypi.org/project/deepcode-hku/)。
 English: [CHANGELOG.md](CHANGELOG.md).
 
+## 2026-10-31 · 保留论文章节中的代码示例
+
+- **代码注释不再变成论文章节。** 反引号或波浪线代码块中的 Python 注释等
+  井号行保持为代码，不再生成错误标题或拆坏传给工作流的论文上下文。
+
 ## 2026-10-29 · 更轻的上下文压缩、目标完成前先检查、Web 端可浏览文件夹
 
 - **先清掉旧的工具输出。** 打开 `agents.defaults.microcompact` 后，长会话接近

tests/test_file_processor_fences.py

@@ -0,0 +1,67 @@
+"""Paper headings must not be inferred from fenced code examples."""
+
+import asyncio
+
+import pytest
+
+from utils.file_processor import FileProcessor
+
+
+@pytest.mark.parametrize("fence", ["```python", "~~~python", "   ```python"])
+def test_fenced_comments_stay_in_their_paper_section(fence):
+    marker = "~~~" if "~" in fence else "```"
+    code = f"{fence}\n# initialize the model\nx = 1\n{marker}"
+    sections = FileProcessor.parse_markdown_sections(
+        f"# Paper\n\n{code}\n\n## Results\nAccurate."
+    )
+    assert len(sections) == 1
+    assert sections[0]["title"] == "Paper"
+    assert sections[0]["content"] == code.strip()
+    assert [s["title"] for s in sections[0]["subsections"]] == ["Results"]
+
+
+@pytest.mark.parametrize(
+    "non_closer", ["```", "~~~~", "```` trailing text", "    ````"]
+)
+def test_only_a_matching_long_enough_fence_closes_code(non_closer):
+    code = f"````python\n{non_closer}\n# still a comment\n`````"
+    sections = FileProcessor.parse_markdown_sections(
+        f"# Paper\n{code}\n## Results\nDone."
+    )
+    assert len(sections) == 1
+    assert sections[0]["content"] == code.strip()
+    assert sections[0]["subsections"][0]["title"] == "Results"
+
+
+def test_unclosed_fence_keeps_later_hash_lines_as_code():
+    sections = FileProcessor.parse_markdown_sections(
+        "# Paper\n```python\n# code comment"
+    )
+    assert len(sections) == 1
+    assert sections[0]["content"] == "```python\n# code comment"
+
+
+def test_backticks_in_info_string_do_not_open_a_fence():
+    sections = FileProcessor.parse_markdown_sections(
+        "# Paper\n```not`a-fence\n## Results"
+    )
+    assert sections[0]["subsections"][0]["title"] == "Results"
+
+
+def test_process_file_input_preserves_code_before_standardization(tmp_path):
+    paper = tmp_path / "paper.md"
+    code = "```python\n# initialize the model\nx = 1\n```"
+    paper.write_text(f"# Paper\n{code}\n## Results\nDone.", encoding="utf-8")
+    result = asyncio.run(FileProcessor.process_file_input(str(paper)))
+    assert len(result["sections"]) == 1
+    assert code in result["standardized_text"]
+    assert result["sections"][0]["subsections"][0]["title"] == "Results"
+
+
+def test_crlf_fences_close_before_the_next_heading():
+    sections = FileProcessor.parse_markdown_sections(
+        "# Paper\r\n```python\r\n# code comment\r\n```\r\n## Results\r\nDone."
+    )
+    assert len(sections) == 1
+    assert sections[0]["subsections"][0]["title"] == "Results"
+    assert "# code comment" in sections[0]["content"]

utils/file_processor.py

@@ -109,8 +109,29 @@ def parse_markdown_sections(content: str) -> List[Dict[str, Union[str, int, List
         sections = []
         current_section = None
         current_content = []
+        fence = None
 
         for line in lines:
+            fence_match = re.match(r"^ {0,3}(`{3,}|~{3,})(.*)$", line)
+            if fence is not None:
+                if (
+                    fence_match
+                    and fence_match.group(1)[0] == fence[0]
+                    and len(fence_match.group(1)) >= len(fence)
+                    and not fence_match.group(2).strip(" \t\r")
+                ):
+                    fence = None
+                if current_section is not None:
+                    current_content.append(line)
+                continue
+            if fence_match and not (
+                fence_match.group(1)[0] == "`" and "`" in fence_match.group(2)
+            ):
+                fence = fence_match.group(1)
+                if current_section is not None:
+                    current_content.append(line)
+                continue
+
             # Check if line is a header
             header_match = re.match(r"^(#{1,6})\s+(.+)$", line)