fix(llm-text): respect quoted reasoning tag delimiters (#6535)

Co-authored-by: Zhangchenhao-111 <285911435+Zhangchenhao-111@users.noreply.github.com>
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>

README.md

@@ -1754,7 +1754,7 @@ Rebuild with `make up` after changing the managed extension set. See
 
 Gateway-generated follow-up suggestions now normalize both plain-string model output and block/list-style rich content before parsing the JSON array response, so provider-specific content wrappers do not silently drop suggestions.
 
-Backend response cleanup preserves unrelated tag names such as `<think-tank>` and `<think:note>` in follow-up suggestions and polished drafts; only the exact `<think>` name (optionally followed by attributes) starts a reasoning block. Self-closing `<think/>` and `<think />` tags are empty reasoning blocks and leave the following answer intact.
+Backend response cleanup preserves unrelated tag names such as `<think-tank>` and `<think:note>` in follow-up suggestions and polished drafts; only the exact `<think>` name (optionally followed by attributes) starts a reasoning block. Self-closing `<think/>` and `<think />` tags are empty reasoning blocks and leave the following answer intact. Delimiters inside quoted attributes, such as `<think note=">"/>`, also leave the following answer intact.
 
 The Web UI composer can polish draft input before sending. The rewrite runs as a short Gateway LLM request using the `input_polish` model configuration, keeps slash skill prefixes such as `/data-analysis`, and only replaces the local draft after the user clicks the polish button; it does not create a thread run or persist a message.
 

backend/packages/harness/deerflow/utils/AGENTS.md

@@ -41,6 +41,8 @@ name; punctuated names such as `<think-tank>`, `<think:note>`, and `<think/other
 are ordinary content. The shared prefix regex owns this boundary rule; do not
 duplicate it in a manual re-check. Self-closing `<think/>` and `<think />` tags
 are empty reasoning blocks: remove the tag without consuming subsequent text.
+Both cleanup functions find opening delimiters outside quoted attributes; a
+quoted `>` cannot terminate the tag or hide an answer after a self-closing tag.
 Keep exact unfinished `<think` prefixes hidden in leading summaries. Coverage:
 `tests/test_think_tag_boundaries.py`.
 

backend/packages/harness/deerflow/utils/llm_text.py

@@ -11,12 +11,27 @@ _THINK_OPEN_PREFIX_RE = re.compile(r"<think(?=[\s>]|/>|$)", re.IGNORECASE)
 _THINK_CLOSE_PREFIX_RE = re.compile(r"</think", re.IGNORECASE)
 
 
+def _find_tag_end(text: str, start: int) -> int:
+    """Find the closing delimiter outside single- or double-quoted attributes."""
+    quote = ""
+    for index in range(start, len(text)):
+        char = text[index]
+        if quote:
+            if char == quote:
+                quote = ""
+        elif char in {"'", '"'}:
+            quote = char
+        elif char == ">":
+            return index
+    return -1
+
+
 def _find_think_open(text: str, start: int) -> tuple[int, int] | None:
     """Find the next complete opening tag without retrying its suffix at each prefix."""
     match = _THINK_OPEN_PREFIX_RE.search(text, start)
     if match is None:
         return None
-    end = text.find(">", match.end())
+    end = _find_tag_end(text, match.end())
     if end < 0:
         return None
     return match.start(), end + 1
@@ -99,7 +114,7 @@ def strip_leading_think_blocks(text: str) -> str:
         opening = _THINK_OPEN_PREFIX_RE.match(text, start)
         if opening is None:
             break
-        end = text.find(">", opening.end())
+        end = _find_tag_end(text, opening.end())
         if end >= 0 and _is_self_closing_think_tag(text, start, end + 1):
             start = end + 1
             continue

backend/tests/test_utils_llm_text.py

@@ -8,10 +8,28 @@ import pytest
 
 from deerflow.utils.llm_text import (
     extract_response_text,
+    strip_leading_think_blocks,
     strip_markdown_code_fence,
     strip_think_blocks,
 )
 
+
+@pytest.mark.parametrize("cleaner", [strip_think_blocks, strip_leading_think_blocks])
+@pytest.mark.parametrize("tag", ['<think note=">"/>', "<think note='>'/>", '<THINK note="a > b" />'])
+def test_self_closing_think_attributes_preserve_answer(cleaner, tag):
+    assert cleaner(tag + '{"answer":"yes"}') == '{"answer":"yes"}'
+
+
+@pytest.mark.parametrize("cleaner", [strip_think_blocks, strip_leading_think_blocks])
+def test_quoted_slash_angle_is_not_a_self_closing_tag(cleaner):
+    assert cleaner('<think note="/>">reasoning</think>answer') == "answer"
+
+
+@pytest.mark.parametrize("cleaner", [strip_think_blocks, strip_leading_think_blocks])
+def test_many_quoted_delimiters_in_empty_think_tag(cleaner):
+    assert cleaner('<think note="' + ">" * 100000 + '"/>answer') == "answer"
+
+
 # ---------------------------------------------------------------------------
 # strip_think_blocks
 # ---------------------------------------------------------------------------