From 7b72fd12476aedc06a993d92c4337e2ceb214bc7 Mon Sep 17 00:00:00 2001
From: Teknium <127238744+teknium1@users.noreply.github.com>
Date: Thu, 3 Sep 2026 02:44:08 -0700
Subject: [PATCH] fix(agent): stream cut mid tool-call markup no longer
 persists as assistant prose (#101899)

GLM-style models serialize tool calls as XML in the text channel; when the
stream drops mid-serialization with finish_reason=stop, the orphan
<arg_key>/<arg_value> fragment (or a bare unclosed <tool_call> opener)
matched neither the complete-block stripper nor the partial-stream guard
and was stored and displayed as ordinary assistant content.

strip_think_blocks (storage boundary) and the CLI display copy now strip
an unterminated block-boundary tool-call opener, or any line carrying
stray argument markup, to end of text. The response then reads as empty
and flows through the existing empty-retry path. Complete blocks and
inline prose mentions are unchanged.
---
 agent/agent_runtime_helpers.py                | 14 ++++++++++
 cli.py                                        |  9 +++++++
 .../test_strip_reasoning_tags_cli.py          | 26 +++++++++++++++++++
 3 files changed, 49 insertions(+)

diff --git a/agent/agent_runtime_helpers.py b/agent/agent_runtime_helpers.py
index 4927938655..1880698e81 100644
--- a/agent/agent_runtime_helpers.py
+++ b/agent/agent_runtime_helpers.py
@@ -100,6 +100,17 @@ _STRAY_TOOL_CALL_CLOSER_PATTERN = re.compile(
     re.IGNORECASE,
 )
 
+# A tool-call opener with no closer, or GLM-style argument markup
+# (<arg_key>/<arg_value>) outside any closed block, means the stream was
+# cut mid-serialization of a text-channel tool call (#101899). The call
+# can't be recovered; strip from the block-boundary opener (or the line
+# holding the first stray argument tag) to the end of the text.
+_UNTERMINATED_TOOL_CALL_PATTERN = re.compile(
+    rf'(?:^|\n)[ \t]*<(?:{"|".join(_TOOL_CALL_TAG_NAMES)})\b[^>]*>.*$'
+    r'|(?:^|\n)[^\n<]*</?arg_(?:key|value)\b.*$',
+    re.DOTALL | re.IGNORECASE,
+)
+
 
 def _ra():
     """Lazy ``run_agent`` reference for test-patch routing."""
@@ -1065,6 +1076,9 @@ def strip_think_blocks(agent, content: str) -> str:
     #     during streaming may still be valuable to the user; matches
     #     OpenClaw's intentional asymmetry.)
     content = _STRAY_TOOL_CALL_CLOSER_PATTERN.sub('', content)
+    # 3c. Tool-call openers or argument markup surviving 1b belong to a
+    #     block that never closed — a mid-serialization stream cut (#101899).
+    content = _UNTERMINATED_TOOL_CALL_PATTERN.sub('', content)
     return content
 
 
diff --git a/cli.py b/cli.py
index 20c160931f..19536ffe8d 100644
--- a/cli.py
+++ b/cli.py
@@ -314,6 +314,15 @@ def _strip_reasoning_tags(text: str) -> str:
         cleaned,
         flags=re.IGNORECASE,
     )
+    # Unterminated opener / stray <arg_key>/<arg_value> markup = stream cut
+    # mid tool-call serialization (#101899); strip to end of text.
+    cleaned = re.sub(
+        r'(?:^|\n)[ \t]*<(?:tool_call|tool_calls|tool_result|function_call|function_calls)\b[^>]*>.*$'
+        r'|(?:^|\n)[^\n<]*</?arg_(?:key|value)\b.*$',
+        '',
+        cleaned,
+        flags=re.DOTALL | re.IGNORECASE,
+    )
     return cleaned.strip()
 
 
diff --git a/tests/run_agent/test_strip_reasoning_tags_cli.py b/tests/run_agent/test_strip_reasoning_tags_cli.py
index 60525990c5..7d32c538cd 100644
--- a/tests/run_agent/test_strip_reasoning_tags_cli.py
+++ b/tests/run_agent/test_strip_reasoning_tags_cli.py
@@ -7,8 +7,21 @@ AIAgent instance. It must stay in sync with run_agent.py::_strip_think_blocks
 for tool-call tag coverage."""
 
 
+from agent.agent_runtime_helpers import strip_think_blocks
 from cli import _strip_reasoning_tags
 
+# GLM text-channel tool call cut mid-serialization by a stream drop (#101899):
+# the first key and call name never arrived, only orphan argument markup.
+_CUT_FRAGMENT = (
+    "Both gates started.\n"
+    "wait</arg_value>\n<arg_key>session_id</arg_key>\n<arg_value>abc</arg_value>\n"
+    "<arg_key>timeout</arg_key>\n<arg_value>59"
+)
+_COMPLETE_WITH_PROSE = (
+    "Use <function> in JS. The arg_key field maps to arg_value.\n"
+    "<tool_call>x<arg_key>a</arg_key><arg_value>1</arg_value></tool_call>\nDone."
+)
+
 
 class TestToolCallStripping:
     def test_tool_call_block_stripped(self):
@@ -26,3 +39,16 @@ class TestToolCallStripping:
     def test_empty_string(self):
         assert _strip_reasoning_tags("") == ""
 
+    def test_cut_tool_call_stripped_to_visible_prefix(self):
+        """Both strippers drop the unrecoverable tail; only prose survives."""
+        assert _strip_reasoning_tags(_CUT_FRAGMENT) == "Both gates started."
+        assert strip_think_blocks(None, _CUT_FRAGMENT).strip() == "Both gates started."
+        assert strip_think_blocks(None, "Waiting.\n<tool_call>process_manage").strip() == "Waiting."
+
+    def test_complete_block_and_inline_prose_mentions_untouched(self):
+        for out in (_strip_reasoning_tags(_COMPLETE_WITH_PROSE),
+                    strip_think_blocks(None, _COMPLETE_WITH_PROSE)):
+            assert "Use <function> in JS. The arg_key field maps to arg_value." in out
+            assert out.rstrip().endswith("Done.")
+            assert "<tool_call>" not in out
+