fix(agent): continue ollama glm truncation replies

run_agent.py

@@ -2103,6 +2103,59 @@ class AIAgent:
         content = re.sub(r'</?(?:think|thinking|reasoning|thought|REASONING_SCRATCHPAD)>\s*', '', content, flags=re.IGNORECASE)
         return content
 
+    @staticmethod
+    def _has_natural_response_ending(content: str) -> bool:
+        """Heuristic: does visible assistant text look intentionally finished?"""
+        if not content:
+            return False
+        stripped = content.rstrip()
+        if not stripped:
+            return False
+        if stripped.endswith("```"):
+            return True
+        return stripped[-1] in '.!?:)"\']}。！？：）】」』》'
+
+    def _is_ollama_glm_backend(self) -> bool:
+        """Detect the narrow backend family affected by Ollama/GLM stop misreports."""
+        model_lower = (self.model or "").lower()
+        provider_lower = (self.provider or "").lower()
+        if "glm" not in model_lower and provider_lower != "zai":
+            return False
+        if "ollama" in self._base_url_lower or ":11434" in self._base_url_lower:
+            return True
+        return bool(self.base_url and is_local_endpoint(self.base_url))
+
+    def _should_treat_stop_as_truncated(
+        self,
+        finish_reason: str,
+        assistant_message,
+        messages: Optional[list] = None,
+    ) -> bool:
+        """Detect conservative stop->length misreports for Ollama-hosted GLM models."""
+        if finish_reason != "stop" or self.api_mode != "chat_completions":
+            return False
+        if not self._is_ollama_glm_backend():
+            return False
+        if not any(
+            isinstance(msg, dict) and msg.get("role") == "tool"
+            for msg in (messages or [])
+        ):
+            return False
+        if assistant_message is None or getattr(assistant_message, "tool_calls", None):
+            return False
+
+        content = getattr(assistant_message, "content", None)
+        if not isinstance(content, str):
+            return False
+
+        visible_text = self._strip_think_blocks(content).strip()
+        if not visible_text:
+            return False
+        if len(visible_text) < 20 or not re.search(r"\s", visible_text):
+            return False
+
+        return not self._has_natural_response_ending(visible_text)
+
     def _looks_like_codex_intermediate_ack(
         self,
         user_message: str,
@@ -9038,6 +9091,17 @@ class AIAgent:
                         finish_reason = stop_reason_map.get(response.stop_reason, "stop")
                     else:
                         finish_reason = response.choices[0].finish_reason
+                        assistant_message = response.choices[0].message
+                        if self._should_treat_stop_as_truncated(
+                            finish_reason,
+                            assistant_message,
+                            messages,
+                        ):
+                            self._vprint(
+                                f"{self.log_prefix}⚠️  Treating suspicious Ollama/GLM stop response as truncated",
+                                force=True,
+                            )
+                            finish_reason = "length"
 
                     if finish_reason == "length":
                         self._vprint(f"{self.log_prefix}⚠️  Response truncated (finish_reason='length') - model hit max output tokens", force=True)