fix(agent): continue ollama glm truncation replies

tests/run_agent/test_run_agent.py

@@ -2202,6 +2202,114 @@ class TestRunConversation:
         assert second_call_messages[-1]["role"] == "user"
         assert "truncated by the output length limit" in second_call_messages[-1]["content"]
 
+    def test_ollama_glm_stop_after_tools_without_terminal_boundary_requests_continuation(self, agent):
+        """Ollama-hosted GLM responses can misreport truncated output as stop."""
+        self._setup_agent(agent)
+        agent.base_url = "http://localhost:11434/v1"
+        agent._base_url_lower = agent.base_url.lower()
+        agent.model = "glm-5.1:cloud"
+
+        tool_turn = _mock_response(
+            content="",
+            finish_reason="tool_calls",
+            tool_calls=[_mock_tool_call(name="web_search", arguments="{}", call_id="c1")],
+        )
+        misreported_stop = _mock_response(
+            content="Based on the search results, the best next",
+            finish_reason="stop",
+        )
+        continued = _mock_response(
+            content=" step is to update the config.",
+            finish_reason="stop",
+        )
+        agent.client.chat.completions.create.side_effect = [
+            tool_turn,
+            misreported_stop,
+            continued,
+        ]
+
+        with (
+            patch("run_agent.handle_function_call", return_value="search result"),
+            patch.object(agent, "_persist_session"),
+            patch.object(agent, "_save_trajectory"),
+            patch.object(agent, "_cleanup_task_resources"),
+        ):
+            result = agent.run_conversation("hello")
+
+        assert result["completed"] is True
+        assert result["api_calls"] == 3
+        assert (
+            result["final_response"]
+            == "Based on the search results, the best next step is to update the config."
+        )
+
+        third_call_messages = agent.client.chat.completions.create.call_args_list[2].kwargs["messages"]
+        assert third_call_messages[-1]["role"] == "user"
+        assert "truncated by the output length limit" in third_call_messages[-1]["content"]
+
+    def test_ollama_glm_stop_with_terminal_boundary_does_not_continue(self, agent):
+        """Complete Ollama/GLM responses should not be reclassified as truncated."""
+        self._setup_agent(agent)
+        agent.base_url = "http://localhost:11434/v1"
+        agent._base_url_lower = agent.base_url.lower()
+        agent.model = "glm-5.1:cloud"
+
+        tool_turn = _mock_response(
+            content="",
+            finish_reason="tool_calls",
+            tool_calls=[_mock_tool_call(name="web_search", arguments="{}", call_id="c1")],
+        )
+        complete_stop = _mock_response(
+            content="Based on the search results, the best next step is to update the config.",
+            finish_reason="stop",
+        )
+        agent.client.chat.completions.create.side_effect = [tool_turn, complete_stop]
+
+        with (
+            patch("run_agent.handle_function_call", return_value="search result"),
+            patch.object(agent, "_persist_session"),
+            patch.object(agent, "_save_trajectory"),
+            patch.object(agent, "_cleanup_task_resources"),
+        ):
+            result = agent.run_conversation("hello")
+
+        assert result["completed"] is True
+        assert result["api_calls"] == 2
+        assert (
+            result["final_response"]
+            == "Based on the search results, the best next step is to update the config."
+        )
+
+    def test_non_ollama_stop_without_terminal_boundary_does_not_continue(self, agent):
+        """The stop->length workaround should stay scoped to Ollama/GLM backends."""
+        self._setup_agent(agent)
+        agent.base_url = "https://api.openai.com/v1"
+        agent._base_url_lower = agent.base_url.lower()
+        agent.model = "gpt-4o-mini"
+
+        tool_turn = _mock_response(
+            content="",
+            finish_reason="tool_calls",
+            tool_calls=[_mock_tool_call(name="web_search", arguments="{}", call_id="c1")],
+        )
+        normal_stop = _mock_response(
+            content="Based on the search results, the best next",
+            finish_reason="stop",
+        )
+        agent.client.chat.completions.create.side_effect = [tool_turn, normal_stop]
+
+        with (
+            patch("run_agent.handle_function_call", return_value="search result"),
+            patch.object(agent, "_persist_session"),
+            patch.object(agent, "_save_trajectory"),
+            patch.object(agent, "_cleanup_task_resources"),
+        ):
+            result = agent.run_conversation("hello")
+
+        assert result["completed"] is True
+        assert result["api_calls"] == 2
+        assert result["final_response"] == "Based on the search results, the best next"
+
     def test_length_thinking_exhausted_skips_continuation(self, agent):
         """When finish_reason='length' but content is only thinking, skip retries."""
         self._setup_agent(agent)