ai-agent-book 精选快照（<2MB 代码与文档，来自 github.com/bojieli/ai-agent-book）

chapter2/system-hint/test_malformed_tool_json.py

@@ -0,0 +1,73 @@
+"""Regression: malformed tool-argument JSON must not abort the ReAct loop."""
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+from agent import SystemHintAgent, SystemHintConfig
+
+
+def test_execute_task_survives_malformed_tool_arguments_json():
+    config = SystemHintConfig(
+        enable_timestamps=False,
+        enable_tool_counter=False,
+        enable_todo_list=False,
+        enable_detailed_errors=False,
+        enable_system_state=False,
+        save_trajectory=False,
+    )
+    agent = SystemHintAgent(
+        api_key="test-key",
+        provider="kimi",
+        config=config,
+        verbose=False,
+    )
+
+    bad_call = SimpleNamespace(
+        id="call-bad",
+        function=SimpleNamespace(
+            name="read_file",
+            arguments='{"file_path": "x.txt",}',  # trailing comma
+        ),
+    )
+    tool_msg = SimpleNamespace(
+        content=None,
+        tool_calls=[bad_call],
+        model_dump=lambda: {
+            "role": "assistant",
+            "content": None,
+            "tool_calls": [
+                {
+                    "id": "call-bad",
+                    "type": "function",
+                    "function": {
+                        "name": "read_file",
+                        "arguments": '{"file_path": "x.txt",}',
+                    },
+                }
+            ],
+        },
+    )
+    final_msg = SimpleNamespace(
+        content="FINAL ANSWER: recovered",
+        tool_calls=None,
+        model_dump=lambda: {
+            "role": "assistant",
+            "content": "FINAL ANSWER: recovered",
+        },
+    )
+    tool_turn = SimpleNamespace(choices=[SimpleNamespace(message=tool_msg)])
+    final_turn = SimpleNamespace(choices=[SimpleNamespace(message=final_msg)])
+    agent.client = MagicMock()
+    agent.client.chat.completions.create = MagicMock(
+        side_effect=[tool_turn, final_turn]
+    )
+
+    result = agent.execute_task("read something", max_iterations=5)
+
+    assert "error" not in result or result.get("error") is None
+    assert result.get("final_answer") == "recovered" or "recovered" in str(
+        result.get("final_answer") or result
+    )
+    tool_roles = [m for m in agent.conversation_history if m.get("role") == "tool"]
+    assert tool_roles
+    assert "Invalid tool arguments" in tool_roles[0]["content"]
+    assert agent.client.chat.completions.create.call_count == 2