feat: handle malformed or invalid response from CodeInterpreterTool

src/crewai/tasks/guardrail_task.py

@@ -61,7 +61,7 @@ class GuardrailTask:
             "You are a expert Python developer"
             "You **must strictly** follow the task description, use the provided raw output as the input in your code. "
             "Your code must:\n"
-            "- Return results with: print((True, data)) on success, or print((False, 'very detailed error message')) on failure. Make sure the final output is beign assined to 'result' variable.\n"
+            "- Return results with: print((True, data)) on success, or print((False, 'very detailed error message')) on failure. Make sure the final output is being assined to 'result' variable.\n"
             "- Use the literal string of the task output (already included in your input) if needed.\n"
             "- Generate the code **following strictly** the task description.\n"
             "- Be valid Python 3 — executable as-is.\n"
@@ -151,7 +151,10 @@ class GuardrailTask:
             return False, result
 
         if isinstance(result, str):
-            result = ast.literal_eval(result)
+            try:
+                result = ast.literal_eval(result)
+            except Exception as e:
+                return False, f"Error parsing result: {str(e)}"
 
         return result
 

tests/test_task_guardrails.py

@@ -227,6 +227,11 @@ print(result)
             "expected_result": False,
             "expected_output": "Invalid output format",
         },
+        {
+            "output": "bla-bla-bla",
+            "expected_result": False,
+            "expected_output": "Error parsing result: malformed node or string on line 1",
+        },
     ],
 )
 @patch("crewai_tools.CodeInterpreterTool.run")
@@ -237,7 +242,7 @@ def test_guardrail_task_execute_code(mock_run, mock_llm, tool_run_output, task_o
 
     result = guardrail(task_output)
     assert result[0] == tool_run_output["expected_result"]
-    assert result[1] == tool_run_output["expected_output"]
+    assert tool_run_output["expected_output"] in result[1]
 
 
 @patch("crewai_tools.CodeInterpreterTool.run")
@@ -361,3 +366,24 @@ def test_guardrail_task_when_docker_is_available(mock_llm, task_output):
         guardrail(task_output)
 
     mock_init.assert_called_once_with(code=ANY, unsafe_mode=False)
+
+
+def test_guardrail_task_when_tool_output_is_not_valid(mock_llm, task_output):
+    guardrail = GuardrailTask(description="Test validation", llm=mock_llm)
+    with (
+        patch(
+            "crewai_tools.CodeInterpreterTool.__init__", return_value=None
+        ) as mock_init,
+        patch(
+            "crewai_tools.CodeInterpreterTool.run", return_value=(True, "Valid output")
+        ) as mock_run,
+        patch(
+            "subprocess.run",
+            return_value=True,
+        ),
+    ):
+        mock_init.return_value = None
+        mock_run.return_value = (True, "Valid output")
+        guardrail(task_output)
+
+    mock_init.assert_called_once_with(code=ANY, unsafe_mode=False)