fix: guard LLM response access before dereferencing choices/content

autoagent/environment/markdown_browser/mdconvert.py

@@ -789,6 +789,8 @@ def _get_mlm_description(self, local_path, extension, client, model, prompt=None
         ]
 
         response = client.chat.completions.create(model=model, messages=messages)
+        if not response.choices or response.choices[0].message is None or response.choices[0].message.content is None:
+            raise ValueError("LLM returned empty or filtered response")
         return response.choices[0].message.content
 
 

autoagent/environment/mdconvert.py

@@ -796,6 +796,8 @@ def _get_mlm_description(self, local_path, extension, client, model, prompt=None
             ]
 
             response = client.chat.completions.create(model=model, messages=messages)
+            if not response.choices or response.choices[0].message is None or response.choices[0].message.content is None:
+                raise ValueError("LLM returned empty or filtered response")
             return response.choices[0].message.content
 
 

docs/translation_updater.py

@@ -61,6 +61,8 @@ def translate_content(content, target_lang):
         ],
     )
 
+    if not message.content or message.content[0].text is None:
+        raise ValueError("LLM returned empty or filtered response")
     return message.content[0].text