What

Guard LLM response access in three files before dereferencing .choices[0] or .content[0].

Why

response.choices[0].message.content is typed Optional[str] in the OpenAI SDK — it returns None (not raises) when finish_reason is "tool_calls" or the provider applies content filtering. Similarly, the Anthropic SDK can return an empty content list on filtered responses, making message.content[0].text unsafe without a length check.

autoagent/environment/mdconvert.py and autoagent/environment/markdown_browser/mdconvert.py (OpenAI):

# before
return response.choices[0].message.content

# after
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 (Anthropic):

# before
return message.content[0].text

# after
if not message.content or message.content[0].text is None:
    raise ValueError("LLM returned empty or filtered response")
return message.content[0].text

Test plan

• Verify translate_content() raises ValueError when given a mock message.content = []
• Verify the mdconvert LLM path raises ValueError when given a mock response.choices = []

🤖 Generated with Claude Code