Pushed a small follow-up (921e314) to harden _anthropic_content_to_system_text against null content, which also resolves the review note from the bot.

Before this, a {"role": "system", "content": null} entry hit message.get("content", ""), which returns None (the key is present, so the "" default never applies), then fell through to str(content) and injected the literal string "None" into the system prompt. The same happened for None blocks inside list content.

The fix returns "" for null content and skips None blocks, so they are dropped instead of polluting the prompt:

if content is None:  # null content must not become the literal "None"
    return ""
...
if block is not None:
    parts.append(str(block))

Added test_system_role_message_with_null_content_ignored as a regression. Full suite is green:

python -m pytest studio/backend/tests/test_anthropic_messages.py -q
86 passed

Thanks for the fix, this is a clean approach. Hoisting in-array system messages into the top level system field is the right call since the downstream anthropic_messages_to_openai converter only reads system and has no system branch in its per message loop, so widening the role literal alone would have left the content stranded.