AI Assistant: investigate responses being wrapped on a code block instead of a normal paragraph

Sometimes the assistant gives a response wrapped around a code block, instead of the normal paragraph, resulting in blocks like this one:


---

The issue is hard to replicate, looks like a random issue.

No files, tests, or entry points are named, and the issue says the problem is difficult to reproduce. Start by reviewing the linked screenshot and investigating the assistant response-rendering path; done means identifying the cause of code-block wrapping and confirming normal paragraph responses no longer become blocks.

Written by the indexing model from the issue text.