Marioario is closing a hidden failure point in chat history

When an assistant goes straight to looking something up, without first writing a sentence, that blank turn can be saved into the conversation record. On a later question, the AI provider may reject the whole reconstructed history rather than answer it.

Marioario's proposed fix removes empty or whitespace-only entries before the history is sent back to the model, while retaining the surrounding exchange. The result is less of the maddening pattern where an initial chat works, then a follow-up produces only a generic streaming error. The change comes with a reproducible case and regression coverage.

The practical point is reliability, not new AI capability: a conversation should survive an assistant choosing action before narration.