We fixed a problem where an AI response could fail before the conversation started. Missing descriptions for writing-policy settings were stopping the preparation of the AI's tools.
We also fixed an unnecessary approval prompt when the AI was only reading writing instructions, which could interrupt the path to saving a manuscript or reference.
Internal identifiers could also appear in progress and response descriptions. We have adjusted the AI's instructions to describe its targets using manuscript headings and reference names.
When a name has not been checked yet, the AI is instructed to use a description such as “the selected manuscript” or “the reference” instead of inventing a name.