Summary

• Normalize plain string message content into { text: ... } before it enters the translation pipeline.
• Replace the corresponding entry in translatedMessages with the normalized translatable message so async translation responses update the translated clone.
• Remove the previous @ts-ignore by constructing the TranslatableMessage through explicit text/developer branches.

Bug

Message.content can be either an array of content objects or a plain string. The translation queue previously used message.content[0] for every text/developer message. For plain strings, that indexes the first character, so "Hello" became "H" instead of a text content object.

That malformed shape meant translation requests were built from the wrong content and downstream logic expected object fields such as text on a string character.

Fix

The translation queue now calls the existing getContentFromContentOrString helper, which converts string content into { text: fullString } and preserves array-backed content by selecting the first content object. The normalized message is assigned back into translatedMessages[i] before translation starts, ensuring response handling mutates the translated clone rather than a temporary object.

Verification

• Confirmed the problematic message.content[0] expression is no longer used in the translation queue.
• Ran pnpm run build successfully.