New reminder tells some models that thinking and the reply share one output cap

What

Claude Code can now add a new kind of reminder, output_limit_note, to the conversation. It is a hidden note to the model, not something you type.

• It tells the model that its thinking and its reply share one cap of about a given number of tokens. Tokens are the small chunks of text the model reads and writes.
• It asks the model to plan its work instead of drafting a deliverable twice.
• It is only added for models chosen by the tengu_jaunty_plum configuration, and only when the cap applies.
• It is not added again if the same text is already in the conversation.
• It is delivered as a hidden user message, the same way existing reminders such as total_tokens_reminder and silent_turn_reminder are.

Nothing has been read about whether tengu_jaunty_plum is on.

Why

For the models it targets, this may change how the model handles long answers. It is meant to keep the model from running out of room partway through a long piece of work.