Feature request for product/service
Cursor IDE
Describe the request
I use Cursor daily for real work: code, files, contracts, and long agent sessions. Please keep the agent behavior. Tool use, edits, git safety, and the other operational instructions are fine. I am not asking to turn the agent into an unrestricted chatbot.
The problem is only the wording of the message I read.
On the same turn, the reasoning text is fluent and specific. The final answer is a second draft: shorter, connectives removed, sentences rebuilt into a report. In English this is mostly a change of tone. In Chinese it breaks the sentence. Words like 因为, 所以, and 但是 are syntax, not decoration. After the rewrite, the answer is on-topic and hard to read. It slows me down and I lose precision. I am considering another coding agent only because of this, even though the usage allowance and the project context I have here are worth keeping.
This is not a request for a new style or a “reply in Chinese” rule. User Rules and always-applied project rules sit beside the built-in instructions and do not replace them. Staff have already said the built-in prompt cannot be disabled (forum 156051, 163434, 171733). Hooks can see the reasoning and the reply, but they cannot replace the message on screen. Asking the same model to translate or polish the answer sends it through the same rewrite again.
Please add one setting for the user-visible reply:
- Off by default, so the current behavior stays.
- When on, the reply uses the model’s own wording in the user’s language. Do not run a second stylistic pass. Do not force a blog-post shape, a lead-with-the-conclusion shape, or a ban on ordinary contrast and connectives.
- Operational instructions stay on.
Control the work. Let the model speak for itself.