Not impossible, but also the thinking process reduces the need for something like that. Anything in the context gets regurgitated back and forth multiple times so the model has multiple opportunities to flag something as being weird.

Incidentally if you actually want to do something like that yourself, I believe OpenRouter has a feature for that. But of course it means you can essentially add up each model’s costs and that’s your new cost per prompt (not exact since output token count will differ, but on average).