Can we please get a model self-awareness system prompt layer? It’s impossible to develop a reliable plugin as different models have such different capabilities and differerent willingness to complete tasks.
I’m trying to finish a plugin that requires multiple tool calls. Medium and up can work now with the plugin fine, but instant can’t (or at least says it can’t, but if you yell at it two/three times, it then proceeds to successfully do it).
Ideally, I’d prefer to just have Instant know it’s Instant and just prompt the user to switch to a higher reasoning model. But, if I try to add any sort of guidance to allow 5.5 Instant to abort the tool usage and refer the user to use a higher model, then all models start being lazy and following the guidance to say they’re not capable and refer you to use a higher model. It would make developing a plugin 100x easier if we could have one set of prompts for model A and another set for all other models. Or even if the agent could just know what model/reasoning it is.