עד כמה משפיע 'Prompt Bloat' על ביצועי Claude?

# מהמקור

Hi guys I have been going back through some of the prompts we use with Claude and realized a few of them have gotten way longer than I remembered and it wasn’t really intentional either like usually Claude does something we don’t want, we add an instruction to prevent it next time then a few weeks later there’s another edge case and another instruction gets added and eventually you end up with a massive prompt where we aren't sure which parts are doing anything useful. They still work well so I’m hesitant to start deleting things just for the sake of making them shorter but I’m also wondering how much unnecessary context we’re sending over and over again (especially for stuff that gets run pretty frequently). Have any of you guys