Persistent Knowledge Retention Challenge with API Updates and Context Limitations

Bug Description
I'm running into real challenges working with claude when i need it to remember and work with updated APIs (in this case OpenAI) where claude's knowledge cutoff causes it to make incorrect assumptions about the model names and how they work, even with me giving it updated documentation - which it sometimes handles fine, but it uses up so much context, that i eventually have to /compact, and then once i need it to start troubleshooting the code, it starts rewriting things based on out of date information again. And reloading the needed context over and over means my context is always running low and my cost of every request is really high. I have spent probably $30 already today and not made much headway since alot of the progress keeps getting undone. I have tried encouraging claude to document what we need, but that just results in more and more documents that even i can't keep track of now, and claude doesn't seem to remember to check when needed. I like what you all are doing, but without much larger context lengths, much cheaper cost to operate, or some way to provide different way to work around these issues, it is easy to get trapped in an expensive and time consuming loop that doesn't yield meaningful results.

Environment Info

• Platform: macos
• Terminal: Apple_Terminal
• Version: 0.2.14
• Feedback ID: ec629314-b599-406f-98fa-c66a74d775f5