I built a prompt compressor that cuts token usage and help prevents Prompt Decay in long sessions

If you use Claude (or any LLM) for long sessions you've probably hit this at some point. You're deep...