Replay Learnings

It greps through files like .claude/LEARNED.md and your session history, then ranks findings by relevance rather than recency. You trigger it by saying "what do I know about" or "remind me about" followed by your task. The output shows concrete learnings with context (like "always mock external APIs in auth tests, applied 8x") and flags previous sessions with high correction rates as warning signs. It's honest about when it finds nothing instead of forcing irrelevant results. Think of it as checking your notes before repeating the same debugging session twice.