Claude-smart, an open-source plugin for Claude Code, stores past command outcomes to extract reusable lessons and apply them to new tasks while cutting token usage over 70 percent

This is kind of interesting: it just not memory, but "smart memory"?

Assuming this works, this could make Claude Code much more effective.

The idea is to "learn" from past experiences, not just "remember". So far, every memory layer is about storing and fetching information, but we haven't seen much about the learning part.

Now, please release this for Codex.