The idea is memory alone does not prevent claude code to make the same mistake twice. It should be able to capture user corrections and turn into learnings that are generalized across different use cases and projects
We have done benchmark against other memory based solutions and found that memory based solution almost never capture any user correction but just storing user mentioned facts. In addition, our context footprint is a fraction of what claude-mem has, which make claude-smart super context efficient.
Try install it with ``` npx claude-smart install ```
and let us know what you think, your feedback is hugely appreciated!