This is a lightweight memory system that lets Claude store learnings in PostgreSQL with embeddings for later retrieval. You invoke it with `/remember` followed by what you learned, and it auto-categorizes the entry into types like WORKING_SOLUTION, ERROR_FIX, or FAILED_APPROACH based on keywords. The real value is accumulating context across sessions so you're not solving the same problem twice. It extracts tags, assigns confidence levels, and returns an ID for each entry. The auto-detection is simple pattern matching on words like "error" or "didn't work", so explicit types are better when you want precision. Most useful in long-running projects where institutional knowledge tends to evaporate between debugging sessions.
npx skills add https://github.com/parcadei/continuous-claude-v3 --skill remember