Hey HN, I built this because I was frustrated re-explaining my setup every Claude Code session and hitting context limits faster than needed. The plugin uses mem0.ai to:
- Automatically retrieve relevant memories before each prompt (UserPromptSubmit hook)
- Store conversation context when sessions end (Stop hook)
- Semantic vector search so only relevant context is injected
Example: Claude now knows I prefer Python, my tech stack, and project context - without me repeating it.
Technical details:
- Pure Python, no dependencies except mem0ai
- ~120 lines of code per hook
- Works with Claude Code's official plugin system
- MIT licensed
Just submitted a PR to the official Anthropic plugin registry: https://github.com/anthropics/claude-plugins-official/pull/52
Happy to answer questions about the implementation or Claude Code plugins in general.