// the find
poteto/brainmaxxing
stupid simple persistent memory and skill improvement
A persistent memory layer for Claude Code built entirely on markdown files and shell hooks. The agent reads a vault of principles and notes at session start, and slash commands (/reflect, /meditate, /ruminate) write back to it as you work. The vault is simultaneously a valid Obsidian notebook, so you can inspect and edit what the agent has learned without any special tooling.
- Plain markdown storage means zero vendor lock-in — the memory survives Claude Code updates, works with grep, and doesn't rot if the project is abandoned
- The /ruminate command mines your existing Claude Code conversation history for corrections and preferences you gave but never explicitly captured — useful bootstrap for people with months of prior sessions
- Obsidian compatibility is load-bearing here, not decorative — once the brain accumulates real content, having a proper search and graph view to audit it matters
- The 16 starter principles are opinionated enough to be immediately useful; you're not starting from a blank slate on day one
- Installation asks you to tell Claude to fetch and install from a GitHub URL — that's a prompt injection surface and there's no package manager, lockfile, or integrity check anywhere
- The hook scripts are bash-only; Windows users hit a wall unless they're already in WSL
- Brain files live in your git repo as plain markdown with no merge strategy — once the agent is actively writing, you'll get conflicts across branches or machines with no resolution path
- With 210 stars and 4 forks and a single author, the 16 starter principles are one developer's opinions baked in as defaults; if your working style differs, you're pruning before you've started