claude-reflect

Table of content

what it does

a self-learning system for Claude Code that captures your corrections, positive feedback, and preferences — then syncs them to CLAUDE.md and AGENTS.md.

your agent doesn’t forget anymore. every time you fix its mistakes, it writes that knowledge into markdown files in your workspace.

why it matters

this is the “your life is a repo” thesis in its purest form.

your agent isn’t a chatbot that forgets you after every session. it’s a system that accumulates knowledge about you as git-trackable text.

your corrections become commits. your preferences become documentation.

the agent becomes a reflection of you. not because someone programmed it that way, but because it learned from working with you.

how it works

  1. you correct Claude Code (e.g., “don’t use semicolons in JavaScript”)
  2. claude-reflect captures that correction
  3. it updates CLAUDE.md or AGENTS.md with the new rule
  4. next time, Claude Code remembers your preference

it’s like having a pair programmer who actually learns from you — and writes down what they learned.

the self.md take

this is what personal AI should look like.

not a generic assistant that treats every user the same. but a system that adapts to you — and stores that adaptation as human-readable, version-controlled text.

you can fork it. you can diff it. you can see exactly what the agent learned and when.

your life is a repo. your agent is learning. what are you teaching it?

repo: https://github.com/BayramAnnakov/claude-reflect
stars: 672
spotted: 2026-02-13