BayramAnnakov/claude-reflect

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

What it solves

Claude Code often forgets corrections or repeating patterns across sessions. claude-reflect creates a permanent memory system that captures user corrections and discovers recurring workflow patterns, turning them into reusable rules and skills so the AI doesn't make the same mistake twice.

How it works

The system operates in two primary modes: a Python-based plugin and a TypeScript-based mod. It uses a hybrid detection approach combining fast regex patterns (to catch phrases like "no, use X") and semantic AI validation (to filter out noise and support multiple languages).

  1. Capture: Hooks automatically detect corrections during a session and queue them.
  2. Process: The user runs /reflect to review the queue and approve the addition of these rules to CLAUDE.md (global or project-specific) or AGENTS.md.
  3. Skill Discovery: The /reflect-skills command analyzes session history to find repeating tasks (e.g., asking for a productivity review multiple times) and suggests creating a dedicated skill file in .claude/commands/.
  4. Routing: Corrections made while using a specific skill are routed back to that skill's definition file to improve it over time.

Who it’s for

Developers using the Claude Code CLI who want to automate the process of documenting project conventions and personal preferences, as well as those using Codex, Cursor, or Aider via AGENTS.md support.

Highlights

  • Permanent Memory: Syncs corrections to CLAUDE.md for persistence across sessions.
  • AI-Powered Skill Discovery: Analyzes history to suggest new /commands based on repeated intent.
  • Multi-language Support: Understands corrections in any language via semantic analysis.
  • Historical Scanning: Can scan all past sessions to recover missed learnings.
  • Semantic Deduplication: Consolidates similar rules to keep memory files clean.

Related

  • Project
    EveryInc/compound-knowledge-plugin

    A Claude Code plugin that implements a structured knowledge-work loop to capture, store, and reuse project learnings in a persistent knowledge base.

  • Project
    supermemoryai/claude-supermemory

    claude-supermemory is a Claude Code plugin that gives the LLM coding agent persistent memory across sessions. It automatically captures conversation turns, recalls relevant past context at prompt time, and shares that memory across tools like Codex and OpenCode via a unified container tag.

  • Project
    anthropics/skills

    A collection of implementation examples and specifications for Claude Skills, allowing users to dynamically load instructions and resources to perform specialized, repeatable tasks.

  • Project
    ReflexioAI/claude-smart

    A self-improvement plugin for Claude Code, Codex, and OpenCode that turns user corrections and successful interactions into durable, reusable skills to prevent AI mistakes.

  • Project
    tigerless-labs/autoharness

    AutoHarness is a Claude Code plugin that automatically extracts, merges, and prunes reusable “skills” from your coding sessions, storing them as plain files and injecting an index at session start. It learns on‑the‑fly, manages skill lifecycles, and provides configurable environment variables for cadence, capacity, and notifications.