This project provides a self-learning system for Claude Code, enabling it to retain user corrections, feedback, and preferences across sessions. It targets Claude Code users seeking to improve AI consistency and tailor its responses to their specific workflows, offering a mechanism to build a persistent, personalized knowledge base for the AI.

How It Works

The system employs a two-stage process. Stage 1 involves automatic capture via hooks that detect corrections and positive feedback using real-time regex patterns. Stage 2, initiated by the user running the /reflect command, involves manual review. A semantic AI validation layer enhances accuracy, supports multi-language understanding, and extracts concise, actionable learnings before they are applied. This hybrid detection approach aims for higher precision and broader applicability than regex alone, with manual review ensuring quality.

Quick Start & Requirements

• Installation:
  1. marketplace add bayramannakov/claude-reflect
  2. claude plugin install claude-reflect@claude-reflect-marketplace
  3. Restart Claude Code.
• Prerequisites: Claude Code CLI installed, Python 3.6+.
• Platform Support: macOS, Linux, Windows (native Python).
• Documentation: Commands serve as primary usage guides; no separate docs/demo links provided.

Highlighted Details

• Historical Scan: /reflect --scan-history processes all past sessions for pre-installation learnings.
• Smart Filtering: Excludes non-actionable feedback, questions, and context-specific instructions, retaining only reusable learnings.
• Duplicate Detection: /reflect --dedupe identifies and consolidates semantically similar entries in CLAUDE.md to maintain a clean knowledge base.
• Multi-Target Sync: Approved learnings are synced to global (~/.claude/CLAUDE.md), project-specific (./CLAUDE.md), and AGENTS.md files.

Maintenance & Community

The repository welcomes contributions via pull requests, with a mention of contributing guidelines. However, specific community channels (e.g., Discord, Slack), notable contributors, sponsorships, or a public roadmap are not detailed in the provided information.

Licensing & Compatibility

The project is licensed under the MIT license. This permissive license generally allows for commercial use and integration within closed-source projects without significant restrictions.

Limitations & Caveats

The system is intrinsically tied to the Claude Code environment and its plugin architecture. The core processing command (/reflect) requires manual human review, which can introduce a workflow bottleneck. The effectiveness of the learned preferences is directly dependent on the quality and consistency of user interactions and corrections within Claude Code sessions.