How does Claude Reflect detect corrections?

It monitors natural language patterns such as 'actually', 'no, use X', or 'remember this' to identify when you are providing specific guidance or fixing an error during a chat.

What is the difference between global and project learnings?

Global learnings are saved to ~/.claude/CLAUDE.md and apply to all your projects, while project learnings are saved to the local directory and only apply to that specific codebase.

Can I see what Claude has queued before applying it?

Yes, you can use the /view-queue command to see pending items or /reflect --dry-run to preview exactly what changes would be made to your documentation.

Are changes applied automatically to my files?

No, corrections are first placed in a queue. You must run the /reflect command to review, edit, and approve which learnings are written to your CLAUDE.md files.

Does this work across different Claude sessions?

Yes. By writing these learnings to CLAUDE.md files, the information is loaded into the context of future sessions, allowing Claude to remember your preferences permanently.

Claude Reflect

Name: Claude Reflect
Author: jeremylongshore

byjeremylongshore

•

896

生产力与工作流

Implements a self-learning system that captures user corrections and syncs them to project-specific or global configuration files.

关于

Claude Reflect is a powerful self-improvement tool designed to make Claude smarter over time by capturing real-time corrections and feedback. It automatically queues instructions when you use phrases like 'actually, use X instead of Y' and allows you to review and sync these learnings into your CLAUDE.md files. This ensures that the AI persists project-specific conventions, model preferences, and unique developer workflows across different sessions, significantly reducing repetitive instructions and personalizing your development environment.

主要功能

Automatic capture of conversational corrections and explicit 'remember this' triggers
History scanning functionality to recover missed learning opportunities from past sessions
896 GitHub stars
Interactive review process for human-in-the-loop validation of queued learnings
Comprehensive queue management with dry-run, skip, and view capabilities
Dual-scope support for both global and project-specific CLAUDE.md updates

使用场景

Correcting recurring AI mistakes regarding project structure or specialized tool preferences
Persisting specific library versions or coding standards across different coding sessions
Building a personalized AI knowledge base for complex codebases without manual documentation

关于

主要功能

Automatic capture of conversational corrections and explicit 'remember this' triggers
History scanning functionality to recover missed learning opportunities from past sessions
896 GitHub stars
Interactive review process for human-in-the-loop validation of queued learnings
Comprehensive queue management with dry-run, skip, and view capabilities
Dual-scope support for both global and project-specific CLAUDE.md updates

使用场景

Correcting recurring AI mistakes regarding project structure or specialized tool preferences
Persisting specific library versions or coding standards across different coding sessions
Building a personalized AI knowledge base for complex codebases without manual documentation