dankefox

dankefox / claude-code-reflection-hook

Public

给 Claude Code 装一个反思系统 — 让 AI 记住自己踩过的坑，越用越聪明

15
6
69% credibility
Found May 28, 2026 at 15 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

This project adds a 'reflection system' to Claude Code, an AI coding assistant. After each time Claude edits code, the tool automatically checks for syntax errors and compares the changes against a database of past mistakes. When it finds a match, it warns itself before repeating the error. The system also learns from repeated corrections—mistakes you catch twice get remembered automatically, while lessons you don't need anymore gradually fade away. It's designed to make Claude Code improve over time by remembering what went wrong before, rather than requiring you to correct the same mistake repeatedly.

How It Works

1
😤 The same mistake, again

You notice Claude Code keeps making the same error it made last week—wrong property names, same typos. You have to correct it every single time.

2
🔍 Finding the fix

You discover a tool that helps Claude Code remember its mistakes so it won't repeat them.

3
⚙️ Setting things up

You add a small hook to your Claude Code settings and create a simple list of mistakes you've caught before.

4
✨ Claude starts learning

Now every time Claude edits code, it automatically checks against your mistake list and warns itself before repeating an error.

5
Two paths of learning
📋
Automatic checks

Python syntax, JSON format, and shell commands are checked automatically after every edit

✏️
Your own rules

You write custom checks for mistakes specific to your project—like checking that UI component names are correct

6
🧠 Getting smarter over time

When you catch a mistake twice, the system automatically adds it to the permanent memory. Mistakes you don't make anymore slowly fade away.

🎉 Fewer repeated mistakes

Claude Code stops making the same mistakes it made before. The more you use it, the smarter it becomes about your project's quirks.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 15 to 15 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is claude-code-reflection-hook?

This is a hook system for Claude Code that builds a persistent memory of past mistakes. Every time Claude Code edits a file, the hook runs syntax checks and matches the changes against a lessons database. If it spots a pattern that caused problems before, it warns you before the mistake propagates. The system is written in Python and integrates through Claude Code's PostToolUse hook mechanism.

Why is it gaining traction?

The killer feature is automatic learning. You do not manually maintain the lessons database forever. The system has a candidate pool that promotes lessons automatically after the same mistake happens twice. Over time, it decays lessons you no longer trigger and merges overlapping rules. It also includes a self-testing regressor that runs automatically when you modify the hook itself, preventing silent breakage. The architecture is straightforward: syntax validation, pattern matching against past errors, and an evolution layer that sharpens the database without manual curation.

Who should use this?

Developers working on long-running Claude Code projects where the same errors keep surfacing will benefit most. If you find yourself repeatedly correcting the same API calls, property names, or import statements, this system captures that institutional knowledge. Teams with project-specific conventions that Claude Code keeps violating will get the most value. It is less useful for one-off tasks or projects where you rarely repeat mistakes.

Verdict

This is an interesting concept with a clever auto-evolution mechanism, but the project is extremely early stage with only 15 stars and a credibility score of 0.699%. The documentation is solid for a hobby project, and the self-testing regressor shows thoughtful engineering, but there is no community backing or production track record yet. Try it on a personal project to see if the workflow fits your needs, but do not bet critical work on it without watching the repo for sustained activity.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.