Raw human-in-the-loop review is overrated. AI has already proven useful enough that I started wondering: why are we still reviewing AI-generated code manually when AI itself can help review it?
When using Claude Code or Codex, I’d first rely on specialized subagents for deeper reviews. But for a quick overview of what actually changed, opening a basic Git diff viewer wasn’t enough for me.
So I built Magic Diff, a tool that uses AI to summarize changes across one or more committed or uncommitted files.
It’s open source, written in Go/Wails/React.
Repository: https://github.com/gsscoder/magik-diff
source & further reading
dev.to — original article
Stop Being Your Agent's QA: A Skill That Makes AI Coding Agents Prove Their Work
If your file-guard hook is registered on Read, it never sees cat
The Glass Box: Inside OpenAI's Computer History & Agentic AI