Your AI coding agent just modified 15 files, added 800 lines of code, and proudly announced:
"All done. Everything works."
Great.
But do you actually understand what it changed?
Meet diff-tour, a free, open-source tool that walks you through code changes the way an experienced software engineer would review them.
It works with Claude Code, OpenAI Codex, and other AI coding assistants.
And instead of dumping a giant Git diff in front of you, it creates an interactive, step-by-step explanation.
Here's what diff-tour does:
Explains the big picture: Why the changes were made and how they fit into your existing codebase.
Shows what matters: Highlights the files you should examine, including related files that weren't modified.
Asks the right questions: What happens if something fails? Can this safely run twice? Will existing functionality still work?
Identifies real problems: Separates must-fix issues from optional improvements and explains why suspicious-looking code might actually be safe.
Teaches you to think like a senior engineer: Sometimes it asks you to predict what could go wrong before revealing the explanation.
Uses actual Git data: Every displayed line of repository code comes directly from Git, not from something the AI invented.
And here's the interesting part.
The entire review runs in read-only mode.
Your AI agent can inspect and explain the code, but it cannot modify your repository during the tour.
The tool uses your existing Claude Code or Codex account, and generating a review typically takes around 2–5 minutes, according to the developers.
How to install diff-tour on macOS or Linux:
curl -fsSL https://raw.githubusercontent.com/init-co-jp/diff-tour/main/install.sh | sh
Then navigate to your Git project and run:
tour
Choose the changes you want to review and let the walkthrough begin.
You can navigate between explanations, explore code, answer questions, and investigate potential problems before merging anything.
My suggestion: add this to your AI coding workflow.
Let Claude Code or Codex implement a feature.
Run tour to understand the changes.
Identify the issues that actually need fixing.
Give those findings to Claude Opus 5.5.
Ask it to explain its proposed fixes before touching any code.
Approve the changes, run the tests, and review the final diff.
Here's a prompt you can use with Opus 5.5 after the review:
"diff-tour reviewed my latest code changes and identified these must-fix issues:
[PASTE FINDINGS HERE]
For each issue, explain how you plan to fix it and which files you'll modify. Wait for my approval before making any changes.
Once approved, implement the fixes, run the relevant tests, and report which issues are resolved and which remain."
Of course, an AI-generated review doesn't replace proper testing or human judgment.
But it gives you something increasingly valuable in the age of agentic coding.
Understanding.
Because generating code is getting ridiculously easy.
Understanding, maintaining, and safely shipping that code is still your responsibility.
And frankly, having a patient senior-engineer-style walkthrough every time your AI agent finishes a task sounds pretty useful.
diff-tour on GitHub:
https://github.com/init-co-jp/diff-tour
Free, open source, MIT licensed.
#ClaudeCode #OpenAICodex #DiffTour #CodeReview #AICoding #VibeCoding #Shorts
YouTube / ABVX