# Git for Vibe Coding and Agentic Engineering

> Source: <https://pub.towardsai.net/git-for-vibe-coding-and-agentic-engineering-43eff72fc0d7?source=rss----98111c9905da---4>
> Published: 2026-07-21 15:01:04+00:00

Member-only story

**Git for Vibe Coding and Agentic Engineering**

## A plain language guide to git, written for people who spend more time reviewing AI agents than writing code by hand.

At some point, Claude Code, Codex, or Antigravity is going to stop mid-task. It will ask permission to run something like:

```
git add src/routes/checkout.ts; git commit -m "fix: handle null cart on checkout" 2>&1; git push 2>&1
```

That’s three commands chained together.

They might seem overwhelming to the untrained eye, but none of them look especially unusual on their own.

You will likely approve it.

That is usually the right call. Honestly, denying it without knowing why will not be helpful either.

Most of the time, nothing bad happens.

Then one day the agent asks about a different command.

It looks just as harmless as the last ten.

But approving it would be detrimental to your project, or even your entire system. For instance:

- A Cursor user
[reported](https://forum.cursor.com/t/cursor-ai-executes-destructive-command-rm-rf-during-development-session/129401?utm_source=chatgpt.com)that the coding agent attempted to run a recursive deletion command`rm -rf ~ && ls -la`

that causes data loss outside the project workspace.
