Skip to main content

How to Stop Your AI Agent From Making Unwanted Code Changes

· 3 min read
Rizel Scarlett
Rizel Scarlett
Staff Developer Advocate

goose, revert this change!

AI agents are often described as brilliant, overeager interns. They're desperate to help, but sometimes that enthusiasm leads to changes you never asked for. This is by design: the large language models powering agents are trained to be helpful. But in code, unchecked helpfulness can create chaos. Even with clear instructions and a meticulous plan, you might hear, "Let me just change this too…" A modification that's either unnecessary or, worse, never surfaced for review.

Sure, you can scour git diff to find and revert issues. But in a multi-step process touching dozens of files, untangling one small, unwanted change becomes a manual nightmare. I've spent hours combing through 70 files to undo a single "helpful" adjustment. Asking the agent to revert is often futile, as conversational memory isn't a snapshot of your codebase.

This problem has a classic engineering solution. We commit early and often to create checkpoints, enabling easy rollbacks and clean collaboration. So, why don't we enforce the same discipline on our AI agents? Here’s the workflow I use with goose to ensure we're creating snapshots of the codebase:

1. Set Up Version Control​

I set up the GitHub CLI (gh). I've found Goose interacts with it flawlessly. The GitHub MCP Server is a good alternative.

2. Branch First​

Always start on a new feature branch. Never let an agent commit directly to main.

3. Set Rules in a Context File​

This is the key. I use a .goosehints or AGENTS.md file with one critical instruction:

"Every time you make a change, make a commit with a clear message."

This does two things: it automates checkpointing so I don't have to babysit the session, and it captures perfect snapshots in time, turning the git history into an undo stack for the entire collaboration.

4. Collaborate with Confidence​

Now I can prompt goose to build, fix, or refactor. If it veers off course or makes a design choice I dislike, I can instantly review the git log or simply say:

"Revert to commit abc123."

The Result​

Then, when I'm happy with the final changes, I can push my code to remote. By integrating this basic software practice, I replace anxiety with awareness. goose gets to be brilliantly helpful, and I get to stay in control.

Try out this method with goose to help you build your next project. Your future self (and your git history) will thank you.