Last week, I kicked off a new series where I revisit Git training material I wrote and taught ten years ago and look at it through the lens of AI agents.
How does it hold up?
In a new video today, I cover revert vs reset. Ten years ago it was a style choice but if you have AI agents working in Git, you need to set careful permissions boundaries.
In the video I show the implications of setting soft boundaries via a CLAUDE.md (or AGENTS.md) file and via a Claude Code hook that strictly enforces what your agent can do.
If you've been around a while, you know that I care a lot about creating guardrails for agents. No YOLO, no one-shotting, no let 'er rip.
Included with the video is the complete CLAUDE.md file and the git-guardrails bash script that you can hook into Claude Code.
