Did it stay within your request?
See where the changes match what you asked for, and where they drift.
Illustrative command: perimitr "Add a dark mode toggle. Keep everything else unchanged.". Results: within requested scope, passed. No unrelated features added, passed. Avoidable complexity detected, review required. Two checks passed, one check needs review.
The code works.
But did it follow your brief?
AI can finish a task and still change more than you intended. perimitr is being built to make that difference visible before you move on.
Did it add something you didn’t ask for?
Catch extra features and changes to parts of your project you meant to leave alone.
Could it have been simpler?
Spot added complexity that may need a reason before you keep it.
A second look
before you accept the change.
Start with what you asked your AI assistant to do. Perimitr checks the local changes against that request, so you can decide what stays.
Give it the brief
Provide the original request.
Check the changes
Review your worktree or staged diff against that request.
Make the call
See possible scope drift, unrequested additions, and avoidable complexity. You decide what to keep.
A clearer way to build with AI is on its way.
perimitr is still in development. The check above illustrates the idea we’re exploring: a second look at local changes against the original request.
Join the free beta waitlist