See what changed and why.
Trace changes across AI codebases and give reviewers the context behind the code.
Trace a change in a document parser
In this sample, a reviewer investigating an AI-assisted parser change needs to see why the code changed, where it changed, and which check covers the new behavior.
Intent
Handle invoices that omit a date so they can be flagged for review.
Code change
A change in
parser/date.pyreturns an empty date and a review flag when the source omits the field.Verification
A linked test fixture checks that a missing date triggers a review flag and leaves the date empty.
Where unvibe fits
Code review
Follow the change from its stated intent to the affected files and related checks.
Regression investigation
Find the changes and decisions associated with a behavior that shifted.
Team handoffs
Give maintainers the context to understand why earlier decisions were made.
Planning a workflow like this?
During review, capture the reason for each behavior change and link the check that verifies it. If there is no related check yet, record that gap so the next reviewer knows what still needs verification.
How to use unvibe
Record the intent
Write down the behavior being changed, why it matters, and the decision reviewers need to verify.
Connect the evidence
Link the decision to the affected file or change and the check that covers it. Record gaps when verification is still pending.
Use the trace in review
Follow the evidence from the decision to the code and test. Revisit it during a regression investigation or handoff.
What to measure
Reviewers can spend less time reconstructing context from scattered notes and discussions. Track time spent finding the reason for a change and the share of changes linked to a verification check.
Estimate monthly savings
Adjust this planning example for your team. Include human review time in your estimates. The calculator values time freed at your hourly labor cost and excludes software and implementation costs. The defaults are assumptions for planning.
Bring your next code review
Before your next release, pick one AI-assisted change that a second engineer should be able to explain. We can walk through the workflow and identify what to measure in a pilot.