bmad-checkpoint-preview

A structured code review for a checkpoint organizes feedback into orientation, a suggested review trail, concern-based walkthroughs, and risk spotting. This approach helps human reviewers focus on design intent, boundary issues, and high-blast-radius spots before merging.

A human-in-the-loop code review workflow guides a reviewer through automated orientation and risk spotting while requiring human judgment for final decisions. It provides a suggested review trail and testing suggestions so humans focus on evaluating design intent and high-risk areas.

To identify high-risk areas in a diff, perform a detail pass that surfaces two to five high-blast-radius spots for targeted attention. This risk analysis focuses your review on boundary issues and critical changes rather than just listing standard code violations.

Spec-driven changes can guide your code walkthrough by providing optional spec artifacts to set context and intent. The review process consumes these specs to summarize intent, compute surface-area stats, and organize findings by concern during the walkthrough.

You need repository access to analyze a branch or commit diff, including the ability to resolve path:line references and read files. This access allows the review process to locate the change, compute surface-area stats, and generate accurate walkthroughs.

The best way to validate behavior after a checkpoint review is to follow actionable testing suggestions that provide observable checks. These testing steps wrap up the review workflow and give you a clear decision prompt to approve, rework, or discuss the changes.