bmad-code-review

Run an adversarial code review by collecting your Git diff, optionally loading a spec, and executing parallel review layers that triage findings into decision-needed, patch, defer, or dismiss buckets.

Spec-aware acceptance auditing checks your code changes against acceptance criteria and related context docs from a story file, persisting the review findings directly back into the relevant story file.

Triage code review findings by normalizing, deduplicating, and classifying the results from parallel adversarial review layers into structured categories: decision-needed, patch, defer, and dismiss.

Yes, you can review staged or uncommitted working tree changes, as the workflow applies to PR, commit, and branch diffs, halting for human confirmation at guided execution checkpoints when targets are unclear.

Automated spec validation requires a workflow resolver step for customization and halts for human confirmation when decisions are needed, meaning it cannot fully auto-apply patches without guided execution checkpoints.

Yes, bmad-code-review uses parallel subagent orchestration to run Blind Hunter, Edge Case Hunter, and Acceptance Auditor layers simultaneously to surface different failure modes.