bmad-code-review

Adversarial code review for branch diffs is automated by running parallel Blind Hunter, Edge Case Hunter, and Acceptance Auditor reviews. The system merges and deduplicates findings, then classifies them into decision_needed, patch, defer, or dismiss buckets for actionable triage.

Adversarial multi-layer code review is a process that runs parallel diff-only, project-aware, and spec-aware assessments to maximize coverage. It surfaces blind spots, edge cases, and acceptance-criteria violations, producing structured triage reports instead of noisy, unstructured feedback.

Code review findings are triaged by merging and deduplicating results from parallel review passes, then classifying each issue into decision_needed, patch, defer, or dismiss buckets. This structured normalization ensures teams act on precise findings with clear handling rules.

Yes, code review can be applied to staged changes, uncommitted changes, branch diffs, specific commit ranges, or user-provided diffs. The system builds the diff from the specified source and optionally halts for confirmation on review scope and chunking large diffs.

Yes, the code review workflow integrates with development artifacts by writing findings to story files, appending deferred work items, and syncing sprint-status updates when available. This connects triaged review findings directly to ongoing development workflows.

Automated code review triage requires human-in-the-loop checkpoints for confirmations on review scope, spec presence, large diff chunking, and decision-needed resolutions. Users must manually resolve decision-needed findings and approve patch applications before the workflow proceeds.