bmad-code-review

Run adversarial code reviews on diffs and triage findings into intent_gap, bad_spec, patch, or defer.

Updated Mar 25, 2026
One-click install
npx skills add https://github.com/JingyiWng/databricks_ai_dev_kit_price_watcher --skill bmad-code-review-jingyiwng
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: bmad-code-review
Source: https://github.com/JingyiWng/databricks_ai_dev_kit_price_watcher/tree/main/_bmad/bmm/workflows/4-implementation/bmad-code-review
Command: npx skills add https://github.com/JingyiWng/databricks_ai_dev_kit_price_watcher --skill bmad-code-review-jingyiwng

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill addresses the challenge of reliably surfacing functional bugs, edge cases, and specification deviations in code changes by running structured, adversarial reviews across multiple independent review layers and producing a concise, actionable triage.

Core Features & Use Cases

  • Parallel adversarial review layers: Runs Blind Hunter (diff-only), Edge Case Hunter (diff + repo), and Acceptance Auditor (diff + spec) to capture different failure modes.
  • Robust input handling: Accepts staged/uncommitted changes, branch diffs, commit ranges, and user-provided unified diffs with validation and chunking for large changes.
  • Structured triage and presentation: Normalizes and deduplicates findings, classifies into intent_gap, bad_spec, patch, or defer, and presents grouped actionable results for engineering and product decisions.

Quick Start

Ask the assistant to run code review on the current diff and include the spec if available.

Frequently Asked Questions about bmad-code-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an adversarial code review on my current git diff?

To run an adversarial code review, apply the Skill to staged, uncommitted, or branch diffs. It executes parallel review layers like Blind Hunter and Edge Case Hunter to surface bugs and structured triage results.

What is an adversarial multi-layer code review?

Adversarial multi-layer code review runs independent parallel assessments, such as Blind Hunter for diff-only bugs and Acceptance Auditor for spec deviations. This approach surfaces diverse functional failures and edge cases missed by single-pass reviews.

Can I review a specific commit range or a manually provided unified diff?

Yes, you can review commit ranges and user-provided unified diffs. The Skill robustly validates and chunks large code changes, normalizing findings across all input formats for consistent structured triage.

How does spec loading work during a code review audit?

Spec loading works by optionally passing specification documents to the Acceptance Auditor review layer. This allows the review to classify code deviations into intent gaps or bad specs rather than just functional patch errors.

How are code review findings categorized and triaged?

Code review findings are normalized, deduplicated, and triaged into four structured categories: intent_gap, bad_spec, patch, or defer. This grouped presentation provides actionable results for engineering and product decisions.

Does this code review approach work without a provided specification?

Yes, the review works without a specification by utilizing the Blind Hunter and Edge Case Hunter layers. These layers analyze diffs and repository context to identify functional bugs and edge cases independently of formal spec loading.