Skip to main content

Dual-Review Skill

The dual-review skill is a Claude Code slash command that orchestrates a complete review-driven development workflow using the Claude MCP bridge.

Overview

The skill automates this workflow:

  1. Explore the codebase to understand the area being changed
  2. Draft an implementation plan
  3. Submit the plan to the reviewer MCP for structured review
  4. Iterate on critical/major issues (up to 3 rounds)
  5. Implement after receiving APPROVED verdict
  6. Post-review the final diff
  7. Summarize with full review history

Setup

The skill file is located at skills/dual-review/SKILL.md. To use it:

  1. Ensure the Claude MCP bridge is running (default: http://localhost:8940/mcp)
  2. Configure claude_reviewer in your .mcp.json:
{
"mcpServers": {
"claude_reviewer": {
"type": "streamable-http",
"url": "http://localhost:8940/mcp"
}
}
}
  1. Use the skill in Claude Code:
/dual-review Implement feature X

How It Works

Phase 1: Exploration

The skill reads relevant files to understand the codebase structure and existing patterns.

Phase 2: Planning

A detailed implementation plan is drafted based on the exploration.

Phase 3: Plan Review

The plan is sent to the review tool on the claude_reviewer MCP server. The reviewer returns a structured verdict.

Phase 4: Iteration

If the verdict is NEEDS_REVISION, the skill addresses critical and major issues and resubmits. This continues for up to 3 rounds.

Phase 5: Implementation

Once the plan is APPROVED, the skill implements the changes.

Phase 6: Post-Implementation Review

The final git diff is sent to the code_review tool for a post-implementation check.

Phase 7: Summary

A complete summary is presented with all review rounds and the final verdict.

Configuration

The skill inherits configuration from the Claude MCP bridge environment variables:

  • CLAUDE_REVIEW_MODEL -- Model used by the reviewer
  • CLAUDE_REVIEW_MAX_TURNS -- Max agentic turns per review
  • CLAUDE_REVIEW_CWD -- Working directory for file access