multi-review

multi-review

Multi-agent code review orchestration with smart agent selection for Claude Code.

Overview

multi-review is a standalone plugin that orchestrates code review agents from multiple plugins:

• feature-dev (Official Anthropic) - General code review with confidence scoring
• pr-review-toolkit (Official Anthropic) - Specialized PR review agents (7 agents)
• superpowers (✅ MIT Licensed) - Framework-specific review guidance

See DEPENDENCIES.md for complete dependency and licensing information.

Installation

# Install from local marketplace
/plugin marketplace add ~/.claude/plugins/multi-review
/plugin install multi-review@local

Quick Start

Interactive Mode

# Interactive mode (prompts for preset)
/multi-review

# Framework-specific review
/multi-review --agents framework

# Custom agent selection
/multi-review --agents custom

Direct Slash Commands (No Questionnaire)

Skip the preset selection and run immediately:

# Quick check - 2 agents (~30s)
/mr-quick

# Thorough review - 4 agents (~2min)
/mr-thorough

# Comprehensive review - 7 agents (~5min)
/mr-comprehensive

Command Equivalents

Direct Command	Equivalent To
/mr-quick	/multi-review --agents quick
/mr-thorough	/multi-review --agents thorough
/mr-comprehensive	/multi-review --agents comprehensive

Additional Commands

# Multi-agent planning & execution
/mr-plan

# Plugin diagnostics
/mr-doctor

Presets

Preset	Agents	Use Case
quick	2	Fast check before commit
thorough	4	Balanced review
comprehensive	7	Complete review before PR
framework	1-2	Framework-specific compliance

All presets use open-source plugins! See DEPENDENCIES.md for details.

Context Detection

The plugin automatically detects:

• Change size: Lines modified (via git diff)
• File types: Tests, type definitions, error handlers
• PR status: Via gh CLI (optional)

Suggestions based on context:

Context	Suggestion
< 50 lines	quick preset
Has tests	Add test analyzer
Has types	Add type analyzer
> 500 lines	comprehensive preset

Hooks

The plugin includes automated hooks (disabled by default):

Post-Write Hook

Runs quick review after file writes.

# Enable in .claude/settings.local.json
{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Write|Edit",
      "hooks": [{
        "type": "command",
        "command": "bash ${CLAUDE_PLUGIN_ROOT}/hooks/post-write.sh",
        "enabled": true
      }]
    }]
  }
}

Pre-Commit Hook

Runs review before git commit.

# Enable in .claude/settings.local.json
{
  "hooks": {
    "PreCommit": [{
      "hooks": [{
        "type": "command",
        "command": "python3 ${CLAUDE_PLUGIN_ROOT}/scripts/pre_commit_check.py",
        "enabled": true
      }]
    }]
  }
}

Session-End Hook

Runs comprehensive review at session end.

# Enable in .claude/settings.local.json
{
  "hooks": {
    "Stop": [{
      "hooks": [{
        "type": "command",
        "command": "python3 ${CLAUDE_PLUGIN_ROOT}/scripts/session_review.py",
        "enabled": true
      }]
    }]
  }
}

Standalone Scripts

All scripts can be run independently:

# Context detection
python3 ~/.claude/plugins/multi-review/scripts/context_detector.py --suggest

# Post-Write review
python3 ~/.claude/plugins/multi-review/scripts/auto_review.py --file path/to/file.py

# Pre-Commit check
python3 ~/.claude/plugins/multi-review/scripts/pre_commit_check.py --strict

# Session review
python3 ~/.claude/plugins/multi-review/scripts/session_review.py

Reports

Review reports are saved to ~/.claude/plugins/multi-review/reports/:

• review_YYYYMMDD-HHMMSS.json - Post-Write reports
• commit_YYYYMMDD-HHMMSS.json - Pre-Commit reports
• session_YYYYMMDD-HHMMSS.json - Session-End reports

Dependencies

Required plugins:

• feature-dev - for feature-dev:code-reviewer
• pr-review-toolkit - for all PR review agents

Optional:

• superpowers - for superpowers:code-review-checklist

Architecture