📑 Table of Contents

• Why ReviewCrew?
• Key Features
• Show, Don't Tell
• Quickstart
• Usage
• Observability Dashboard
• Architecture
• Background & Credibility
• Roadmap
• Contributing
• License

🎯 Why ReviewCrew?

Instead of noisy, generic AI reviews that flag every minor styling nit, agentic-code-review acts like a senior engineer. It uses a multi-agent orchestration model.

Each agent (Security, Bugs, Git History, etc.) runs in parallel with its own model and narrow mandate. Findings are scored 0–100 by confidence, filtered against a shared noise list, deduped, and reported to your terminal or posted as inline GitHub PR comments.

Extended re-implementation of Anthropic's official code-review plugin: featuring the same phase structure and confidence-gating philosophy, plus security & test-coverage reviewers, a local-diff mode, and a rich observability dashboard.

✨ Key Features

• 🚀 Parallel Fan-out: Six specialized agents review your code simultaneously.
• 🛡️ Aggressive Noise Filtering: A shared confidence-scoring rubric ensures you only see high-signal feedback.
• 💻 Local Diff Mode: Review uncommitted changes right in your terminal—no GitHub required.
• 💬 GitHub Native: Pass a PR number and --comment to automatically post deduped, inline review comments.
• 📊 Observability Dashboard: Launch a beautiful local web UI to inspect the exact findings and confidence scores of every agent.

🎬 Show, Don't Tell

Watch the agents successfully review a complex PR, catch bugs standard linters miss, and filter out the noise.

(To be added)

🚀 Quickstart

Get up and running in under a minute via Claude Code.

Prerequisites:

• Claude Code installed (npm install -g @anthropic-ai/claude-code)
• Ensure your environment has the required Anthropic API keys configured.

# 1. Add the marketplace
/plugin marketplace add MANOJ21K/agentic-code-review

# 2. Install the plugin
/plugin install agentic-code-review@reviewcrew

💻 Usage

Run reviews directly from your Claude Code terminal.

# Review your uncommitted local diff
/agentic-code-review:code-review                 

# Review GitHub PR #123 (Terminal report only)
/agentic-code-review:code-review 123             

# Review GitHub PR #123 & post inline PR comments
/agentic-code-review:code-review 123 --comment   

# Strict scan: Only run the security agent with a high confidence threshold
/agentic-code-review:code-review --focus security --threshold 90 

📊 Observability Dashboard

Multi-agent systems shouldn't be a black box. Visualize exactly how the agents performed, what they found, and why the orchestrator filtered certain items out.

After running a code review, launch the dashboard:

/agentic-code-review:dashboard

This spins up a local server and opens a sleek, glassmorphic UI in your browser displaying the full telemetry of the run.

🏗️ Architecture

flowchart TD
    O[Orchestrator] -->|Fans out diff| R1[Bug Reviewer]
    O -->|Fans out diff| R2[Security Reviewer]
    O -->|Fans out diff| R3[Compliance Reviewer]
    O -->|Fans out diff| R4[Git History Reviewer]
    O -->|Fans out diff| R5[Test Coverage Reviewer]
    O -->|Fans out diff| R6[Custom Reviewer]
    
    R1 --> S[Scoring Engine]
    R2 --> S
    R3 --> S
    R4 --> S
    R5 --> S
    R6 --> S
    
    S -->|Confidence > Threshold| F[False-Positive Filter]
    F --> D[Deduplication]
    D --> Rep[Terminal / PR Comment / Telemetry JSON]

🧠 Background & Credibility

This architecture is heavily informed by hands-on experience building complex platforms like CoAgentics. Developing real-world multi-agent systems—including insights gained as a finalist at the Google Cloud Agentic AI Day hackathon—shaped the orchestration, context passing, and confidence-gating logic of this tool. It is built to handle generative AI at scale, tailored for senior developers who demand reliable, low-noise code reviews.

🗺️ Roadmap

• Initial release with 6 specialized agents
• Observability UI and telemetry logging
• Support for custom, user-defined agent injection
• Integration with CI/CD pipelines (GitHub Actions)

🤝 Contributing

We welcome contributions! Please see our CONTRIBUTING.md for details on how to get started.

Look out for issues labeled good first issue or help wanted to jump right in. High-impact areas include adding support for new languages, optimizing agent prompts, and enhancing the observability dashboard.

📜 License

This project is licensed under the MIT License.

📁 Layout

.claude-plugin/marketplace.json       marketplace manifest
code-review-plugin/
├─ .claude-plugin/plugin.json         plugin manifest
├─ commands/                          orchestrator & dashboard launchers
├─ dashboard/                         observability UI source (HTML/CSS/JS)
├─ agents/                            7 reviewer subagents
├─ skills/                            confidence-scoring, review-reporting
└─ README.md                          full documentation