Bounty #4: open-reviewer — Multi-Agent PR Review System

[chenchong0215]

Bounty #4 — PR Review Agent ($150)

🎯 Overview

open-reviewer is a multi-agent PR review system that goes beyond single-script approaches by orchestrating specialized review agents in parallel.

🏗️ Architecture

PR URL → review-coordinator (orchestrator)
            ├→ security-reviewer (OWASP, injection, auth)
            ├→ code-quality-reviewer (testing, maintainability, perf)
            └→ Combined Report with Confidence Score

✨ Key Differentiators

1. Multi-Agent Parallel Review — Security and quality reviews run simultaneously via OpenClaw skill-native agents
2. Structured Confidence Scoring — Formula: base(0.85) - size_penalty - finding_penalties, with clear High/Medium/Low labels
3. Heuristic Analysis Engine — Pre-scan detects 20+ patterns (SQL injection, hardcoded secrets, missing error handling, etc.) without API calls
4. JSON + Markdown Output — Machine-readable output for CI integration
5. GitHub Action Workflow — Automated PR reviews on every pull_request event

📦 Deliverables

File	Purpose
bin/claude-review.py	CLI tool with heuristic analysis engine (16KB)
.claude/agents/review-coordinator.md	Orchestrator agent with confidence scoring
.claude/agents/reviewers/security-reviewer.md	Security analysis (CRITICAL/HIGH/MEDIUM/LOW)
.claude/agents/reviewers/code-quality-reviewer.md	Quality analysis (testing, maintainability)
workflows/pr-review.yml	GitHub Action for automated reviews
tests/test_review.py	14 unit tests (all passing)
samples/sample-requests-7401.md	Real review: psf/requests#7401 (confidence 0.85 High)
samples/sample-uv-19322.md	Real review: astral-sh/uv#19322 (confidence 0.40 Low, 2818 lines)

🧪 Usage

# CLI
python agents/open-reviewer/bin/claude-review.py --pr https://github.com/owner/repo/pull/123

# With JSON output
python agents/open-reviewer/bin/claude-review.py --pr https://github.com/owner/repo/pull/123 --json

# GitHub Action
# See workflows/pr-review.yml — automatically reviews on pull_request events

✅ Requirements Checklist

• CLI tool (claude-review --pr <url>)
• Structured Markdown output (summary, risks, suggestions, confidence score)
• Tested on 2+ real PRs (Add tests for hasattr/getattr usage on LookupDict psf/requests#7401, Sync latest Python releases astral-sh/uv#19322)
• README with architecture and usage documentation
• GitHub Action workflow
• Unit tests (14 passing)

🔬 Tested Against Real PRs

1. Add tests for hasattr/getattr usage on LookupDict psf/requests#7401 — Confidence: 0.85 (High) — Small focused PR, clean code
2. Sync latest Python releases astral-sh/uv#19322 — Confidence: 0.40 (Low) — 2818-line mega-diff, many findings detected