vibe-reviewer-dev

Vibe Code Reviewer

Multi-agent code review system for Claude Code. 7 specialized auditors detect 41 cataloged antipatterns across your codebase and generate prioritized reports.

What It Does

Vibe Reviewer audits your entire codebase (not just diffs) looking for structural problems: missing error handling, god classes, hardcoded secrets, unbounded LLM calls, missing health checks, and 35 more antipatterns across 7 domains.

Each review produces a markdown report saved to reports/ with findings sorted by severity, evidence, and actionable recommendations.

Quick Start

# Review current directory
/vibe-review

# Review a specific project
/vibe-review ~/Developer/my-project

# Run only specific auditors
/vibe-review --skills=schema-auditor,resilience-auditor

# Show only critical findings
/vibe-review --severity=critical

How It Works

/vibe-review
    │
    ▼
┌─────────────────────────┐
│  1. Detect Project Stack │  ← Languages, frameworks, ORMs, DB, async patterns
│     (stack-detection.md) │
└────────────┬────────────┘
             │
             ▼
┌─────────────────────────┐
│  2. Select Auditors      │  ← Always: architecture, hygiene, production-readiness
│                          │  ← Conditional: schema, async, cost, resilience
└────────────┬────────────┘
             │
             ▼
┌─────────────────────────┐
│  3. Launch Skills        │  ← Each auditor runs as a Task subagent
│     (parallel via Task)  │  ← Returns JSON array of findings
└────────────┬────────────┘
             │
             ▼
┌─────────────────────────┐
│  4. Validate & Aggregate │  ← Validate against finding-schema.json
│                          │  ← Reject invalid findings
│                          │  ← Deduplicate by file+line
│                          │  ← Sort by severity → confidence
└────────────┬────────────┘
             │
             ▼
┌─────────────────────────┐
│  5. Generate Report      │  ← Mandatory structure (report-template.md)
│     → reports/           │  ← Saved as vibe-review-YYYYMMDD-HHMMSS.md
└─────────────────────────┘

The 7 Auditors

Auditor	Domain	Antipatterns	Runs When
schema-auditor	Database & ORM	god-table, missing-indexes, data-duplication, enum-table, missing-constraints, n+1-schema	DB/ORM detected
architecture-auditor	Structure & layers	spaghetti-code, god-class, circular-dependency, layer-violation, business-logic-in-controllers, infrastructure-leakage	Always
resilience-auditor	Error handling	silent-failure, generic-exception-catching, missing-timeout, no-retry-logic, missing-circuit-breaker, unvalidated-external-input	HTTP/external I/O detected
async-auditor	Concurrency	fire-and-forget, mixed-sync-async, race-condition, thread-pool-exhaustion, callback-hell	Async patterns detected
cost-auditor	LLM & API costs	unbounded-llm-calls, redundant-api-calls, missing-caching, oversized-prompts, no-pagination	LLM/API calls detected
hygiene-auditor	Code quality	code-duplication, magic-numbers, dead-code, inconsistent-naming, missing-docstrings, overly-complex-function	Always
production-readiness-auditor	Ops & deploy	missing-health-checks, no-metrics-monitoring, hardcoded-configuration, missing-logging, no-rate-limiting, missing-tests, no-backup-strategy	Always

Finding Schema

Every finding is validated against finding-schema.json. Required fields:

{
  "finding_id": "schema-001",
  "skill": "schema-auditor",
  "antipattern": "god-table",
  "severity": "critical",
  "confidence": 0.95,
  "location": {
    "file": "src/models/user.py",
    "line": 15,
    "function": "User"
  },
  "evidence": "User model has 47 fields spanning auth, preferences, billing, and profile",
  "recommendation": "Split into UserAuth, UserPreferences, UserBilling, UserProfile with 1:1 relationships",
  "schema_version": "1.1.0",
  "catalog_version": "1.1.0"
}

The antipattern field is a closed enum — only the 41 cataloged names are valid. Auditors cannot invent new antipattern names.

Severity Levels

Level	Meaning	Action	Example
🔴 Critical	Security, data loss, production breakage	Block deployment	Hardcoded API key, silent exception swallowing
🟡 Important	Logic errors, maintainability risks, perf issues	Fix in next sprint	N+1 queries, missing retry logic, god class
🟢 Desirable	Style, optimization, documentation	Tech debt backlog	Magic numbers, missing docstrings

Report Output

Reports are saved to ~/.claude/plugins/vibe-reviewer/reports/ with this structure: