Open Review

A battle-tested methodology for AI-assisted code review, delivered as a harness, skills, and presets.

Open Review is not just a code review tool. It provides a structured reasoning process that guides AI agents to produce consistent, high-quality reviews. The quality depends on the model, context, instructions, and codebase — we provide the structure.

Three Modes of Operation

| Mode | Context | What We Deliver | |---|---|---| | CI/CD | No agent present | Harness runs automated reviews via GitHub Action | | Local CLI | Developer wants quick check | Harness script for terminal use | | Existing Agent | Developer uses Cursor/Claude Code/Open Code | Methodology as agent instructions (skill) |

Quick Start

Zero Config

npm install -g open-review
open-review review --diff main

Works immediately with built-in methodology, auto-detected presets, and default settings.

With Init Wizard

cd your-project
open-review init

This creates:

• .open-review/config.yml — Configuration file
• .open-review/presets/ — Framework-specific presets (if detected)
• Agent skill files (.cursorrules, CLAUDE.md, AGENTS.md)
• .github/workflows/open-review.yml — GitHub Action workflow

Set API Key

export OPEN_REVIEW_API_KEY=your_key_here

Configuration

.open-review/config.yml:

version: "1.0"

review:
  methodology: default      # 'default' (built-in) or path to custom file
  presets: auto             # 'auto' (detect) or [laravel, react]
  conventions: auto         # 'auto' (discover), path, or inline text

# LLM Settings (optional)
llm:
  provider: anthropic
  model: claude-sonnet-4-20250514

# Output Settings (optional)
output:
  format: human
  timezone: America/New_York

Conventions

The conventions field supports three values:

# Auto-discover from codebase
conventions: auto

# Path to conventions file
conventions: ./docs/standards.md

# Inline text
conventions: "We use HasRealm for tenancy. Controllers are CRUD-only."

CLI Commands

open-review init

Interactive wizard to set up Open Review in your project.

open-review publish

Copy built-in methodology files to .open-review/methodology/ for customization.

open-review publish

open-review review

Run a code review.

# Review changes against main
open-review review --diff main

# Review staged changes
open-review review --diff staged

# JSON output for agents
open-review review --diff main --json

# Ephemeral focus
open-review review --diff main --prompt "Focus on authentication"

Agent Skills

Open Review generates thin skill files that point to your config:

• .cursorrules — For Cursor
• CLAUDE.md — For Claude Code
• AGENTS.md — For Open Code
• .ai/instructions.md — Generic

Skills are identical across projects. The agent reads the skill, reads the config, and executes the review.

Presets

Framework-specific review guides that add awareness at step 3 of the methodology.

Built-in presets: Laravel, React, Next.js, Vue.

# Auto-detected during init
open-review init

# Or specify in config
review:
  presets: [laravel, react]

Presets are copied to .open-review/presets/ where agents can read them.

GitHub Action

For automated PR reviews:

name: Open Review
on:
  pull_request:
    types: [opened, synchronize, reopened]

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: elliottlawson/open-review-action@v1
        with:
          api_key: ${{ secrets.OPEN_REVIEW_API_KEY }}

Architecture

open-review/
├── methodology/              # Core reasoning process (source of truth)
│   ├── core.md
│   ├── output-discipline.md
│   └── communication-style.md
│
├── skills/                   # Agent-specific instruction files
│   ├── cursor.md
│   ├── claude-code.md
│   ├── open-code.md
│   └── generic.md
│
├── presets/                  # Framework-specific review guides
│   ├── laravel.md
│   ├── react.md
│   ├── nextjs.md
│   └── vue.md
│
├── src/                      # The harness (TypeScript implementation)
│   ├── config/
│   ├── core/
│   ├── cli/
│   └── output/
│
└── templates/                # Config and workflow templates
    ├── config.yml
    └── workflow.yml

Environment Variables

• OPEN_REVIEW_API_KEY — API key for the configured LLM provider

License

MIT