https://github.com/albertoarena/llm-review-panel
PHP CLI that runs the same code/plan review across multiple LLM command-line tools (Claude Code, OpenCode, Gemini, Codex, Aider, Qwen) in parallel, then synthesizes them into one balanced result.
https://github.com/albertoarena/llm-review-panel
ai ai-tools aider claude-code cli code-review codex developer-tools gemini llm llm-tools ollama opencode php symfony-console
Last synced: 3 days ago
JSON representation
PHP CLI that runs the same code/plan review across multiple LLM command-line tools (Claude Code, OpenCode, Gemini, Codex, Aider, Qwen) in parallel, then synthesizes them into one balanced result.
- Host: GitHub
- URL: https://github.com/albertoarena/llm-review-panel
- Owner: albertoarena
- License: mit
- Created: 2026-06-17T07:32:42.000Z (22 days ago)
- Default Branch: main
- Last Pushed: 2026-07-03T05:49:58.000Z (6 days ago)
- Last Synced: 2026-07-03T06:26:53.602Z (6 days ago)
- Topics: ai, ai-tools, aider, claude-code, cli, code-review, codex, developer-tools, gemini, llm, llm-tools, ollama, opencode, php, symfony-console
- Language: PHP
- Homepage: https://albertoarena.github.io/llm-review-panel/
- Size: 206 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# llm-review-panel
[](https://github.com/albertoarena/llm-review-panel/actions/workflows/ci.yml)
[](https://github.com/albertoarena/llm-review-panel)
[](https://www.php.net/)
[](LICENSE)
A PHP CLI that runs the same review across multiple LLM command-line tools
(Claude Code, OpenCode, Gemini CLI, Codex, Aider, Qwen) in parallel, then
synthesizes their independent reviews into one consolidated result.
It exists to automate a manual workflow: ask several different LLMs to review
the same markdown plan, then compare viewpoints instead of trusting one model.
**Documentation: https://albertoarena.github.io/llm-review-panel/**
## How it works
```
plan.md ──> [rubric + plan] ──> N reviewers in parallel ──> synthesizer ──> consolidated review
│ │
checkpoint 2 checkpoint 3
```
Three human checkpoints: confirm the prompt, review raw outputs, accept the
synthesis. `--yes` skips them; `--dry-run` prints the resolved commands without
spawning anything.
## What it is not
- Not a framework app. Symfony Console + Symfony Process only.
- Not an API client. Reviewers are local CLI binaries you already have
installed and authenticated. The tool spawns processes; it never makes HTTP
calls to model APIs.
- Not per-provider code. One generic reviewer class, driven entirely by
`config.json`. Adding a new reviewer is a config edit.
## Requirements
- PHP 8.3+
- Composer
- At least one reviewer CLI on `PATH` (Claude Code, OpenCode, etc.). See
`docs/CONFIG.md` for the supported list.
## Quick start
```bash
composer install
cp config.example.json config.json
# edit config.json: enable the reviewers you have installed
bin/llm-review-panel review path/to/plan.md --dry-run # verify config
bin/llm-review-panel review path/to/plan.md
```
## Docs
- `CLAUDE.md` — constraints and conventions (for agents working on this repo)
- `docs/DESIGN.md` — architecture, flow, checkpoints, persistence
- `docs/CONFIG.md` — config schema and supported reviewers
- `docs/SCHEMA.md` — reviewer JSON output contract
- `docs/IMPLEMENTATION.md` — phased build plan
- `docs/CI.md` — CI and release automation
## Status
Working. All build phases are implemented and tested (CI runs Pint plus Pest on
PHP 8.3/8.4/8.5). The tool runs end-to-end against real reviewer CLIs. Early
days: expect rough edges and pre-1.0 changes.