https://github.com/salmundani/ralphex
Ralphex is a Claude Code plugin that uses Claude Code as planner + implementer, and Codex as code reviewer
https://github.com/salmundani/ralphex
ai anthropic claude claude-code claude-code-plugin code-review codex developer-tools openai
Last synced: about 2 months ago
JSON representation
Ralphex is a Claude Code plugin that uses Claude Code as planner + implementer, and Codex as code reviewer
- Host: GitHub
- URL: https://github.com/salmundani/ralphex
- Owner: salmundani
- License: mit
- Created: 2026-02-21T02:57:01.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2026-04-24T18:00:30.000Z (about 2 months ago)
- Last Synced: 2026-04-24T20:04:45.513Z (about 2 months ago)
- Topics: ai, anthropic, claude, claude-code, claude-code-plugin, code-review, codex, developer-tools, openai
- Homepage:
- Size: 29.3 KB
- Stars: 2
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
```
⠀⠀⠀⠀⠀⠀⣀⣤⣶⡶⢛⠟⡿⠻⢻⢿⢶⢦⣄⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⢀⣠⡾⡫⢊⠌⡐⢡⠊⢰⠁⡎⠘⡄⢢⠙⡛⡷⢤⡀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⢠⢪⢋⡞⢠⠃⡜⠀⠎⠀⠉⠀⠃⠀⠃⠀⠃⠙⠘⠊⢻⠦⠀⠀⠀⠀⠀⠀
⠀⠀⢇⡇⡜⠀⠜⠀⠁⠀⢀⠔⠉⠉⠑⠄⠀⠀⡰⠊⠉⠑⡄⡇⠀⠀⠀⠀⠀⠀
⠀⠀⡸⠧⠄⠀⠀⠀⠀⠀⠘⡀⠾⠀⠀⣸⠀⠀⢧⠀⠛⠀⠌⡇⠀⠀⠀⠀⠀⠀
⠀⠘⡇⠀⠀⠀⠀⠀⠀⠀⠀⠙⠒⠒⠚⠁⠈⠉⠲⡍⠒⠈⠀⡇⠀⠀⠀⠀⠀⠀
⠀⠀⠈⠲⣆⠀⠀⠀⠀⠀⠀⠀⠀⣠⠖⠉⡹⠤⠶⠁⠀⠀⠀⠈⢦⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠈⣦⡀⠀⠀⠀⠀⠧⣴⠁⠀⠘⠓⢲⣄⣀⣀⣀⡤⠔⠃⠀⠀⠀⠀⠀
⠀⠀⠀⠀⣜⠀⠈⠓⠦⢄⣀⣀⣸⠀⠀⠀⠀⠁⢈⢇⣼⡁⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⢠⠒⠛⠲⣄⠀⠀⠀⣠⠏⠀⠉⠲⣤⠀⢸⠋⢻⣤⡛⣄⠀⠀⠀⠀⠀⠀⠀
⠀⠀⢡⠀⠀⠀⠀⠉⢲⠾⠁⠀⠀⠀⠀⠈⢳⡾⣤⠟⠁⠹⣿⢆⠀⠀⠀⠀⠀⠀
⠀⢀⠼⣆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣼⠃⠀⠀⠀⠀⠀⠈⣧⠀⠀⠀⠀⠀
⠀⡏⠀⠘⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⣠⠞⠁⠀⠀⠀⠀⠀⠀⠀⢸⣧⠀⠀⠀⠀
⢰⣄⠀⠀⠀⠉⠳⠦⣤⣤⡤⠴⠖⠋⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢯⣆⠀⠀⠀
⢸⣉⠉⠓⠲⢦⣤⣄⣀⣀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣀⣀⣀⣠⣼⢹⡄⠀⠀
⠘⡍⠙⠒⠶⢤⣄⣈⣉⡉⠉⠙⠛⠛⠛⠛⠛⠛⢻⠉⠉⠉⢙⣏⣁⣸⠇⡇⠀⠀
⠀⢣⠀⠀⠀⠀⠀⠀⠉⠉⠉⠙⠛⠛⠛⠛⠛⠛⠛⠒⠒⠒⠋⠉⠀⠸⠚⢇⠀⠀
⠀⠀⢧⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⠇⢤⣨⠇⠀
⠀⠀⠀⢧⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣤⢻⡀⣸⠀⠀⠀
⠀⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢹⠛⠉⠁⠀⠀⠀
⠀⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀⠀⢠⢄⣀⣤⠤⠴⠒⠀⠀⠀⠀⢸⠀⠀⠀⠀⠀⠀
⠀⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀⠀⡇⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀⠘⡆⠀⠀⠀⠀⠀
⠀⠀⠀⡎⠀⠀⠀⠀⠀⠀⠀⠀⢷⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀⠀⡇⠀⠀⠀⠀⠀
⠀⠀⢀⡷⢤⣤⣀⣀⣀⣀⣠⠤⠾⣤⣀⡘⠛⠶⠶⠶⠶⠖⠒⠋⠙⠓⠲⢤⣀⠀
⠀⠀⠘⠧⣀⡀⠈⠉⠉⠁⠀⠀⠀⠀⠈⠙⠳⣤⣄⣀⣀⣀⠀⠀⠀⠀⠀⢀⣈⡇
⠀⠀⠀⠀⠀⠉⠛⠲⠤⠤⢤⣤⣄⣀⣀⣀⣀⡸⠇⠀⠀⠀⠉⠉⠉⠉⠉⠉⠁⠀
RRRRRRRRRRRRRRRRR lllllll hhhhhhh
R::::::::::::::::R l:::::l h:::::h
R::::::RRRRRR:::::R l:::::l h:::::h
RR:::::R R:::::R l:::::l h:::::h
R::::R R:::::R aaaaaaaaaaaaa l::::lppppp ppppppppp h::::h hhhhh eeeeeeeeeeee xxxxxxx xxxxxxx
R::::R R:::::R a::::::::::::a l::::lp::::ppp:::::::::p h::::hh:::::hhh ee::::::::::::ee x:::::x x:::::x
R::::RRRRRR:::::R aaaaaaaaa:::::a l::::lp:::::::::::::::::p h::::::::::::::hh e::::::eeeee:::::eex:::::x x:::::x
R:::::::::::::RR a::::a l::::lpp::::::ppppp::::::ph:::::::hhh::::::h e::::::e e:::::e x:::::xx:::::x
R::::RRRRRR:::::R aaaaaaa:::::a l::::l p:::::p p:::::ph::::::h h::::::he:::::::eeeee::::::e x::::::::::x
R::::R R:::::R aa::::::::::::a l::::l p:::::p p:::::ph:::::h h:::::he:::::::::::::::::e x::::::::x
R::::R R:::::R a::::aaaa::::::a l::::l p:::::p p:::::ph:::::h h:::::he::::::eeeeeeeeeee x::::::::x
R::::R R:::::Ra::::a a:::::a l::::l p:::::p p::::::ph:::::h h:::::he:::::::e x::::::::::x
RR:::::R R:::::Ra::::a a:::::a l::::::lp:::::ppppp:::::::ph:::::h h:::::he::::::::e x:::::xx:::::x
R::::::R R:::::Ra:::::aaaa::::::a l::::::lp::::::::::::::::p h:::::h h:::::h e::::::::eeeeeeee x:::::x x:::::x
R::::::R R:::::R a::::::::::aa:::al::::::lp::::::::::::::pp h:::::h h:::::h ee:::::::::::::e x:::::x x:::::x
RRRRRRRR RRRRRRR aaaaaaaaaa aaaallllllllp::::::pppppppp hhhhhhh hhhhhhh eeeeeeeeeeeeeexxxxxxx xxxxxxx
p:::::p
p:::::p
p:::::::p
p:::::::p
p:::::::p
ppppppppp
```
Ralphex is a Claude Code plugin that creates an automated development loop between two AI systems: **Claude Code** (planner + implementer) and **OpenAI Codex** (code reviewer). The loop repeats until Codex gives a clean review.
I usually prefer running a single agent model, but from my coding sessions, I feel that Claude is really good at planning and executing, while Codex really excels at code review.
## Installation
### Prerequisites
- [Claude Code](https://code.claude.com/docs)
- [OpenAI Codex CLI](https://github.com/openai/codex) installed and authenticated
### Install from marketplace
```shell
/plugin marketplace add salmundani/ralphex
/plugin install ralphex@salmundani-ralphex
```
### Usage
Once installed, run the skill from Claude Code:
```shell
/ralphex
```
Or for just code review:
```shell
/ralphex-review
```
To configure settings like the base branch, Codex model, or reasoning effort:
```shell
/ralphex-settings
```
## How it works
I was constantly doing the following loops manually, this plugin just automates them:
```shell
$ /ralphex
Prompt → Plan → Implementation → Commit → Codex Review → Claude filters reviews → End
↑ ↓ ↑ ↓
└──┘ └────── loop if there are corrections ────────┘
```
Sometimes I prefer to do the plan mode outside of the loop, and just do the code review loop at the end:
```shell
$ /ralphex-review
Prompt → Codex Review → Claude filters reviews → Implementation → Commit → End
↓ ↑
└────────── if no corrections ────────────┘
```
### Claude Code filters reviews
Codex reviews have a high chance of false positives. Therefore, there is a step where Claude Code filters the reviews to only include the ones that are actually useful.
- ✅ **Accept**: Real bugs, security issues, correctness problems.
- ❌ **Reject**: Factually wrong suggestions, misunderstandings, or purely stylistic nitpicks.
- ⏭️ **Defer**: Valid observations that are out of scope for the current task (e.g. pre-existing issues).
### Caveats
**IMPORTANT:** If you clear the context when accepting the plan in `/ralphex`'s Plan Mode, Claude will NOT do the commit → review → implement cycle at the end. To avoid this, you can accept the plan without clearing the context.
Eventually I will have to do a more robust solution using stop hooks.