Ralph Codex

Persistent operator-loop skill for OpenAI Codex.

This repository publishes a single skill, ralph-codex, which keeps a repo-local native-hook loop alive across turns until one verified outcome is complete.

Install

Official skills.sh flow:

npx skills add https://github.com/Skyline-23/ralph-codex -a codex -g -y

Fallback if npx skills does not resolve correctly in your npm environment:

npm exec --package=skills@latest -- skills add https://github.com/Skyline-23/ralph-codex -a codex -g -y

Use

Start the loop from the target repository:

bash scripts/ralph_loop.sh start --goal "Ship one verified outcome"

Optional guardrail:

bash scripts/ralph_loop.sh start --goal "Ship one verified outcome" --max-turns 12

Inspect or stop:

bash scripts/ralph_loop.sh status
bash scripts/ralph_loop.sh stop
bash scripts/ralph_loop.sh resume

Notes

• The skill definition lives in SKILL.md.
• Native hooks are experimental and intended for macOS/Linux-style environments.
• Loop state is written under .codex-ralph/ in the target repository.