AgentSkillsCN

ralph-codex

基于 snarktank/ralph 的 Ralph 式自主代理循环，适配 OpenAI Codex CLI；适用于以 PRD 为导向的迭代开发（Ralph 循环、prd.json）。

SKILL.md
--- frontmatter
name: ralph-codex
description: Scaffold a Ralph-style autonomous agent loop (from snarktank/ralph) adapted for OpenAI Codex CLI; use for PRD-driven iterations (ralph loop, prd.json).
license: MIT
metadata:
  author: Y4rd13
  version: "1.0.0"
user-invocable: true

ralph-codex

The Job

Scaffold the Ralph loop into an existing repo under ./ralph/, including:

  • •ralph.sh, prompt.md, CODEX.md, prd.json.example, progress.txt
  • •.codex/rules/ralph.rules

How to run

After scaffolding:

  1. •Copy prd.json.example → prd.json and fill in your stories.
  2. •Run:
bash
cd ralph
./ralph.sh --tool codex 20

Notes

  • •The loop is intentionally simple: one story per iteration, progress stored in progress.txt, completion signaled by <promise>COMPLETE</promise>.
  • •Codex is invoked in non-interactive mode via codex exec --full-auto (see ralph.sh).