Token导航 LogoToken导航

Agent Skill · 研究检索

ralph-loop(拉尔夫·鲁普)

ralph-loop 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

ralph-loop 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证信息还不明确,第三方扫描有未通过项,按安装量在研究检索类的 73,675 个 Skill 里排第 55,093。

总安装 490周安装 20GitHub Stars 2下载 158核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:ralph-loop(拉尔夫·鲁普)
来源仓库:https://github.com/benjaming/ai-skills
仓库路径:skills/ralph-loop
安装命令:
npx skills add https://github.com/benjaming/ai-skills --skill ralph-loop
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/benjaming/ai-skills --skill ralph-loop

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

ralph-loop 创建自主迭代的任务处理循环,基于明确验收标准持续优化执行过程。适用于 Codex、Claude、Cursor、Gemini CLI,适用于 backlog 处理与 spec-driven 开发场景。包含任务系统、进度日志与知识积累机制,实现经验复用与效率持续提升。

初始化前需定义清晰终止条件与验收准则,避免无限循环或目标漂移。ralph-loop 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/ralph-loop

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/benjaming/ai-skills --skill ralph-loop;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

ralph-loop 怎么安装?

在终端执行 npx skills add https://github.com/benjaming/ai-skills --skill ralph-loop。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

ralph-loop 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

ralph-loop 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

ralph-loop 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

ralph-loop 用的人多吗?

公开数据是累计安装 490 次,近一周安装 20 次,下载 158 次,GitHub 2 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

ralph-loop 安全吗?

第三方扫描有未通过项,风险较高。除非你确认过源码,否则不建议直接安装到常用环境。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
ralph-loop

拉尔夫·鲁普 · 本页

4902需要联网未明确未通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Ralph Loop Creator

Create autonomous iterative loops that process a backlog of tasks with clear acceptance criteria. Named after the Ralph Wiggum pattern for spec-driven autonomous development.

Overview

A Ralph loop consists of:

  • Task System - Native Claude Code task management with dependency tracking
  • Prompt (prompt.md) - Instructions for each iteration
  • Runner (ralph.sh) - Bash script that loops until completion
  • Progress (progress.txt) - Execution log
  • Knowledge (knowledge.md) - Accumulated learnings

Quick Start

To create a new Ralph loop, gather requirements then run the init script:

python ~/.claude/skills/ralph-loop/scripts/init_ralph.py <output-dir>

Workflow

Step 1: Gather Requirements

Before creating a loop, clarify these parameters with the user using AskUserQuestion:

Loop Goal:

  • What is the global objective of this loop?
  • What domain/codebase does it operate on?

Task Structure:

  • What kind of tasks/items will be tracked?
  • What dependencies exist between tasks?
  • What metadata should each task have?

Iteration Process:

  • What steps should each iteration perform?
  • What tools/commands are needed?
  • What files should be read/written?

Acceptance Criteria:

  • How to validate a task is completed?
  • What checks should pass before marking done?
  • Should there be automated validation (tests, builds, typechecks)?

Exit Conditions:

  • When should the loop terminate? (all tasks done, specific marker, max iterations)
  • What completion marker to use? Default: <promise>COMPLETE</promise>

Step 2: Select or Create Template

Available templates in ~/.claude/skills/ralph-loop/templates/:

TemplateUse Case
migration.mdMigrating components, APIs, dependencies
exploration.mdCodebase analysis, documentation generation
testing.mdTest generation, coverage improvement
refactoring.mdCode cleanup, pattern enforcement
custom.mdBlank template for unique workflows

Step 3: Generate Loop Files

After gathering requirements, generate the loop structure:

python ~/.claude/skills/ralph-loop/scripts/init_ralph.py \
  --template migration \
  --output ./my-loop \
  --goal "Migrate React components from styled-components to Tailwind"

Or interactively:

python ~/.claude/skills/ralph-loop/scripts/init_ralph.py ./my-loop

Step 4: Create Tasks Using Task System

After generating the loop, create tasks using the native Task tools:

TaskCreate:
  subject: "Migrate Button component"
  description: "Convert Button from styled-components to Tailwind CSS classes"
  activeForm: "Migrating Button component"

Set up dependencies between tasks:

TaskUpdate:
  taskId: "2"
  addBlockedBy: ["1"]  # Task 2 waits for Task 1

Step 5: Run the Loop

cd <output-dir>
./ralph.sh [max-iterations]

Default max iterations: 50

File Structure

<output-dir>/
├── ralph.sh          # Executable bash loop
├── prompt.md         # Iteration instructions (uses Task tools)
├── progress.txt      # Execution log
├── knowledge.md      # Accumulated learnings
└── logs/             # Per-run log files

Task System Integration

The Ralph loop uses Claude Code's native Task management system:

Tools:

  • TaskCreate - Create new tasks with subject, description, activeForm
  • TaskList - View all tasks with status/owner/blockers
  • TaskGet - Fetch full task details by ID
  • TaskUpdate - Change status, set dependencies, mark complete

Status values: pending, in_progress, completed

Dependencies:

  • blocks - Tasks that cannot start until this one completes
  • blockedBy - Tasks that must complete before this one can start

Benefits over JSON backlog:

  • Native dependency tracking
  • Status validation (can't mark complete if tests fail)
  • Parallel agent coordination via owner field
  • Progress spinner shows activeForm text

Prompt Template Structure

# Ralph Agent: {Goal}

## Configuration
- List paths, files, tools

## Your Task (One Iteration)

### Step 1: Load Context
Use TaskList to get all tasks, read knowledge.md

### Step 2: Determine State
Check if all tasks are completed → output <promise>COMPLETE</promise>

### Step 3: Select Target
Use TaskList to find first task with:
- status: "pending"
- blockedBy: empty (no unresolved dependencies)
Use TaskUpdate to set status: "in_progress"

### Step 4-N: Execute
Task-specific steps using TaskGet for full details

### Final Step: Update
Use TaskUpdate to set status: "completed"
Update progress.txt and knowledge.md

## Completion
When all tasks completed: <promise>COMPLETE</promise>

Creating Initial Tasks

After generating the loop, create tasks for the backlog. Example for a migration loop:

# Create first task
TaskCreate:
  subject: "Migrate Button component"
  description: "Convert Button.tsx from styled-components to Tailwind. Update imports, replace styled() calls with className props using Tailwind utilities."
  activeForm: "Migrating Button component"

# Create dependent task
TaskCreate:
  subject: "Migrate Modal component"
  description: "Convert Modal.tsx. Depends on Button migration since Modal uses Button internally."
  activeForm: "Migrating Modal component"

# Set dependency
TaskUpdate:
  taskId: "2"
  addBlockedBy: ["1"]

Tip: Use TaskUpdate with addBlockedBy to ensure tasks run in correct order.

Best Practices

  1. One task per iteration - Keep iterations focused
  2. Clear acceptance criteria - Define what "done" means
  3. Use dependencies - Set blockedBy for ordered tasks
  4. Incremental commits - Commit after each successful task
  5. Knowledge accumulation - Document learnings for future iterations
  6. Fail fast - Stop on errors, don't silently continue
  7. Idempotent operations - Safe to re-run if interrupted

Troubleshooting

IssueSolution
Loop never completesCheck completion marker in prompt.md matches ralph.sh
Tasks skippedVerify status field updates in prompt.md
Context overflowReduce prompt.md size, move details to knowledge.md
Permission errorsEnsure ralph.sh has execute permissions

References

平台分布

Codex

35.11%
按下载量换算55

Claude

30.2%
按下载量换算48

Cursor

19.27%
按下载量换算30

Gemini CLI

9.57%
按下载量换算15

安全审计

Gen Agent Trust Hub

未通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills