Design the loop. Let agents run it.
设计闭环，让Agent去跑。
1 Rex Hub + 4 Feedback Loops + 7 Quality Gates
一个以Rex为中枢、以闭环为驱动的AI Agent执行框架。
An AI Agent execution framework driven by a Rex hub and closed-loop feedback.
RexLoop 是一个结构化的AI Agent执行框架，核心理念是：将复杂的AI任务分解为可执行、可审核、可学习、可进化的闭环系统。通过一个中枢调度器（Rex Hub）协调四条反馈闭环，并在关键节点设置质量门禁，确保Agent的输出始终可控、可追溯、可改进。
RexLoop is a structured AI Agent execution framework. Its core philosophy: decompose complex AI tasks into executable, auditable, learnable, and evolvable closed-loop systems. A central dispatcher (Rex Hub) coordinates four feedback loops, with quality gates at critical nodes to ensure Agent output is always controllable, traceable, and improvable.
┌─────────────────────────────┐
│ REX HUB │
│ (Central Dispatcher) │
└──────┬──────┬──────┬────────┘
│ │ │
┌──────────────────┘ │ └──────────────────┐
▼ ▼ ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ EXECUTION LOOP │ │ REVIEW LOOP │ │ LEARNING LOOP │
│ 执行闭环 │ │ 审核闭环 │ │ 学习闭环 │
│ Execute → Check │ │ Review → Approve│ │ Learn → Adapt │
└────────┬────────┘ └────────┬────────┘ └────────┬────────┘
│ │ │
▼ ▼ ▼
┌─────────────────────────────────────────────────────────────────┐
│ STRATEGIC LOOP / 战略闭环 │
│ Plan → Execute → Measure → Evolve │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ 7 QUALITY GATES / 7个质量门禁 │
│ [G1] Input Validation [G2] Plan Review │
│ [G3] Execution Check [G4] Output Verification │
│ [G5] Review Approval [G6] Learning Validation │
│ [G7] Strategic Alignment │
└─────────────────────────────────────────────────────────────────┘
| 组件 / Component | 说明 / Description |
|---|---|
| Rex Hub (Rex 中枢) | 中央调度器，负责任务分发、闭环协调和质量门禁管理 |
| Execution Loop (执行闭环) | Execute → Check — 任务执行与即时校验 |
| Review Loop (审核闭环) | Review → Approve — 人工/AI审核与批准流程 |
| Learning Loop (学习闭环) | Learn → Adapt — 从历史执行中提取经验并调整策略 |
| Strategic Loop (战略闭环) | Plan → Execute → Measure → Evolve — 战略层面的PDCA循环 |
| 7 Quality Gates (7个质量门禁) | 分布在各闭环关键节点的质量检查点 |
自动化循环的设计与编排引擎，支持定时触发、事件驱动和链式执行。
Design and orchestration engine for automation loops — supporting scheduled triggers, event-driven execution, and chained workflows.
基于工作树的并行任务隔离机制，每个Agent在独立的工作空间中执行任务，互不干扰。
Tree-based parallel task isolation — each Agent operates in an independent workspace without interference.
将可复用的知识和能力封装为标准化技能模块，支持跨Agent共享和版本管理。
Encapsulate reusable knowledge and capabilities as standardized skill modules — supporting cross-Agent sharing and version management.
通过标准化插件接口接入外部工具和服务，扩展Agent的能力边界。
Extend Agent capabilities through standardized plugin interfaces for external tools and services.
层级式子代理架构配合多层级记忆系统，实现复杂任务的分解与上下文持久化。
Hierarchical sub-agent architecture paired with multi-level memory systems for complex task decomposition and context persistence.
# Clone the repository
git clone https://github.com/VokoForge/rexloop.git
cd rexloop
# Install dependencies
npm install # or pip install -r requirements.txt
# Initialize your first Rex Hub
npx rexloop init
# Design your first loop
npx rexloop create --type execution --name "my-first-loop"
# Run it
npx rexloop run --loop my-first-loopNote: RexLoop is framework-agnostic. It works with any AI Agent runtime — OpenAI, Anthropic, local LLMs, or custom backends.
| Project | Description |
|---|---|
| loop-engineering | Loop Engineering 橙皮书 — 从提示词到闭环工程方法论 |
| managing-ai-agents | "给AI当老板" — 21章AI Agent团队搭建实战手册 |
| ai-prism | AI Prism — 多维视角下的AI工程实践 |
| voko-skills | AI Agent Skills Collection — 6大类技能模块 |
This project is licensed under the MIT License.
Built by VokoForge · Open Source · Community Driven