——蓝紫罗兰）

** 一个人工智能原生的后端工程知识库。**

以神经知识图谱组织——每个文件都是一个具有明确目的、激活条件和因果关系的神经元。为人工智能代理（Claude、ChatGPT、Cursor、Codex、MiMoCode 等）构建，仅动态激活相关知识，应用质量门，并生成带有推理痕迹的生产级决策。

---

目录

• 这是什么？
• 其工作原理 — 神经结构
• 认知管道
• 路由与激活
• 四大优质发动机
• 证据协议
• 代码层级
• 建筑林特
• 里面是什么
• 知识领域
• 协议与引擎
• 大脑
• 战术手册
• 清单
• 完整性工具
• 安装
• 代理人应如何使用
• 优势与当前局限
• 许可证

---

这是什么？

这是一个纯Markdown知识库，零依赖关系。它被称为“Neural”，因为它被建模成一个大脑：

• Neurons = 单个 Markdown 文件，每个文件包含一段重点的后端知识。
• 根神经元 （ROOT.md]） = 告诉代理应读取哪些子神经元的路由器。

BACKEND ARCHITECT NEURAL

An AI-native backend engineering knowledge base.

Organized as a neural knowledge graph — every file is a neuron with a defined purpose, activation conditions, and causal relationships. Built for AI agents (Claude, ChatGPT, Cursor, Codex, MiMoCode, and others) to dynamically activate only the relevant knowledge, apply quality gates, and produce production-grade decisions with reasoning traces.

Repository: https://github.com/mobinhasanghasemi/backend-skill

---

Table of Contents

• What is this?
• How it Works — Neural Architecture
  • Cognitive Pipeline
  • Routing & Activation
  • The 4 Quality Engines
  • Evidence Protocol
  • Code Tiers
  • Architecture Linter
• What’s Inside
  • Knowledge Domains
  • Protocols & Engines
  • Brain
  • Playbooks
  • Checklists
  • Integrity Tools
• Installation
• How the Agent Should Use It
• Strengths & Current Limitations
• License

---

What is this?

This is a pure Markdown knowledge base with zero dependencies. It’s called “Neural” because it’s modeled as a brain:

• Neurons = individual Markdown files, each containing a focused piece of backend knowledge.
• Root neurons (ROOT.md) = routers that tell the agent which child neurons to read.
• Graph = explicit connections between neurons (depends_on, influences, conflicts_with, etc.).
• Engines = always-on quality layers that filter every decision through security, simplicity, performance, and reliability lenses.
• Evidence = every claim carries a confidence label (VERIFIED / SUPPORTED / INFERRED / EXPERIMENTAL / UNCERTAIN), preventing hallucination.

The agent never reads the whole tree. It follows the graph, activating only the neurons relevant to the question.

---

How it Works — Neural Architecture

Cognitive Pipeline

When the agent receives a question, it follows this pipeline (defined in BRAIN.md):

USER INPUT
    │
    ▼
PERCEPTION              ← parse problem, language, scope
    ▼
PROBLEM CLASSIFICATION  ← what kind of problem is this?
    ▼