Token导航 LogoToken导航TokenDH.com
研究检索操作浏览器clawhub未标认证来源可访问clear审计提醒

web-architecture网络架构

Agent Skill

web-architecture 用于查找、检索和筛选相关信息,适合在 OpenClaw 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。

总安装

29,426

周安装

1,264

GitHub Stars

公开资料未说明

下载量

10,314
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

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:web-architecture(网络架构)
来源仓库:https://github.com/michaelmonetized/web-architecture
安装命令:
openclaw skills install web-architecture
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 OpenClaw 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。

ClawHubOpenClaw
openclaw skills install web-architecture

简介

适用于复杂 TypeScript/Next.js/Convex 项目的多代理编排。分阶段构建、功能验证、无混乱地委托给子代理的完整剧本。

SKILL.md

name
web-architecture
description
Multi-agent orchestration for complex TypeScript/Next.js/Convex projects. Phased builds, functional verification, the full playbook for delegating to sub-agents without chaos.

Web Architecture

Multi-agent development workflow for TypeScript/Next.js/Convex projects.

*Born from: 29 agents, 50K lines, 212 errors, 1 hard lesson*


⚠️ SUB-AGENT POLICY (READ FIRST)

No Timeouts

Sub-agents MUST run without timeout limits. A 10-minute timeout that cuts off an agent mid-implementation leaves broken, partial code. Let agents finish.

Completion Means Working, Not Compiling

"Build passes" is necessary but NOT sufficient.

Before marking ANY phase complete, verify:

  1. Functions actually work — Call them, verify data flows
  2. UI actually renders data — Not just loading spinners forever
  3. User flows complete end-to-end — Click through, verify state changes persist
  4. Error states are handled — Not just happy path

The Lesson

An agent produced 15K lines of "working" code that:

  • ✅ Compiled with zero TypeScript errors
  • ✅ Passed bun run build
  • ❌ Had ZERO actual functionality
  • ❌ All data was mocked or hardcoded
  • ❌ Every button was a no-op

Self-grade: 5/10 — A prototype, not a product.


The Core Lesson

Single agent with full context > Many agents with partial context

29 parallel agents wrote 50K lines of code that didn't compile. Why?

  • No schema coordination → duplicate table definitions
  • No type contracts → frontend expected user.role, backend returned profile.plan
  • No initialization → npx convex dev never ran, no generated types
  • No integration checkpoints → errors discovered only at the end

The fix: One agent with full context rewrote the entire Convex backend in 11 minutes.


When to Use Multi-Agent

Good for parallel work:

  • Marketing pages (after design system exists)
  • Documentation files (independent)
  • Isolated features with clear contracts

Bad for parallel work:

  • Schema design (needs single owner)
  • Core type definitions (must be shared)
  • Interconnected backend functions
  • Component library (needs consistency)

The Workflow

Phase 0: Bootstrap (SEQUENTIAL — One Agent)

Must complete before spawning ANY other agents.

  1. Initialize project structure
  2. Initialize Convex: npx convex dev --once
  3. Create complete schema.ts (ALL tables)
  4. Run npx convex dev to generate types
  5. Create CONTRACTS.md (all data shapes)
  6. Create shared types in lib/types.ts
  7. Verify: bun run build passes

Deliverables:

  • [ ] convex/schema.ts — Complete, no TODOs
  • [ ] convex/_generated/ — Types generated
  • [ ] CONTRACTS.md — API shapes documented
  • [ ] lib/types.ts — Shared frontend types
  • [ ] bun run build — Passes with 0 errors

Phase 1: Foundation Documents (CAN BE PARALLEL)

Only spawn AFTER Phase 0 completes.

AgentOutputDependencies
Tech RequirementsTECH-REQ.mdNone
ComplianceCOMPLIANCE.mdNone
Design PrinciplesDESIGN.mdNone
Coding StandardsSTANDARDS.mdNone

Rule: These agents READ the schema. They do NOT modify it.


Phase 2: Backend Implementation (SEQUENTIAL or CAREFUL PARALLEL)

Option A: Single Backend Agent (Recommended)

  • One agent implements all Convex functions
  • Consistent patterns, no conflicts

Option B: Parallel with File Locks

  • Each agent owns specific files
  • NO shared file writes
  • Must reference CONTRACTS.md

Functional Requirements:

  1. Test CRUD operations — Create, read, update, delete
  2. Verify queries return data — Not empty arrays
  3. Check mutations persist — Data survives refresh
  4. Test auth guards — Protected functions reject unauthorized
  5. Verify indexes work — Queries return correct filtered data

Phase 3: Component Library (SEQUENTIAL)

Single agent builds the component library.

Why? Components reference each other. Parallel work creates duplicate components with different APIs.

Functional Requirements:

  1. Interactive states work — Buttons trigger onClick
  2. Form components submit — Not just styled divs
  3. Loading/error states exist
  4. Accessibility basics — Labels, ARIA, keyboard nav
  5. Consistent API — All components follow same patterns

Phase 4: Features & Pages (CAN BE PARALLEL)

Now safe to parallelize because schema is locked, types exist, components exist.

AgentScopeCan Modify
Admin Suite/app/(admin)/**Own files only
Support Portal/app/(support)/**Own files only
Marketing Pages/app/(marketing)/**Own files only
User Flows/app/(app)/**Own files only

Rules:

  1. Read schema, types, contracts — don't modify
  2. Use existing components — don't recreate
  3. Write to assigned directories only

Functional Requirements:

  • [ ] Page loads without console errors
  • [ ] Data appears (not mock/placeholder)
  • [ ] Forms submit and persist data
  • [ ] Can complete full user flow (create → view → edit → delete)
  • [ ] Refresh preserves state

Red flags (NOT complete):

  • // TODO comments in business logic
  • Hardcoded arrays instead of useQuery
  • onClick handlers that console.log instead of mutate
  • "Coming soon" placeholders in core features

Phase 5: Integration & QA (SEQUENTIAL)

  1. bun run build (must pass)
  2. npx convex dev --once (must pass)
  3. Generate sitemap from routes
  4. Route crawl & 404 check
  5. Browser smoke test (all routes return 200)
  6. End-to-end flow verification

E2E Verification Checklist:

Auth Flow:

  • [ ] Sign up creates user in database
  • [ ] Sign in authenticates and redirects
  • [ ] Protected routes redirect to sign-in

Core CRUD Flow:

  • [ ] Create: Form submits → record appears
  • [ ] Read: List shows real data
  • [ ] Update: Edit form saves → changes persist
  • [ ] Delete: Remove action → record gone

Directory Structure

project/
├── convex/
│   ├── schema.ts            # 🔒 Phase 0 only
│   ├── _generated/          # 🔒 Auto-generated
│   └── [domain].ts
├── lib/
│   ├── types.ts             # 🔒 Phase 0 only
│   └── utils.ts
├── components/
│   ├── ui/                  # Component library agent
│   └── [domain]/            # Feature agents
├── app/
│   ├── (admin)/             # Admin agent
│   ├── (app)/               # App agent
│   └── (marketing)/         # Marketing agents
└── CONTRACTS.md             # 🔒 Phase 0 only

🔒 = Locked after Phase 0. Agents read, don't modify.


Agent Spawn Order

1. Bootstrap Agent (MUST COMPLETE FIRST)
   └── schema.ts, types, contracts
   
2. Doc Agents (parallel)
   ├── TECH-REQ.md
   ├── COMPLIANCE.md
   └── DESIGN.md
   
3. Backend Agent (single)
   └── All convex/*.ts functions
   
4. Component Agent (single)
   └── All components/ui/*
   
5. Feature Agents (parallel, isolated directories)
   ├── Admin Suite
   ├── Support Portal
   ├── Marketing Pages
   └── User Flows
   
6. Integration Agent (single)
   └── Final build, fixes, QA

Anti-Patterns

Spawn all agents at once — No coordination, duplicate work

Let agents invent types — Use CONTRACTS.md, not imagination

Skip Phase 0 — "We'll figure out the schema later" = disaster

Parallel schema writes — One owner only

Frontend before backend types — Generates type mismatches

No build checkpoints — Errors compound


Related Files

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

79.07%
按下载量换算8,155

安全审计

VirusTotal

通过

ClawScan

可疑

Static analysis

未展示

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills