convez.ai
AgenteAutomaçãoAvançado
atharvak161

Claude Organisation Framework

atharvak161/claude-org-framework

Organização de IA completa: clone, customize e rode em 10 minutos

Ver no GitHub
InicialPluginScore · de 0 a 100

Sistema operacional multi-agente de nível produção construído sobre Claude Code com 40+ agentes distribuídos em 12 departamentos (Engenharia, Produto, QA, DevOps, Design, Financeiro, RH, Legal, Marketing e mais). Coordena automaticamente fluxos de trabalho complexos via um Chief of Staff inteligente que interpreta objetivos em linguagem natural, delega tarefas aos agentes certos, gerencia dependências e aplica quality gates.

Instalação

$npx convez-ai atharvak161/claude-org-framework

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 1 mês
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Automatizar fluxos de desenvolvimento end-to-end (design → engenharia → QA → deploy)
  • Gerenciar projetos complexos com coordenação automática entre múltiplos times
  • Executar análises de segurança, testes de performance e code reviews em paralelo
  • Prototipagem acelerada de produtos SaaS com geração de landing pages + APIs

Veredito

Framework robusto e bem-pensado para quem quer delegar orquestração de projetos inteiros a agentes Claude. Excelente para prototipar processos empresariais de IA, mas requer familiaridade com Claude Code e disciplina na estrutura dos prompts dos agentes. Monitoramento em tempo real é um diferencial prático.

README

A tradução para o português deste README ainda está na fila.

Claude Organisation Framework

A production-grade multi-agent operating system built on top of Claude Code. Run a full engineering organisation — Chief of Staff, VP Engineering, Product, QA, Security, DevOps, Design, Finance, HR, Legal, Marketing and more — all coordinated through a single intelligent command layer.

Clone it, customise it, and have your own AI-powered organisation running in under 10 minutes.


What this is

Every "agent" is a CLAUDE.md file that turns Claude Code into a specific role. The Chief of Staff coordinates everything. You give it a goal in plain English — it breaks it down, spawns the right agents, manages dependencies, enforces quality gates, and delivers back to you.

You → Chief of Staff → Senior PM → [VP Eng → Frontend Dev → Backend Dev → QA → DevOps]
                                  ↘ Director Security ↗

The live monitor (monitor.html) shows every agent working in real time.


Quick start

# 1. Clone
git clone https://github.com/atharvak161/claude-org-framework.git
cd claude-org-framework

# 2. Set your workspace path
# Open WORKSPACE.md — replace the example path with your absolute path to this folder

# 3. Open Claude Code inside the chief_of_staff directory
cd agents/chief_of_staff
claude

# 4. Give it a goal
# "Build me a landing page for a SaaS product"
# "Set up a REST API with JWT authentication"
# "Review and fix all bugs in src/"
# "Design a full design system for a fintech app"

Live Monitor

Open monitor.html in any browser (no server needed — just double-click it). It polls org/ every 2 seconds and shows every agent action live.

To enable full file-level activity tracking, run this in a separate terminal:

bash org/watch_poll.sh

Agent roster — 40+ agents across 12 departments

Command

AgentRole
agents/chief_of_staffProgramme director. Coordinates everything. Start here.
agents/guide_explainerTranslates technical output to plain English for the Owner

Engineering

AgentRole
agents/engineering/vp_engineeringEngineering leadership and strategy
agents/engineering/frontend_developerUI/UX implementation + motion design (GSAP, Three.js, React)
agents/engineering/backend_developerAPIs, services, data models
agents/engineering/mobile_developeriOS/Android — React Native / Expo
agents/engineering/solution_architectSystem design, ADRs
agents/engineering/dev_team_leadSprint planning, code ownership
agents/engineering/code_reviewerCode quality gate before merge
agents/engineering/technical_writerDocs, READMEs, changelogs
agents/engineering/db_engineerSchema, migrations, query optimisation
agents/engineering/integration_engineerThird-party APIs, webhooks, OAuth

Product

AgentRole
agents/product/product_managerRequirements, roadmap, prioritisation
agents/product/ui_designerDesign system, tokens, components, visual QA
agents/product/ux_designerUser flows, wireframes, usability
agents/product/business_analystRequirements analysis, user stories

QA

AgentRole
agents/qa/director_qaQA strategy, sign-off gate
agents/qa/full_stack_testerEnd-to-end testing
agents/qa/test_automation_engineerCI test suites, automation frameworks
agents/qa/performance_testerLoad testing, Core Web Vitals

Security

AgentRole
agents/security/director_securitySecurity posture, mandatory sign-off
agents/security/penetration_testerVulnerability discovery
agents/security/security_engineerHardening, secure code review

DevOps

AgentRole
agents/devops/director_devopsInfrastructure strategy
agents/devops/cicd_engineerPipelines, GitHub Actions
agents/devops/infra_engineerCloud, IaC (Terraform, CDK)
agents/devops/container_engineerDocker, Kubernetes
agents/devops/sreReliability, SLOs, incident response
agents/devops/monitoring_engineerObservability, alerting, dashboards

More departments

agents/finance/ (8 agents) · agents/design/ · agents/hr/ · agents/legal/ · agents/marketing/ · agents/sales/ · agents/strategy/ · agents/data/ · agents/career/ · agents/operations/ · agents/pm/ · agents/pr/ · agents/research/ · agents/support/


Install Claude Skills (optional — strongly recommended)

These skills auto-activate when doing UI/frontend work. Install once, improve every session:

mkdir -p ~/.claude/skills

# Anthropic frontend-design (prevents generic AI-slop output)
git clone --depth=1 https://github.com/anthropics/skills.git /tmp/a-skills
cp -r /tmp/a-skills/skills/frontend-design ~/.claude/skills/

# Vercel bundle: web guidelines + React best practices + composition patterns
git clone --depth=1 https://github.com/vercel-labs/agent-skills.git /tmp/v-skills
cp -r /tmp/v-skills/skills/web-design-guidelines ~/.claude/skills/
cp -r /tmp/v-skills/skills/react-best-practices ~/.claude/skills/
cp -r /tmp/v-skills/skills/composition-patterns ~/.claude/skills/

# UI/UX Pro Max: 50 styles, 97 palettes, 57 font pairings, 99 UX guidelines
git clone --depth=1 https://github.com/nextlevelbuilder/ui-ux-pro-max-skill.git /tmp/uxpro
cp -r /tmp/uxpro/.claude/skills/ui-ux-pro-max ~/.claude/skills/

# Bencium innovative UX designer: 28,000-char UX reference document
git clone --depth=1 https://github.com/bencium/bencium-claude-code-design-skill.git /tmp/ben
cp -r /tmp/ben/bencium-innovative-ux-designer ~/.claude/skills/

File structure

├── agents/                   All agent CLAUDE.md files — the org itself
│   ├── chief_of_staff/       Start here
│   ├── engineering/
│   ├── product/
│   ├── qa/
│   ├── security/
│   ├── devops/
│   ├── finance/
│   ├── design/
│   └── ... (12 departments)
├── org/                      Shared state — agents write here, monitor reads here
│   ├── ACTIVITY.md           Real-time agent action log
│   ├── LIVE.md               Per-file live feed (written before every file touch)
│   ├── DECISIONS.md          Every decision recorded before implementation
│   ├── STATUS.md             Current sprint state
│   ├── BLOCKERS.md           Active blockers + escalation path
│   └── watch_poll.sh         File activity watcher (run to power the monitor)
├── review/                   Delivery gates
│   ├── SIGN_OFFS.md          Security + QA + DevOps + Docs sign-offs required
│   └── DELIVERY_SUMMARY.md   Plain-English delivery report for the Owner
├── monitor.html              Live dashboard — open in browser, no server needed
├── WORKSPACE.md              Master path reference + directory ownership map
└── README.md                 This file

How the Chief of Staff works

1. You give a goal in plain English:

Build a full-stack task management app with authentication

2. Chief of Staff produces a Master Execution Plan (MEP):

  • Departments involved
  • Risks and failure modes identified upfront
  • Success criteria that are measurable
  • Definition of done (not just working — tested, secured, documented, deployed)
  • Timeline estimate
  • Parallel workstreams mapped

3. You approve the MEP. Work begins.

Agents coordinate via shared files in org/. All decisions go to org/DECISIONS.md before implementation. All activity is logged to org/ACTIVITY.md immediately.

4. Quality gates before anything reaches you:

  • Security sign-off (review/SIGN_OFFS.md)
  • QA sign-off
  • DevOps confirms deployment stable
  • Technical Writer confirms docs complete

5. You receive a plain-English delivery report in review/DELIVERY_SUMMARY.md.


Customisation

WhatHow
Your nameReplace Atharva / Owner in agents/chief_of_staff/CLAUDE.md
Workspace pathUpdate the path in WORKSPACE.md to your absolute path
Add an agentCopy any folder under agents/, update CLAUDE.md, add to Chief of Staff's reporting chain
Remove an agentDelete the folder, remove from chief_of_staff/CLAUDE.md direct reports
Add a departmentCreate agents/[dept]/[role]/CLAUDE.md — model it on an existing agent

Principles

  • Every agent is a CLAUDE.md — readable, editable, version-controlled, no black boxes
  • All coordination is written — nothing assumed, everything logged in org/
  • Agents delegate, never implement — the Chief of Staff never writes a single line of code
  • Quality gates are mandatory — Security + QA + DevOps sign-off before anything ships
  • The monitor shows everything — open monitor.html, see exactly what every agent is doing in real time

Built with Claude Code. Clone it. Ship things.

Ver o original no GitHub

Bom para

Relacionados