convez.ai
AgenteAutomaçãoAvançado
WizardingCode-io

wzrdxOS

WizardingCode-io/wzrdxos

SO open-source para times de agentes IA com orquestração nativa Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Sistema operacional open-source para times de agentes de IA, construído nativamente sobre primitivas do Claude Code (Skills, Workflows, subagentes). Oferece orquestração leve e modular sem engine de orquestração customizado, com foco em skills curados, workflows de qualidade e base de conhecimento como cidadão de primeira classe.

Instalação

$npx convez-ai WizardingCode-io/wzrdxos

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
0
Forks
0
Acompanhando
0
Saúde
62%
Último commit
há 4 meses
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Orquestração de múltiplos agentes especializados em departamentos
  • Execução de workflows com verificação adversarial e síntese citada
  • Ingestão e RAG de base de conhecimento integrada a Claude Code via MCP
  • Composição modular de skills reutilizáveis entre agentes

Veredito

Projeto ambicioso que simplifica orquestração de agentes aproveitando nativamente primitivas do Claude Code, evitando overhead de engines customizados. Ainda em M1 (foundations), com arquitetura híbrida TS/Python promissora, mas requer experiência sólida em arquitetura de sistemas para adoção.

README

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

wzrdxOS

The open-source operating system for AI agent teams.

wzrdxOS is a lightweight, native-first framework for running a company of AI agents on top of Claude Code primitives — skills, workflows, subagents and memory — without the weight of a bespoke orchestration engine.

It is a clean-slate successor to ArkaOS, built around three target inspirations:

  • Native Claude Code primitives — the Skill tool, the Workflow tool, native subagents and plan mode do the orchestration. wzrdxOS resolves who does the work and with which workflow; it does not reinvent orchestration.
  • Modular skills (superpowers-style) — small, composable, disciplined skills instead of monolithic flows.
  • Quality workflows — fan-out + adversarial verification + cited synthesis, judge panels, loop-until-dry.

Why wzrdxOS

ArkaOS grew into a heavy Python engine with a non-negotiable 13-phase flow, blocking enforcement hooks, 270+ skills and 65 agents. wzrdxOS keeps the company with departments metaphor but makes it lean:

  • Opt-in flow, no blocking enforcement. Quality comes from workflow patterns, not gates that stall the work.
  • Curated skills, not accumulated. A designed taxonomy plus a generator.
  • Few, strong agents. A small roster of high-quality specialists.
  • Knowledge base as a first-class citizen. A purpose-built ingestion / RAG pipeline exposed to Claude Code over MCP.

Architecture

A hybrid monorepo with a clean runtime boundary:

wzrdxOS
├─ packages/core   (TypeScript)  registry · routing · conductor · CLI · artifact generator
├─ services/kb     (Python)      ingestion · embeddings · vector index · RAG → MCP server
└─ artifacts/                    Claude Code artifacts (curated + generated)
   ├─ skills/      SKILL.md (superpowers-style)
   ├─ agents/      agent / persona definitions
   └─ workflows/   Workflow scripts (reusable quality patterns)
  • @wzrdx/core (TS): source of truth for the registry (skills, agents, departments, workflows), the natural-language routing resolver, a light conductor, profile/config, the KB client, light telemetry, and the wzrdx CLI. It generates and installs Claude Code artifacts into ~/.claude/.
  • @wzrdx/kb (Python, via uv): the isolated worker for everything TS does poorly — ingestion (YouTube/Whisper, articles, URLs, files), chunking, embeddings, a vector index and semantic search, exposed to Claude Code as the wzrdx-kb MCP server.

Status

M1 — Foundations & skeleton. This milestone delivers firm ground and the artifact format contracts, not domain functionality. See docs/roadmap.md for the full phased plan and the dedicated design sessions for the Knowledge Base (M2) and the Org / Teams roster (M3).

Quick start (development)

pnpm install
pnpm build
pnpm wzrdx -- doctor    # validate the environment
pnpm wzrdx -- status    # list registered skills / agents / workflows

Requirements

  • Node.js >= 22 + pnpm 10
  • Python 3.13 + uv (for services/kb)

License

MIT © WizardingCode

Ver o original no GitHub

Bom para

Relacionados