convez.ai
AgenteAutomaçãoAvançado
itohnobue

Orchestrator OpenCode

itohnobue/orchestrator-opencode

Orquestrador paralelo de agentes IA para refatoração e correção de código em escala

Ver no GitHub
BomPluginScore · de 0 a 100

Sistema avançado de orquestração de agentes para automação de tarefas complexas de desenvolvimento. Coordena até 110+ agentes especializados trabalhando em paralelo, com verificação adversarial de saídas e fluxo de trabalho dinâmico adaptado à complexidade da tarefa.

Instalação

$npx convez-ai itohnobue/orchestrator-opencode

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
9
Forks
1
Acompanhando
0
Saúde
42%
Último commit
há 2 dias
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Refatoração multi-arquivo com verificação automática de regressões
  • Correção de bugs críticos em codebases complexas com validação adversarial
  • Implementação de features grandes decompondo em subtarefas paralelas
  • Análise de segurança e performance com agentes especializados por domínio

Veredito

Solução robusta para automação de tarefas de desenvolvimento que vai além de edições simples, com arquitetura bem pensada de verificação adversarial para reduzir alucinações. Exige configuração inicial e integração com projeto existente, mas entrega valor significativo em tarefas complexas e críticas.

README

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

Orchestration Workflow

A parallel AI agent orchestrator for OpenCode. Instead of doing work itself, the lead decomposes your task, spawns specialist agents to do the actual work in parallel, verifies their output through an adversarial pipeline, and delivers production-ready results — all automatically. Works with any LLM provider.

Why use it

OpenCode is great at single-file edits, but complex tasks overwhelm a single context window. This workflow solves that:

  • Parallel execution — Up to 3 specialist agents work simultaneously on different parts of your task
  • Built-in verification — Every finding is adversarially checked against source code before it becomes a fix. False positives are caught and dropped automatically
  • Smart scoping — The planner researches your project first, classifies the task on 5 axes (size, domain breadth, ambiguity, severity, change type), then builds a custom workflow selecting only the stages the task actually needs. A cosmetic fix uses a handful of agents; a critical multi-file refactor gets full adversarial verification
  • Domain experts — 110+ specialized agents (from python-pro to security-reviewer to ios-pro), each with domain-specific checklists and anti-patterns

Quick Start

git clone https://github.com/itohnobue/orchestrator-opencode
cd orchestrator-opencode
./install.sh /path/to/your/project   # macOS/Linux
# or: .\install.ps1 C:\path\to\project   (Windows)

The installer copies the orchestration files into your project. Open your project with OpenCode — the workflow activates automatically when you give it a non-trivial task.

How it works

You ask: "Add dark mode to settings" or "Fix the payment race condition"
         │
         ▼
      Planner     Researches your codebase, finds relevant files,
         │        classifies the task, builds a custom workflow
         ▼
     Specialist    Up to 3 agents run in parallel — each gets a
      Agents       focused prompt with specific files, a domain
          │        persona, and questions they must answer
         ▼
   Verification   Every finding is adversary-checked against source code.
         │        Only confirmed issues survive. False positives are dropped.
         ▼
      Fixes       Confirmed issues are fixed by domain specialists,
         │        then independently reviewed for regressions
         ▼
   Deliverable    Clean commits, passing tests, verified code

Everything runs autonomously — the lead coordinates, agents do the work, verification catches mistakes.

Key concepts

Lead — The orchestrator. It doesn't write code. It researches your task, picks the right agents, writes their prompts, spawns them, and routes their findings through verification.

Agents — Specialist AI workers. Each one gets a narrow, well-defined task with a specific persona (swift-pro, code-reviewer, debugger). They work independently and write structured reports.

Verification — Before any finding becomes a fix, it goes through adversarial checking. An extraction agent deduplicates findings, then severity-routed agents try to falsify each one against the actual source code. Only findings that survive become fixes. This is what prevents hallucinated bugs from being "fixed."

Dynamic workflow — No fixed pipeline. The planner classifies your task on 5 axes (size, domain breadth, ambiguity, severity, change type) and assembles a custom stage plan. A cosmetic text change skips discovery and uses a single agent. A critical security fix gets full adversarial verification with multiple verification passes.

Requirements

  • OpenCode CLI
  • At least one LLM provider configured in ~/.config/opencode/opencode.json
  • uv (auto-installed if missing — handles Python dependencies for tools)

License

MIT

Ver o original no GitHub

Bom para

Relacionados