convez.ai
AgenteDevOps & InfraestruturaAvançado
drewrad8

Strategos

drewrad8/strategos

Orquestrador web para agentes IA em tmux com hierarquias e monitoramento live

Ver no GitHub
InicialPluginScore · de 0 a 100

Dashboard web para orquestrar agentes IA de coding (Claude Code, Gemini CLI) em sessões tmux. Dispara workers isolados, monitora terminals em tempo real, organiza hierarchias de agentes e permite coordenação autônoma com modo Bulldoze para tasks de longa duração.

Instalação

$npx convez-ai drewrad8/strategos

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
57%
Último commit
há 5 meses
Versão
v2.0.0
Issues abertas
0
Contribuidores
1

Casos de uso

  • Coordenar múltiplos agentes Claude/Gemini em paralelo para debugging de bugs
  • Organizar pesquisa, implementação e testes em hierarquias (GENERAL > COLONEL > especialistas)
  • Automação de fluxos de desenvolvimento com continuação autônoma em modo Bulldoze
  • Monitoramento visual de progresso com health status e terminal live

Veredito

Strategos é uma ferramenta sofisticada para quem trabalha com múltiplos agentes IA de coding e precisa coordenação em escala. A integração MCP com VSCode, Claude Code e Cursor abre possibilidades avançadas, mas requer Node.js, tmux e CLIs de IA instalados — não é para iniciantes. Prático para equipes de desenvolvimento que rodam experimentos de automação de código.

README

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

Strategos

A web dashboard for managing AI coding agents. Spawn Claude Code or Gemini CLI workers in tmux sessions, watch their terminals live, organize them into hierarchies, and let them coordinate autonomously.

Install

git clone https://github.com/drewrad8/strategos.git
cd strategos
npm install
npm run build
npm run server

Open http://localhost:38007.

Requirements

  • Node.js 20+
  • tmux
  • Claude Code CLI (npm i -g @anthropic-ai/claude-code) or Gemini CLI

What it does

You give workers a task. They run in isolated tmux sessions. You see their terminal output live in your browser.

Workers can be organized into hierarchies — a GENERAL spawns COLONELs who manage IMPL/TEST/RESEARCH specialists. Progress bubbles up automatically via the Ralph protocol: workers signal in_progress, done, or blocked with a completion percentage.

Bulldoze mode keeps workers going autonomously — when a worker stalls, it gets a continuation prompt. Useful for long-running tasks where you don't want to babysit.

The dashboard shows worker cards, a tree view of hierarchies, live terminals, health status, and project organization. Keyboard shortcuts for everything (press ?).

Spawning workers

From the UI, click "Spawn Worker" and pick a template.

From the API:

curl -X POST http://localhost:38007/api/workers/spawn-from-template \
  -H "Content-Type: application/json" \
  -d '{
    "template": "impl",
    "label": "IMPL: fix-login-bug",
    "projectPath": "/path/to/your/project",
    "task": {"description": "Fix the login timeout bug in auth.js"}
  }'

Templates: research, impl, test, review, fix, colonel, general

Configuration

All via environment variables. The defaults work out of the box.

VariableDefaultWhat it does
PORT38007Server port
THEA_ROOT/theaWhere to scan for projects
STRATEGOS_API_KEY(none)Set this to require auth (min 16 chars)
ENABLE_OLLAMA_SUMMARIESfalseAI-powered worker summaries via local Ollama

See docs/STRATEGOS_USAGE_GUIDE.md for the full API reference.

IDE Integration (MCP)

Strategos includes an MCP server that lets AI assistants in your IDE manage workers directly. Works with VSCode Copilot, Claude Code, Claude Desktop, Cursor, Cline, and Continue.

18 tools (spawn, kill, monitor, signal), 3 resources (workers, health, tree), and 3 prompts (deploy, briefing, review).

VSCode (Copilot Chat)

The repo includes .vscode/mcp.json — open the project in VSCode and the Strategos MCP server is available automatically. If connecting via Remote-SSH, stdio transport works since the MCP process runs on the remote machine.

Edit .vscode/mcp.json to adjust the path if your install location differs:

{
  "servers": {
    "strategos": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/mcps/strategos/dist/server.js"],
      "env": {
        "STRATEGOS_URL": "http://localhost:38007"
      }
    }
  }
}

Claude Code / Claude Desktop

Copy .mcp.example.json to .mcp.json and update the path:

{
  "mcpServers": {
    "strategos": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/mcps/strategos/dist/server.js"],
      "env": {
        "STRATEGOS_URL": "http://localhost:38007"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global) using the same mcpServers format as above.

Using MCP Prompts

Once connected, use slash commands in your AI chat:

  • /strategos.deploy — Spawn a new worker with Commander's Intent
  • /strategos.briefing — Get a full operational overview
  • /strategos.review — Review a worker's output and dismiss it

MCP Resources

Attach live data as context in your chat via "Add Context > MCP Resources":

  • Active Workers — all running workers with status and progress
  • System Health — server health, worker counts by status/health
  • Worker Hierarchy — parent-child tree view
  • Worker Output — terminal output from any specific worker

Running as a service

cp strategos.service ~/.config/systemd/user/strategos.service
# Edit paths in the file to match your setup
systemctl --user enable --now strategos

Uses KillMode=process so server restarts don't kill your running workers.

Development

npm run dev    # Server on :38007, Vite HMR on :38008
npm test       # Playwright e2e tests

License

ISC

Ver o original no GitHub

Bom para

Relacionados