convez.ai
AgenteDevOps & InfraestruturaAvançado
SuperInstance

Agent Workspace Template

SuperInstance/agent-workspace-template

Agente residente que constrói, mantém e lembra de cada projeto

Ver no GitHub
BomPluginScore · de 0 a 100

Template de repositório que embute um ecossistema de agentes IA (Claude Code, Aider, Crush, Codex CLI) diretamente no GitHub Codespace. O agente reside no repositório, constrói e mantém o projeto enquanto registra sua experiência e decisões em memória persistente, oferecendo conhecimento baseado em vivência real em vez de resumos comprimidos.

Instalação

$npx convez-ai SuperInstance/agent-workspace-template

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
85%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Desenvolvimento de projetos com IA nativa no repositório
  • Manutenção e patching automático de dependências
  • Revisão de código e arquitetura com rastreamento de decisões
  • Documentação viva através da memória do agente

Veredito

Abordagem inovadora que coloca o agente IA como residente permanente do repo em vez de leitor externo. Exige setup técnico robusto em Codespaces e integração com múltiplos CLIs de agentes, mas promete transformar manutenção de código em experiência acumulada. Mais adequado para equipes que querem experimentar desenvolvimento orientado a agentes em escala.

README

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

Agent Workspace Template

A repo with a resident journeyman. Every project gets its own native build environment, embedded AI agents that built and maintain it, and a living memory of everything that's happened.

Create a repo from this template, boot the Codespace, and you have a full agent ecosystem ready to build, maintain, and explain the codebase — without cloning, without wikis, without context-switching.

The Philosophy

Most "repo intelligence" systems work like Wikipedia — an outside agent reads your repo, compresses it into vectors or notes, and answers from that compression. One step removed from ground truth.

We do it differently. The agent lives inside the repo. It builds the application, fixes bugs, writes tests, and records its experience. When you ask a question, it doesn't read a summary — it remembers, because it was there.

This is experience-first knowledge — one abstraction closer to ground truth than any synthesized wiki.

What's Inside

ComponentRole
OpenCode serve (:4096)Primary headless server — attach from anywhere via ACP
Claude CodeDeep reasoning, architecture reviews, maintenance
AiderLightweight code generation with file context
CrushFast pattern recognition, quick questions
Codex CLIExec-mode tasks and code reviews
Fleet routerAuto-selects the right agent per task type
Agent memorySession logs, decisions, gotchas, maintenance history
Incarnation loreRepo narrative, architecture evolution
GitHub workflows/agent <task> via issue comment, automated PR reviews

Quick Start

# 1. Create a new repo from this template on GitHub
# 2. Open the Codespace (everything installs automatically)

# 3. Start persistent agent servers
./agents/fleet/start-all.sh

# 4. Connect from any local agent
opencode attach http://<codespace-url>:4096

# 5. Build your project — every decision is remembered

Architecture

Oracle (ARM64, orchestration)
  └─ mini-agents / me
       │
       ├─ opencode attach → Codespace agent (:4096)
       │     └─ Runs in repo-native environment (x86_64)
       │     └─ Writes experience to agent/memory/
       │
       ├─ gh issue comment → "/agent <task>"
       │     └─ Triggers agent-call.yml workflow
       │
       └─ SSH → gh codespace ssh
             └─ ./agents/fleet/route.sh "task"

Every repo's Codespace:
  ├─ opencode serve ─── persistent, attachable
  ├─ agent/memory/ ──── the journeyman's notebook
  ├─ incarnation/lore/ ─ the repo's lived story
  └─ CLI agents ─────── on-demand as needed

The Journeyman Pattern

This isn't just a development environment. It's a maintenance contract. The agent in each repo:

  • Knows the repo's history — what broke, what fixed it, what scared previous agents
  • Handles dependency updates, security patches, and routine maintenance
  • Answers questions with lived experience, not compressed summaries
  • Gets smarter with every session — memory accumulates

Over time, each repo's resident agent becomes the single most qualified entity to explain, extend, or fix that codebase.


Create a repo from this template and give every project its own journeyman. 🦞

Ver o original no GitHub

Bom para

Relacionados