convez.ai
MCP ServerProdutividadeIntermediário
timurgaleev

vibestack — Workflows de Engenharia Sênior para Agentes de IA

timurgaleev/vibestack

Workflows profissionais de engenharia para agentes de IA, sem lock-in

Ver no GitHub
BomPluginScore · de 0 a 100

MCP server que instala um pacote portável de slash commands em assistentes de código (Claude Code, Cursor, Kiro). Fornece workflows estruturados como /office-hours, /tdd, /review e /ship que guiam desde a ideia até o merge de um pull request seguindo práticas de engenharia sênior.

Instalação

$npx convez-ai timurgaleev/vibestack

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
8
Forks
0
Acompanhando
0
Saúde
71%
Último commit
há 1 semana
Versão
v1.58.2
Issues abertas
0
Contribuidores
1

Casos de uso

  • Transformar ideias vagas em código pronto para produção
  • Implementar práticas de revisão cruzada entre modelos antes de deploy
  • Estruturar processos de TDD e revisão de segurança com IA

Veredito

Solução bem arquitetada e agnóstica que eleva a qualidade do output de assistentes de código ao impor workflows disciplinados. Ponto forte é portabilidade e zero lock-in; requer agente que suporte Agent Skills spec.

README

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

vibestack


What is this? (30-second version)

If you use an AI coding assistant (Claude Code, Cursor, or Kiro), vibestack adds a menu of expert workflows you trigger by typing /. Instead of "write me some code," you get /office-hours to shape the idea, /review to catch bugs, and /ship to open a clean pull request — each one a structured, opinionated process, not a vague prompt.

For engineers: 50+ portable SKILL.md workflows — planning, TDD, security audit, cross-model review, debugging, release. Same source installs into Claude Code, Cursor, and Kiro. No lock-in, no telemetry, state stays in ~/.vibestack/.

Think of it as turning a junior-level "just do it" assistant into one that plans, gets a second opinion, tests, and ships like a team would.


Install in 30 seconds

git clone https://github.com/timurgaleev/vibestack ~/.claude/skills/vibestack
~/.claude/skills/vibestack/install

That's it. Open a new session of your agent and type /office-hours:

Before we dig in — what's your goal with this?

  Building a startup (or thinking about it)
  Hackathon / demo — time-boxed, need to impress
  Open source / research — building for a community
  Learning — teaching yourself to code, leveling up

Pick a mode and it walks you through the right questions, then saves a design doc you can hand straight to /plan-eng-review. Every skill works like this — guided, structured, no filler. If /office-hours clicks, the rest will too.


Why people use it

  • 🚀 Idea → shipped, guided the whole way. /office-hours → /plan-eng-review → /tdd → /review → /ship. The chain is built for you.
  • 🧠 A real second opinion. Reviews run a cross-model check (a different AI) automatically, so two models have to agree before you ship.
  • 🔒 Yours, private. No telemetry, no accounts, no cloud. Everything lives on your machine in ~/.vibestack/.
  • 🔁 No lock-in. One source installs into Claude Code, Cursor, and Kiro alike. Switch tools, keep your workflow.
  • 📋 Copy, run, done. Two commands to install, git pull && ./install to update. Plain bash, zero runtime dependencies.

How it works

One set of skill files installs into whichever agent you use:

flowchart LR
  S["📄 One SKILL.md source"] --> I(["./install"])
  I --> A["Claude Code"]
  I --> B["Cursor"]
  I --> C["Kiro"]
  A --> U["Type /command → expert workflow runs"]
  B --> U
  C --> U

Each skill is a plain SKILL.md file your agent discovers and exposes as a /command. Install writes the file plus links to its helpers — so git pull && ./install is the entire update story.


A taste of the skills

50+ skills across planning, shipping, QA, design, security, and iOS. A few highlights:

CommandWhat it does
/office-hoursBrainstorm an idea into a concrete design doc
/plan-eng-reviewPressure-test a plan — architecture, data, risk
/tddTest-driven development, red-green-refactor
/reviewPre-merge review — correctness, security, tests
/shipMerge base, run tests, review, version bump, open the PR
/investigateSystematic debugging — no fix without a confirmed root cause
/csoSecurity audit — OWASP Top 10 + threat model

👉 Full list of every skill: docs/skills.md


More

./install --target=all      # Claude Code + Cursor + Kiro, non-interactive
./install --dry-run         # Preview every change, write nothing
git pull && ./install       # Update
~/.claude/skills/vibestack/uninstall --target=all   # Remove

Heads-up: the safety skills (/careful, /freeze, /guard) enforce hard blocks on Claude Code. On Cursor/Kiro they fall back to a soft LLM nudge — details in the compatibility audit above.


Ver o original no GitHub

Bom para

Relacionados