convez.ai
MCP ServerAutomaçãoAvançado
CertyPro

GitHub Control Plane para Agentes IA

CertyPro/certy-gh600-github-control-plane-lab

Orquestre agentes IA através de issues, labels e workflows do GitHub

Ver no GitHub
BomPluginScore · de 0 a 100

MCP server que transforma GitHub em um painel de controle para orquestração de agentes autônomos. Usa issues como intenção, labels para roteamento, pull requests para mudanças e GitHub Actions como executor, garantindo governança e auditoria em cada decisão.

Instalação

$npx convez-ai CertyPro/certy-gh600-github-control-plane-lab

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
100%
Último commit
há 2 meses
Versão
—
Issues abertas
5
Contribuidores
1

Casos de uso

  • Automação de tarefas de desenvolvimento com aprovação humana obrigatória
  • Roteamento de trabalho baseado em labels para diferentes agentes/papéis
  • Governança determinística e auditável de mudanças de código
  • Parseamento de issues em planos de execução para agentes

Veredito

Excelente para equipes que já usam GitHub e querem orquestração de agentes com governança built-in. Requer compreensão de GitHub Actions e workflows; o foco em auditoria e aprovação humana é a maior força aqui.

README

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

Certy GH-600: GitHub control plane lab

A public, fully runnable and tested TypeScript training repository for the Certy GH-600 GitHub Agentic AI Developer course. It teaches GitHub as the control plane for operating, supervising, and governing autonomous agents within a GitHub-driven delivery workflow.

What this lab teaches

GH-600 3.0 is about running agents through the primitives you already use to ship software. GitHub coordinates agents through:

  • Issues - the durable unit of intent, parsed into a plan before any action.
  • Labels - the routing signal that maps a task to an agent role.
  • Pull requests - the unit of change an agent produces and where it reports back.
  • Actions - the execution trigger and runtime for agent steps.
  • Checks - the execution gate; a change cannot merge until checks pass.
  • Reviews and CODEOWNERS - the human governance gate over sensitive files.

The through-line is governance: nothing merges silently, risky or protected work always needs a human, and every decision is deterministic and auditable.

Layout

src/automation/        Pure, deterministic control-plane functions
  labelRouter.ts         Map a label to an agent role
  issueToPlan.ts         Parse an issue body into a plan object
  approvalCheck.ts       Decide if a task needs human approval
  prCommenter.ts         Format an agent summary PR comment
tests/                 node:test suites for each module
docs/                  The operating model, lifecycle, taxonomy, approval flow
examples/              A worked agent task and the resulting PR comment
labs/                  Eight hands-on labs from the GH-600 brief
.github/               The control plane itself
  workflows/ci.yml         Real CI: install, test, typecheck
  workflows/agent-*.yml    Teaching examples, gated to workflow_dispatch
  ISSUE_TEMPLATE/          Agent task and bug report forms
  pull_request_template.md
  CODEOWNERS               Example governance map
  dependabot.yml

Run it

Requires Node 22 or newer.

npm install --no-fund --no-audit
npm test
npm run typecheck

The code is pure TypeScript with no runtime dependencies, no network, and no API calls, so every result is deterministic and reproducible.

The workflows

ci.yml is the real continuous-integration workflow and runs on every push and pull request. The agent-triage.yml, agent-review.yml, and human-approval.yml workflows are teaching examples: they are valid YAML but gated to on: workflow_dispatch so they never run automatically and never error on push. Each one explains, in comments, how a live trigger (an issue event, a PR event, a review event) would dispatch agent work under least-privilege permissions.

Licence

MIT. See LICENSE. Author: Certy (Coded Vision Design).

Ver o original no GitHub

Bom para

Relacionados