convez.ai
AgenteAutomaçãoAvançado
napheir

Governance Core

napheir/governance-core

Governança multi-agente pronta para Claude Code com hooks, propostas e constitucionalismo

Ver no GitHub
InicialPluginScore · de 0 a 100

Infrastructure reutilizável de governança multi-agente para projetos Claude Code. Fornece camada completa de conformidade com ganchos de segurança, fluxo de propostas, disciplina de encerramento, iteração constitucional e sincronização entre clones. Instale e execute `governance-core install` para equipar seu projeto imediatamente com controles, auditoria e coordenação de agentes.

Instalação

$npx convez-ai napheir/governance-core

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
2
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 1 dia
Versão
v0.53.0
Issues abertas
1
Contribuidores
1

Casos de uso

  • Coordenação segura de múltiplos agentes Claude em projetos complexos
  • Auditoria e rastreamento de decisões e modificações via fluxo de propostas
  • Aplicação de políticas de conformidade e proteção de dados em tempo real

Veredito

Ferramental robusto para equipes que precisam governar sistemas multi-agente em produção. A abordagem constitucional + ganchos de segurança + auditoria é rigorosa, mas requer autorização do mantenedor e consentimento para candidatos. Excelente para projetos onde conformidade não é negociável.

README

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

governance-core

PyPI License

Reusable multi-agent governance infrastructure for Claude Code projects.

Drop-in package providing a complete governance layer (constitution clauses, safety hooks, proposal workflow, wrap-up discipline, cross-clone sync) so you can focus on your project's business logic instead of re-inventing multi-agent coordination from scratch.

What you get

Install this package + run governance-core install in a new project, and your project immediately has:

  • 5 safety hooks (PreToolUse + PostToolUse): scope-guard, edit-write-guard, session-boundary-guard, command-guard, sensitive-data-guard — all configurable via .governance/config.json
  • Proposal workflow (/proposal skill): classify gate, draft / submit / approve / complete / archive state machine with audit trail
  • Wrap-up discipline (/wrap-up skill): STATE.md + git + knowledge publishing + skill learning in one command
  • Constitution iteration (/iterate-constitution skill): structured constitutional change workflow with red-line audit
  • Cross-clone coordination (/sync-infra, /sync-repos): physical-files sync + cross-repo git operations
  • 17 constitution clauses (art_00..art_16 + appendix) covering ritual, config management, contracts, scope governance, test/prod unification, git discipline, artifacts, constitutional protection, wrap-up discipline, memory staleness, etc.

Authorization

governance-core is source-available: you may read, audit, and fork the source freely. Using it to govern a project — running governance-core install / upgrade — requires a valid authorization code issued by the maintainer. install verifies the code offline (no network) against a public key bundled in the package; without a valid code the governance layer is not materialized.

install also requires candidate-uplink consent: improved skills, hooks, and mechanisms detected in your project may be packaged as candidates and uploaded to governance-core's public repository for review. In the current version this consent is mandatory — declining it aborts the install.

governance-core install --project-root . --auth-code GC1.<...> --accept-candidate-uplink

To obtain an authorization code, contact the maintainer. See LICENSE for terms.

Quick start

This package is one half of a 2-piece distribution. The other half is multi-agent-template — a cookiecutter template + bootstrap CLI that generates a new project skeleton.

Typical workflow:

# One-time setup (per machine)
pip install cookiecutter
pip install governance-core
pip install multi-agent-bootstrap

# Per-project: one-line bootstrap
multi-agent-bootstrap new my-project \
    --agents core,data \
    --ritual-phrase "Acknowledged"

# Verify
cd ~/workshop-claude/my-project
governance-core doctor --project-root .

See docs/architecture.md for the full picture (generic vs business boundary, config injection mechanism, upgrade flow, cross-clone coordination).

Standalone CLI usage

You can also install governance-core directly into an existing multi-agent project (without cookiecutter):

cd /path/to/your/project
pip install governance-core

# Write .governance/config.json with your project config
mkdir -p .governance
cat > .governance/config.json <<EOF
{
  "project_name": "my-project",
  "ritual_phrase": "OK",
  "core_agent_name": "core",
  "agents": [
    {"name": "core", "branch": "master", "clone_dir": "agent-core"}
  ]
}
EOF

# Render clauses + install hooks/skills (requires an authorization code)
governance-core install --project-root . --auth-code GC1.<...> --accept-candidate-uplink

# Validate
governance-core doctor --project-root .

Subsequent updates: git pull this repo, pip install -e . --upgrade, then governance-core upgrade --project-root /path/to/your/project (this preserves your .governance/config.json but refreshes clauses/hooks/skills).

Example content disclaimer

Some clauses and knowledge docs in this package contain examples drawn from the upstream project where governance-core was first developed (domain terminology, pipeline names, broker/API references, agent name conventions). These are explanatory examples, not requirements. Your project's .governance/config.json supplies its own agent names, ritual phrase, and clause keywords; the package's logic is project-agnostic.

A v1.0 release will template-ize all example tables; v0.1.0 ships with disclaimers attached to mixed clauses and methodology docs.

Project status

v0.1.0-alpha (2026-05):

  • API may break between minor versions (0.1.x)
  • Stable from 1.0.0 onwards
  • Bug reports + PRs welcome

License

MIT — see LICENSE.

Related

  • multi-agent-template — companion cookiecutter template + bootstrap CLI
  • Trade Agent (the project where this package was first developed) — private
Ver o original no GitHub

Bom para

Relacionados