convez.ai
AgenteAutomaçãoAvançado
Dich01

Claude Code Multi-Agent Research

Dich01/claude-code-multi-agent-research

Pesquisa empírica de enforcement mecânico para multi-agentes Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Relatório técnico empírico sobre construção de plugins multi-agente confiáveis com Claude Code. Documenta discrepâncias CLI críticas, identifica exit code 2 como único mecanismo de bloqueio confiável, e fornece framework para orquestração complexa com até 112 agentes em produção.

Instalação

$npx convez-ai Dich01/claude-code-multi-agent-research

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
28%
Último commit
há 8 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Construir sistemas multi-agente com garantia de execução ordenada
  • Implementar enforcement de TDD em workflows automatizados
  • Bloquear operações inseguras em orquestração de agentes
  • Resolver discrepâncias CLI em plugins Claude Code
  • Escalar sistemas de agentes em produção (100+ agentes)

Veredito

Pesquisa valiosa e pragmática que expõe limitações reais de abordagens baseadas em prompts para multi-agentes. Crítica para quem precisa de guarantias mecânicas em orquestração complexa, mas exige compreensão profunda de CLI e exit codes.

README

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

Building Reliable Multi-Agent Claude Code Plugins

An Empirical Investigation of Enforcement Mechanisms

An empirical investigation into what actually works when building multi-agent plugins for Claude Code — and what doesn't, despite what the documentation says.

Why This Exists

Prompts are suggestions. The LLM can ignore them silently.

When you build a multi-agent system that needs to guarantee execution, order, enforce TDD, or block unsafe operations, you need mechanical enforcement — not prose.
This report documents 6 empirical tests, 12 GitHub issues, and community patterns from systems with up to 112 agents in production.

Key Findings

FindingStatus
exit code 2 in PreToolUse hooks blocks tool callsWorks
exit code 1 blocks tool callsDoes not block (by design)
permissionDecision: "deny" blocks tool callsDocs say yes, issue #4669 says no
Subagents can spawn subagentsNo — absolute restriction
Blocked Write/Edit? Claude bypasses via Bash (sed, echo >, python3 -c)Known vector — requires additional Bash hook
Hooks can trigger infinite loops without recursion guards3 open issues (#10205, #9579, #9704)
Agent frontmatter (hooks, skills) works in teammatesNo — bug #30703

Validated Patterns

  • Hub-and-spoke orchestration: A central "brain" agent as main
    session (--agent namespace:brain) spawns specialist subagents
  • Recursion guards: Temporary flag files prevent infinite hook
    loops when hooks propagate to subagents
  • Bash bypass defense: When Write/Edit are blocked, Claude uses sed, python3 -c, or echo > instead — a second hook on
    Bash closes this vector
  • Prerequisite gates: Hooks verify artifact existence before
    allowing specialist agents to execute

Empirical Validation

These patterns were validated through a complete Express-to-NestJS
migration of a multi-tenant production project:

  • 11 agents spawned with real parallelism (up to 3 concurrent)
  • Quality gate rejected first pass (44% coverage), forced correction to 93%
  • 422 tests, zero regressions, TypeScript strict with zero errors

Community References

  • Blake Crosley: 95 hooks in production over 9 months
  • kenryu42/claude-code-safety-net: semantic analysis + 5-level
    recursive wrapper detection
  • wshobson/agents: 112 agents, 16 orchestrators
  • Issue #29795: 5-layer QA system built from 68 documented failures

Full Report

Credits

Author: Diego Cheloni https://github.com/Dich01/

Date: March 14, 2026

Environment: Claude Code CLI (March 2026), Claude Opus 4.6

Ver o original no GitHub

Bom para

Relacionados