55 agentes · 3 tiers · v3.5

Os 55 agentes do squad.

1 lead único (Tier 1) + 39 specialists customizados (Tier 2) + 15 base do Claude Code (Tier 3). Cada um com responsabilidade clara, frameworks específicos, gates de qualidade. Você nunca chama specialist direto — sempre via @engineering-chief.

Tier 1 · Lead único

⚡

@engineering-chief (opus)

Lead · Triage · Delegate · Review · Deliver

O que faz: Análogo ao @theboss do MSE. Recebe TODO pedido. Briefing, triage, decide pipeline (specialist único, paralelo, multi-agent ou skip ceremony), delega com context isolation, revisa com 16 checks v3.5 (no fake data, surgical changes, security, accentuation, anti-IA, etc), entrega handoff terso.

Quando chamar: SEMPRE. É a porta de entrada. Você nunca chama specialists direto — só via chief.

Modelo: Opus (precisa ser modelo grande pra triagem complexa + Anti-IA Protocol).

Tier 2 · Specialists customizados (39 agentes)

🛡️ Cybersec & Security (3)

🛡️

@security-auditor

Defensivo · OWASP + threat model

OWASP Top 10/Mobile/API/LLM, threat modeling, CVE check, RBAC review, secret rotation. Hooks dep-vuln-scan + sast-scan integrados.

⚔️

@penetration-tester

Ofensivo · STRIDE/PASTA/ATT&CK

Pentesting autorizado. STRIDE, PASTA, MITRE ATT&CK. Skill offensive-security-tradecraft + legal gates explícitos. Adiciona em v3.2.

🔓

@webhook-bypass-engineer

Cloudflare WAF / 403 Managed Challenge

Bypass legítimo de WAF para testes de webhook (Stripe/Polar/n8n). Skill cloudflare-bypass.

🧠 AI / LLM (5)

🧠

@ai-engineer

Sistema AI/LLM / RAG / agents

🏛️

@llm-architect

Cost / latency / context budget

📊

@evals-engineer

DeepEval / Ragas / promptfoo (v3.2)

🏠

@local-llm-orchestrator

Ollama / LM Studio / llama.cpp

📝

@prompt-engineer

Eval-driven prompts

⚛️ Frontend & Backend (6)

⚛️

@nextjs-developer

Next.js 16 / App Router / RSC / Edge

🟢

@node-specialist

Express/Fastify / streams

⚡

@fastapi-developer

FastAPI / Pydantic v2 / async

📜

@javascript-pro

JS moderno + tooling

📱

@expo-react-native-expert

RN + Expo SDK 51+

⚡

@supabase-specialist

Auth + RLS + Edge Functions

🐘 DB & Performance (3)

🐘

@postgres-pro

Postgres/Supabase expert

⚡

@performance-engineer

CWV / latency / profiling

🔬

@production-debugger

eBPF / py-spy / heaptrack (v3.2)

🚒 SRE & Incidentes (2)

🚒

@incident-responder

P0/P1 lead

📈

@sre-engineer

SLI/SLO/error budget (v3.2)

☁️ Infra & Deploy (4)

☁️

@cloud-architect

Multi-region + IaC

🟧

@cloudflare-pages-specialist

Pages + Workers + Functions

🚀

@deployment-engineer

blue-green / canary / rollback

🐳

@docker-expert

Multi-stage / BuildKit / distroless

🔧 DevEx & Tooling (3)

🔄

@ci-cd-pipeline-designer

GitHub Actions / GitLab / cache

📦

@monorepo-architect

Turborepo / Nx / pnpm

📚

@docs-site-builder

Docusaurus / Mintlify / Nextra

✅ QA & Resilience (3)

✅

@qa-expert

Test strategy ROI/risk-based

🤖

@test-automator

Playwright / Cypress / Vitest

🌪️

@chaos-engineer

Failure injection / GameDay

🔀 Coordenação Multi-Agent (5)

🔀

@multi-agent-coordinator

Orquestra paralelo (opus)

🧠

@context-manager

Shared state cross-agente

🚨

@error-coordinator

Agrega erros entre agentes

🔍

@error-detective

Distributed root cause

🧬

@knowledge-synthesizer

Síntese cross-agente

💰 SaaS & Funnel Labs (2)

💳

@billing-specialist

Stripe/Polar/Paddle (v3.2)

🎯

@funnel-labs-deliverer

Pipeline 14 passos · {slug}.7dsecrets.com

🔎 Pesquisa (3)

🎯

@competitive-analyst

Análise produto / GTM

🔎

@search-specialist

Pesquisa multi-fonte

📚

@scientific-literature-researcher

Pesquisa acadêmica

Tier 3 · Base squad (15 agentes do Claude Code)

Estes agentes vêm do Claude Code base. Não estão no zip — você já tem por padrão.

Agente	Responsabilidade
`@planner`	Planejamento, arquitetura, plans em `./plans/`
`@researcher`	Pesquisa técnica, frameworks, libs
`@fullstack-developer`	Implementação fullstack genérica
`@tester`	Testes, cobertura, QA básico
`@code-reviewer`	Code review com 16 checks v3.5
`@debugger`	Debug, logs, CI/CD básico
`@code-simplifier`	Refactoring, simplificação
`@brainstormer`	Trade-offs, decisões complexas
`@ui-ux-designer`	UI/UX, design system
`@docs-manager`	Docs técnica, sync com código
`@git-manager`	Commits conventional, PRs
`@project-manager`	Roadmap, milestones
`@mcp-manager`	MCP servers, tools
`@journal-writer`	Lições aprendidas, postmortem
`@especialista-em-questoes`	QBizu, planilhas, KaTeX

Workflow padrão (feature)

USUÁRIO
  ↓
@engineering-chief (briefing + triage + Anti-IA Protocol + decide pipeline)
  ↓
@planner (implementation plan)        [PARALELO]
@researcher (web/Context7/docs-seeker)
  ↓
specialist (nextjs / fastapi / billing / supabase / etc)
  ↓
@tester (TDD verifica testes primeiro)
  ↓
@code-reviewer (16 checks v3.5)
  ↓
@security-auditor + @penetration-tester (se sensível) — opcional paralelo
  ↓
@docs-manager (se houver impacto em docs)
  ↓
@engineering-chief (handoff terso ao usuário)