claude-plugins/automacao/skills/n8n-chatbot/SKILL.md

/n8n-chatbot - Chatbots e AI Agents n8n

Criar chatbots e workflows com AI usando LangChain.

---

Uso

/n8n-chatbot create <descrição>    # Criar chatbot
/n8n-chatbot agent <tipo>          # Criar AI agent
/n8n-chatbot rag <knowledge_base>  # Criar sistema RAG

---

Workflow Obrigatório

1. Health check       → mcp__n8n__n8n_health_check()
2. Pesquisar template → mcp__n8n__search_templates({ task: "ai_automation" })
3. Se template OK     → mcp__n8n__n8n_deploy_template()
4. Se criar do zero   → Seguir fluxo abaixo

---

Nodes LangChain

Core

Node	Uso
@n8n/n8n-nodes-langchain.agent	AI Agent principal
@n8n/n8n-nodes-langchain.chainLlm	Chain LLM simples
@n8n/n8n-nodes-langchain.chainRetrievalQa	RAG Q&A

Modelos

Node	Provider
@n8n/n8n-nodes-langchain.lmChatOpenAi	OpenAI GPT
@n8n/n8n-nodes-langchain.lmChatAnthropic	Claude
@n8n/n8n-nodes-langchain.lmChatOllama	Ollama local

Memória

Node	Tipo
@n8n/n8n-nodes-langchain.memoryBufferWindow	Últimas N mensagens
@n8n/n8n-nodes-langchain.memoryPostgresChat	PostgreSQL
@n8n/n8n-nodes-langchain.memoryRedisChat	Redis

Tools

Node	Função
@n8n/n8n-nodes-langchain.toolCalculator	Cálculos
@n8n/n8n-nodes-langchain.toolCode	Executar código
@n8n/n8n-nodes-langchain.toolHttpRequest	Chamar APIs
@n8n/n8n-nodes-langchain.toolWorkflow	Chamar workflows

Vector Stores (RAG)

Node	Sistema
@n8n/n8n-nodes-langchain.vectorStoreSupabase	Supabase
@n8n/n8n-nodes-langchain.vectorStorePinecone	Pinecone
@n8n/n8n-nodes-langchain.vectorStorePgVector	PostgreSQL

---

Chatbot Básico

Estrutura

Webhook/Trigger
  ↓
LLM Chat Model (OpenAI/Claude)
  ↓
Memory (Buffer)
  ↓
Agent
  ↓
Resposta

Implementação

// 1. Modelo LLM
mcp__n8n__get_node({
  nodeType: "@n8n/n8n-nodes-langchain.lmChatOpenAi",
  detail: "standard"
})

// 2. Memória
mcp__n8n__get_node({
  nodeType: "@n8n/n8n-nodes-langchain.memoryBufferWindow",
  detail: "standard"
})

// 3. Agent
mcp__n8n__get_node({
  nodeType: "@n8n/n8n-nodes-langchain.agent",
  detail: "standard"
})

---

RAG (Retrieval Augmented Generation)

Estrutura

Documentos
  ↓
Embeddings
  ↓
Vector Store
  ↓
Query (pergunta)
  ↓
Retriever
  ↓
LLM (resposta contextualizada)

Configuração Vector Store

mcp__n8n__validate_node({
  nodeType: "@n8n/n8n-nodes-langchain.vectorStoreSupabase",
  config: {
    mode: "insert",  // ou "retrieve"
    tableName: "documents",
    queryName: "match_documents"
  },
  mode: "minimal"
})

---

AI Agent com Tools

Estrutura

Input
  ↓
Agent
  ├── Tool: Calculator
  ├── Tool: HTTP Request
  ├── Tool: Code
  └── Tool: Workflow
  ↓
Output

Configuração Agent

mcp__n8n__validate_node({
  nodeType: "@n8n/n8n-nodes-langchain.agent",
  config: {
    agentType: "conversationalAgent",
    systemMessage: "Tu és um assistente prestável...",
    options: {
      returnIntermediateSteps: true
    }
  },
  mode: "minimal"
})

---

Exemplos Práticos

Chatbot de Suporte

Webhook (mensagem cliente)
  ↓
Memory PostgreSQL (histórico)
  ↓
Vector Store (docs suporte)
  ↓
Chain Retrieval QA
  ↓
Webhook Response

Agent CRM

Webhook (comando)
  ↓
Agent
  ├── Tool: Pesquisar clientes
  ├── Tool: Criar lead
  └── Tool: Actualizar tarefa
  ↓
Slack (resultado)

Resumo de Documentos

Webhook (upload PDF)
  ↓
PDF Extract
  ↓
Text Splitter
  ↓
LLM Chain (resumo)
  ↓
Email (enviar resumo)

---

Templates Recomendados

// AI templates
mcp__n8n__search_templates({
  searchMode: "by_task",
  task: "ai_automation"
})

// Por keyword
mcp__n8n__search_templates({
  searchMode: "keyword",
  query: "chatbot langchain openai"
})

---

Credenciais Necessárias

Provider	Credencial	Node
OpenAI	API Key	lmChatOpenAi
Anthropic	API Key	lmChatAnthropic
Supabase	URL + Key	vectorStoreSupabase
Pinecone	API Key	vectorStorePinecone

---

Boas Práticas

Prática	Razão
System prompt claro	Define comportamento
Temperatura baixa (0.1-0.3)	Respostas consistentes
Memória limitada	Performance
Retry on error	Resiliência
Logs de conversas	Debug e melhoria

---

Troubleshooting

Problema	Solução
Respostas inconsistentes	Baixar temperatura
Contexto perdido	Verificar memória
RAG não encontra	Verificar embeddings
Timeout	Aumentar limite
Token limit	Resumir contexto

---

Skill v1.0 | Descomplicar®