rag-implementation

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). O fluxo runtime descrito no Workflow 1 (Make.com Webhook “WebHook Custom” → endpoint /query_rag que retorna “context” → prompt do Claude usando pergunta do módulo 1 e chunks recuperados do ChromaDB) ingere texto livre do usuário (“question”) não confiável diretamente na consulta ao RAG e no prompt.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). Make.com HTTP module example calls the external endpoint http://seu_servidor_rag:5000/query_rag at runtime to fetch retrieved chunks that are injected into the LLM prompt (i.e., external content directly controls the prompt/context).