Aulafy
Courses/Production agents with LangGraph and n8n/Human approvals and permissions

Human approvals and permissions

Human review is not a failure of automation. It is the safety belt that lets you use agents on real tasks without turning every mistake into an incident.

  • Classify actions by risk before automating them.
  • Create approval checkpoints for high-risk tasks.
  • Apply least-privilege permissions to tools and credentials.

Permission matrix

  • Read: allowed when the data is required.
  • Create draft: allowed with logging.
  • Send or publish: requires approval.
  • Delete, pay, or change permissions: forbidden at first.
TerminalCode
low_risk:
  - resumir
  - clasificar
  - crear borrador
medium_risk:
  - actualizar CRM
  - crear ticket
high_risk:
  - enviar email
  - publicar contenido
  - emitir factura
forbidden:
  - borrar datos
  - cambiar permisos
  - exfiltrar secretos