N8n-as-code: Stop n8n AI hallucinations with TS decorators

I spent the last few weeks building an abstraction layer for n8n to fix a specific problem: LLM hallucinations in AI IDEs.

When you ask Cursor or Windsurf to build an n8n workflow, it often breaks the JSON schema. By bringing the n8n ontology into TypeScript with decorators, we provide the AI with the strict context it needs to be 10x more reliable.

It's a pure GitOps approach to automation. Curious to see if others here are moving away from 'No-Code' UI towards 'As-Code' for better AI reliability.