Output Parser Node Gets Overidden by Phantom Prompt

Describe the problem/error/question

I have a simple workflow (snippet of a larger automation) that takes in info from a user form to create a script for a subject video, e.g. “The World of Tiger Sharks.” The first couple of nodes correctly capture the user data. But when the data gets to the 4th node, the Script Writer node, something weird happens. The data is correctly captured and it looks like it sends the correct output about Tiger Sharks. But then a SECOND phantom prompt fires off, which overrides the first prompt. It seems like this second prompt is some hidden system prompt or something and it ends up changing the output from the intended subject of Tiger Sharks to something random like How to Create a Perfect YouTube Script Fast!

Here are pics of that node’s logs, where you can see the TWO instances of the LLM:



What is the error message (if any)?

No error message. The workflow fires off without error BUT my prompt gets overidden by a phantom prompt.

Please share your workflow

Share the output returned by the last node

Information on your n8n setup

n8n version: 1.92.2
Database (default: SQLite): SQLite
n8n EXECUTIONS_PROCESS: own
Running n8n via: Docker (self-hosted)
Operating system: Debian GNU/Linux 12 (bookworm) 6.1.0-34-amd64 (Docker host)

Anyone have any tips to help solve this problem?