Where I used it
In my 2025 Self-healing project, n8n sends formatted error information to Claude, extracts the response, and saves a cause analysis and suggested fix prompt in Notion. The workflow supports troubleshooting; it does not show an automatic repair or a verified recovery rate.
The role I would give it
I would use Claude to question my assumptions, review a model explanation, and help inspect code. I want it to find where my reasoning is weak, not simply agree with my conclusion.
How I would guide it
I would give it the objective, source documents, assumptions, and calculations separately. I would ask which conclusions the evidence supports, which need more data, and which change under another reasonable assumption.
What I would check
I would open the cited pages and reproduce the calculations. When two AI tools agree, I would still check the source: agreement between models is not independent proof.
Keep my own voice
I would ask for plain language, then rewrite the final note myself. If I cannot explain a sentence or calculation, it is not ready to publish.