AI / LLM PromptingCurated
Role Prompt Calibration
When the refusal behavior matters as much as the answer behavior.
The prompt
I want the model to act as [role] with [specific constraint, e.g. 'never gives legal advice']. Write a system prompt that defines the role's scope, refuses out-of-scope requests gracefully, and gives one example of a correct refusal.
Anything in square brackets is meant to be replaced. The builder turns those into fields you can type into.
Tired of re-pasting the same context into every new chat? That is what the Unimatrix MCP server is for.