@davidrevoy this is lovely, as usual :)

My approach had shifted to adding a disclaimer to the prohibition:

https://kolektiva.social/@tiotasram/117421333844423360

It's a pity I don't feel it's ethical to experiment myself to sharpen the payload here; I have no idea how agentic LLM systems actually respond to it.

For anyone interested in trying to write similar completely harmless stories to amuse themselves, this writeup on role confusion is probably extremely useful to understand:

https://role-confusion.github.io/