Why do some of these prompts fail with some AI providers?

LLMs implement the Context URL protocol in a soft manner and subject to other filters and policies (such as avoiding prompt injection). Sometimes this can trip up their compliance with the spec, which is understandable.