I would not build a prompt library before capturing prompts that already produced publishable work. Libraries die when they are theoretical.
Start from wins
Save prompts tied to article types you ship weekly.
Include context blocks
Voice rules, audience, internal links, fact sources.
Version lightly
Date and owner. Delete stale prompts.
Review monthly
Kill prompts that produce generic drafts.
A prompt library is an ops tool, not a documentation hobby.