Your team has a prompt library. It's a Notion page, or a Slack thread, or three people's private notes. Nobody owns it, nothing in it is versioned, half of it was written for a model that has since been replaced, and no one can tell you which prompts still work.

Twenty minutes on treating prompts as a maintained asset instead of accumulated debris.

Why prompt libraries decay faster than code — model updates, silent behaviour drift, and the fact that a broken prompt fails quietly rather than loudly
Curation over collection: I started from a public library of 18,140 prompts and cut it to roughly forty that a team would actually use. Volume is the enemy
Minimum viable structure: ownership, a last-verified date, the model it was written against, and an example of correct output
How to test a prompt — the same instinct as a regression test, applied to something non-deterministic
When to delete rather than fix, which is more often than anyone likes

Short, practical, and immediately applicable to any team that has more than one person using AI tools.