There are two types of prompt libraries.
The first is a Notion page someone wrote at 11pm titled “awesome ChatGPT prompts.” Sixty templates that all start with “You are an expert.” Nobody uses it after week two.
The second is the prompt you typed in a hurry six weeks ago that produced the document your team actually shipped. You can't find it now. You'd give a lot to find it now.
The library that matters writes itself
Shelvia treats every imported chat as a candidate source for prompts. When a prompt appears to have produced useful output , based on length, structure, your own reactions in the chat, and whether anything saved-as-decision came out of it. Shelvia proposes saving it.
You don't collect prompts. You confirm the ones that earned a spot.
Tied to the work, not floating free
A saved prompt in Shelvia carries its provenance: which import it came from, which AI tool ran it, what came out, which project it helped. When you reuse the prompt, you can pull up the original output to remember the shape of the answer.
Six months from now, when you're onboarding a new collaborator and need to show them how the team operates, you can hand them a small library of prompts that have been pressure-tested by your actual work. Not a starter pack. A practice.
Skills are the next step up
When you find yourself reaching for the same prompt with small variations. different inputs, same structure. Shelvia can promote it to a Skill: a parameterized template you can rerun on new inputs. The prompt graduates from a saved string to a small, reusable workflow.
That graduation is where personal AI use becomes professional AI use.
