Prompt Larder

Prompt Larder is a deliberately small prompt library. The pitch is in the word small. Most prompt collections compete on volume, listing thousands of untested prompts scraped from wherever, and volume is exactly the wrong metric for something you have to trust.

The stated standard is that every prompt on the shelf was actually run, and that each one is labelled with the model it was run against and the reason it works.

Labelling a prompt with its model is the detail that matters. A prompt tuned on one model frequently degrades on another, and almost no prompt library records which one it was written for.

What It Is Set Up To Cover

A curated prompt collection with provenance attached. Model noted, rationale explained, tested before shelving.

Published in twelve languages, with indexed paths for Spanish, French, German, Japanese, Korean, Chinese, Portuguese, Hebrew, Hindi, Indonesian, and Vietnamese alongside English.

Where It Is Right Now

Homepage, about page and author page are live. The archive is empty at the time of listing, so the shelf has nothing on it yet.

The standard described above is the site's stated policy, not something that can be checked against published prompts today.

Who It Is Aimed At

People who have downloaded a mega-pack of prompts, found most of them useless, and would prefer a short list that works.

Anyone who has had a prompt work well on one model and fall apart on another and wants that variable tracked.