About PromptAttic

PromptAttic is an independent publication. We publish prompt recipes. Terse, opinionated, model-tagged. Each one is built for a single job: shipping a production AI feature. Not prompt theory. Not chat transcripts. Recipes you copy, read the receipt on, and put behind a real product.

Every recipe carries a model tag, GPT, Claude, Gemini, or open-source, because a prompt that wins on one model often loses on another. We say which model we wrote it for, what it costs, where it breaks, and what to swap when you move it. The voice is deliberately flat: senior engineer commit messages turned into prose. No throat-clearing, no hype, no filler.

The editorial focus is narrow on purpose. A recipe earns a slot only if it ships something. If it cannot survive contact with a real workload, latency, token budget, eval gates, it does not run. We would rather publish ten recipes that hold up than a hundred that read well and fail in staging. Each one names the model version it was tested against and the date it was run, so you can judge how stale it has gone before you trust it.

We also document failure modes. The prompt that looked airtight in the playground and fell apart under concurrency. The schema the model honored ninety-eight times and quietly broke on the ninety-ninth. The cache that saved nothing because the prefix changed on every call. A recipe is only finished when we know where it cracks, and we publish that part too.

Editorial is led by Sam Q. and Roo Iyer, who write, test, and tag the recipes against live models before anything is published. Bylines are named on purpose. You should know who stood behind a prompt before you put it in production.

Everything here is meant to be copied. Take any prompt and run it on whatever stack you like. We do not gate recipes, and we do not ask you to sign up to read one. The whole point is that the recipe leaves with you.

Have a prompt that ships, or a failure mode worth documenting? Use our contact form. We read everything.