About

Most advice about prompting is vague, unfalsifiable, or copied from the same listicle. And most tools that check your prompt want an account, an email address, and a copy of everything you type.

This site is an attempt at the opposite of both.

What it does

The checker runs nine specific checks against a prompt and tells you which structural mistakes it contains — a missing output format, a dangling pronoun, data pasted without delimiters, phrasing that will make a model overtrigger. It runs entirely in your browser, so it is free, instant, and needs no account and no key.

The builder goes the other way: answer a few questions and it assembles a prompt that passes all nine checks. That also runs in your browser, at no cost, because assembling a prompt from stated requirements needs no AI at all.

The guides explain each check, and — this is the part we care about — every claim in them is demonstrated rather than asserted. When a guide says a prompt fails the delimiter check, the real checker runs on that real prompt in front of you and fails it. The examples cannot drift out of date, because they are not screenshots.

The prompt library is a set of templates that all pass every check, each one explained: not just the prompt, but why every section of it is there.

What makes it different

It is a linter, not an oracle. It does not invent a score out of 100 from heuristics, because heuristics are not entitled to that judgement. It tells you which specific practices are present and which are missing, and lets you decide whether each matters for your task. A tool that cries wolf gets ignored — and then it catches nothing.

We never receive your prompt. The checks run on your machine. There is no server of ours that could receive it. (There is third-party advertising code on the page, and the privacy page is candid about what that does and does not mean.)

It works with whatever model you already pay for. Claude, ChatGPT, Gemini. The full AI review runs on your own chat subscription, so we are not in the business of reselling you tokens — and you do not need an API key to get it.

How it stays free

Advertising. That is the whole answer.

The site costs almost nothing to run — the checks execute on your machine, and the full review spends the subscription you already have rather than ours. Ads cover the domain and the hosting. There is no premium tier, no upsell, and no plan to add one.

Who made it

A small independent project, built because the existing tools all wanted an account and none of them would say why a prompt was bad in a way you could check.

Found a bug, disagree with one of the checks, or think a rule fires when it should not? That last one is genuinely valuable — a false positive is the worst failure this tool can have. Tell us.