Shiplog

Prompt coach · browser only

Find out if your prompt
is worth sending.

Paste a prompt and see what a model would make of it: a score, what that score is measured against, and the single change worth making first. A worked example is already graded below, so you can read a real result before you paste anything.

Already have an export of your own history? Read your whole history instead — the same rubric across weeks of your prompts, naming the single habit worth changing first. It is read in your browser too.

Grade a prompt

Would a model answer this prompt well?

Paste a prompt, or the few turns around it, and grade it against the same rubric AI FinOps grades an imported corpus with.

Start here

Grade one prompt against a bundled rubric, in this tab, with no sign-in and nothing uploaded.

What you get, what it is measured against, and what to do first appear here in a moment.

A worked example, already graded

This result grades a bundled synthetic example written for this page. It is not your text, not a customer’s, and not anyone’s real prompt. Grade your own prompt below and this example steps aside for your result.

The graded example appears here in a moment.

You can revise and re-grade in this session. Grade once, edit, and grade again to see the previous and revised scores, what moved, and one next step. Both grades use the bundled rubric in this tab; no prompt is sent to a model. A text-free result envelope from the previous grade is kept in memory only until you clear the panel or close the tab.

Label turns “User:” and “Assistant:” to grade an exchange, or paste one prompt on its own.

Model fit is one of three axes. Left unspecified, it abstains rather than assuming a tier.

Nothing you paste leaves this tab. Your text is graded in this browser: no sign-in, no upload, no storage, and no request to a model.

What this reads, what it returns, and what stays out

Coaching is computed in this browser from bundled static client-side code. No request is sent for coaching, no persistence is implemented, and no credential, customer record, or live integration is read.

The worked example, the field list, and the four states this workflow can return are rendered here from the coaching contract when the page’s scripts load.

Result presentation specimen — every state, drawn

Seven states of one reading order: the answer first, what it is measured against second, exactly one move third, and the evidence behind a disclosure. Every case below is built from the bundled synthetic samples — nothing you type is read, rendered, or measured here.

The states are rendered here from the coaching contract when the page’s scripts load.

Also on this site · separate demo

Grading a whole organization’s prompts?

This page grades one text. AI FinOps scores a provider export for intent, efficiency, and model fit, then attributes the spend to the org chart. Your files do not leave that tab either.

Open AI FinOps