The VIBES Framework · LibVibes

Vision

Decide what you want and who it's for before you open a chat window.

Instruct

Give the AI your colors, your data, your constraints, and where it will live.

Build

Generate in small chunks and iterate conversationally, one feature at a time.

Evaluate

Run it, read it, click everything on a phone and a laptop. Verify links and facts.

Ship

Host on GitHub Pages, embed in Google Sites, or deploy as an Apps Script web app.

What each step really means

VVision. Finish this sentence before you type a word to the AI: “This is for ___ so they can ___.” One audience, one job. A sharp vision is also your finish line — it tells you when you're done.
IInstruct. Whatever the AI doesn't know, it invents — fake hours, generic colors, placeholder books. Feed it the real specifics: your data, your campus colors, and the one constraint that matters most (“no logins,” “one HTML file”).
BBuild. Get the skeleton working first, confirm it runs, then add one feature per turn. Small steps mean that when something breaks, you know exactly which change caused it.
EEvaluate. Looking right is not the same as working. Click every button, test on a real phone, feed it weird input, and verify any facts (citations, Dewey ranges) against a real source.
SShip. A tool nobody can find helps nobody. Put the working result where your people already are — a GitHub Pages link, a Google Sites embed, a QR code at the shelf, a loop on the lobby monitor.

VIBES is a loop, not a staircase. You'll cycle Build → Evaluate → Build several times before you Ship — that's the process working. The AI's first answer is a starting point, never the finish line. Each prompt in the library includes a VIBES block tuned to that specific build.