The Prompt Library: 120 Prompts That Hold Up at Work

As of July 2026, the prompt library is live: 120 copy-paste prompts across ten professions, from research and engineering to sales, education, and personal productivity. Twelve per category, every one written to be used as-is.

What a library entry actually contains

Most prompt collections are one-line wishes ("write a great blog post"). Each entry here is a complete working prompt: a numbered method, real constraints, and an explicit output shape, plus three things around it:

• Variables: {{placeholders}} with a description and a concrete example for each, so you know exactly what to fill in.
• Why it works: the mechanism behind the prompt (the structure it forces, the failure it prevents), so you can adapt it without breaking it.
• A model suggestion where the task rewards a specific strength: a coding model for SQL generation, a fast model for transcript summaries.

The prompts encode working discipline, not magic words: the meeting summarizer refuses to invent deadlines nobody said, the market sizing outline labels every unverified number, the SQL prompt restates your request before writing a line.

Run it in one click

Every prompt page has a Run in idapt button. It opens a new chat with the prompt prefilled and stops there: nothing sends until you press send, so you can fill the variables first. From the same composer you can pick any of 200+ models, or run the prompt across several side by side (here is how comparing models on one prompt works).

Receipts, not promises

Flagship prompts carry a receipts band linking to Benchmarks, where a filled-in version of the same prompt has been run on real models in production, with the results published. When a prompt has no published outputs yet, the band simply does not appear. You see what the prompt produces before you spend a token on it.

Where this fits

The library joins the free tools on the public site: calculators and checkers for the numeric questions, prompts for the working ones. Both are free, no signup needed to read and copy everything.

Good to know

• Prompts are capped at 1,500 characters so the one-click run always works; longer workflows belong to agents, not prompts.
• The library is English-only for now.
• Every entry carries a visible last-reviewed date and joins the quarterly review sweep.

Start with your profession's category at /prompts, copy the first prompt that matches a task you already have this week, and judge it on the output.