PROMPT GUIDE
The Replit Agent is a different animal: it proposes a build plan, writes the code, runs it, and self-tests — inside a real cloud dev environment. Prompting it well means thinking like a product owner handing work to a competent developer. Here is how, plus a free generator that writes these prompts for you.
Generate your prompt — free, no sign-up
Start with "show me your build plan before building." You catch wrong assumptions in ten seconds instead of after a full build.
The agent self-checks against your requirements — give it checkable ones. "Login works with email and password; wrong password shows an error" beats "add auth."
Replit has auth and Postgres ready. Say "use the built-in authentication" instead of describing login mechanics — it wires the whole flow, sessions included.
End with "run the app, test against these requirements, and report what passed." A failing self-test is a free bug report before you ever open the preview.
Like every agent, it does best with a coherent chunk — "the booking flow" — not three unrelated screens crammed together.
Add [FEATURE]. First, show me your build plan before writing code. Requirements: 1. [Requirement with a checkable outcome] 2. [Requirement with a checkable outcome] Use the built-in auth/Postgres where applicable. After building: run the app, test against the requirements above, and report which passed.
Usually unstated assumptions — it guessed something you assumed it knew. Ask for the plan first, and give acceptance criteria it can check itself against. Both habits surface wrong assumptions before code is written.
Yes — that is its best trick. Prompt it to run the app, test your requirements, and report results. When something fails, paste the failure back and ask for the fix. The free generator bakes this loop into every prompt.
Implementation detail. Do not prescribe libraries or file structure — the agent knows its environment. Spend your words on outcomes and checkable requirements.