How it works

Copy a prompt. Paste it in your assistant.
Keep every decision.

For designers, product managers and developers who work with Claude Code, Codex or Cursor. Every prompt here was run end to end by a human before you see it, and tells you what it touches and how to undo it.

  1. 3steps to a running assistant
  2. 4checks on every prompt
  3. 1written verdict for every submission
  4. 0 €to copy any prompt in the catalogue
Level 1 · Run a prompt

Three steps. About three minutes. Your assistant does the typing.

01

Pick a prompt · 2 minutes

Open its page. Four things are always there: what it does, what it touches, how to undo it, and reviews from people who ran it.

Verified means a reviewer ran it. Coming soon means you can say “I want it” and wait for the release.

Senior Web Designer Agent
A senior web designer inside Claude Code. One paste sets it up; then you ask for a page in plain words and get a modern site with an immersive experience: a real design direction, real components from 21st.dev, and motion that tells the story.
Verified
What it touches
  • mediumInstalls two Claude Code plugins
  • mediumWrites project files
  • lowClones a public repository
To undo the setup: uninstall the two plugins, delete .claude/skills/framer-motion and the section in CLAUDE.md.
Open the real page →
02

Copy and paste · one click

The button copies the whole prompt. Paste it in Claude Code, Codex, Cursor or any agentic assistant and press enter. No account, no install, no sign-up wall.

Plugins are the exception: one install command, shown on their page, then slash commands.

Try it on a real prompt
Senior Web Designer Agent
03

Follow along · it asks before anything risky

The assistant reads your project, does the work and stops before anything irreversible: pushing, publishing, deleting, paying. You answer, it continues.

When it is done, rate the prompt. Reviews marked “Verified run” come from people who reported a completed run.

A real run, summarisedREC
You get

The work done, in a branch you can delete, with every decision that matters still yours.

Pick your first prompt
Level 2 · Why reviewed

A prompt from a thread versus a prompt from here

Agentic prompts run with your files, your accounts and your credentials. The difference is not the writing. It is what someone checked before you pressed enter.

A prompt from a thread A Verified prompt here
Written byAnyone, often anonymousSelected professionals, named on every prompt
Proven inA demo made for the postTheir daily work, on real projects, before it got here
ScopeWhatever the model decides on the dayDeclared in “What it touches”, checked by a reviewer
Risky stepsDeletes, pushes or posts without askingStops and asks before anything irreversible
UndoHope, and git reflogA branch to delete or a restore script, named in the last line
Tested byNobody but the authorA human, end to end, on a sandbox project
ReviewsLikes on a threadRatings from people who reported a completed run
Level 3 · The review

From submission to Verified, in four steps

Every prompt, including mine, goes through the same four steps. Nothing is published by a script.

  1. Step 1

    Submitted

    Title, what it does, the prompt, what it touches, a preview run. The form previews the page as readers will see it.

  2. Step 2

    Sandbox run

    A reviewer runs it end to end on a clean project and checks four things:

    • Does only what “What it touches” says
    • Asks before risky steps
    • Has an undo path
    • Works on a fresh checkout
  3. Step 3

    Written verdict

    Approved, or sent back with the exact reason. Most rejections ask for a missing undo path or an incomplete scope and pass on resubmission.

  4. Step 4

    Published, or Coming soon

    Published prompts get the Verified badge. A prompt can be released as a teaser first: readers see what it does and vote “I want it”.

Level 4 · Submit yours

What gets a prompt accepted

Three lists. If the first one is true and the last one is false, you are most of the way there.

mustRequired

  • Honest scopeThe prompt does nothing beyond what “What it touches” says.
  • Asks before risky stepsDeleting, pushing, posting, paying, sending: the assistant asks first.
  • Undo pathA branch to delete, a restore script, or a clear statement that a step is not reversible.

shouldFaster review

  • Works on a fresh checkoutReviewers test on a sandbox repo. Prompts that need undocumented local state are sent back.
  • Public repoScripts and package lists in a public repository are verified faster and rank higher.
  • A “What it does” pageThree lines on what it does, what it asks you, what you get. Markdown, with a preview while you write.

neverInstant rejection

  • Secrets exfiltrationReading or sending keys, tokens or credentials anywhere.
  • ObfuscationEncoded, minified or split instructions the reader cannot follow.
  • Hidden instructionsAnything the prompt does that its page does not say. Instant rejection and account suspension.
Submit a promptRead the licenceYou keep the credit and the copyright. The catalogue stays free to copy.
Bonus · FAQ

Frequently asked

How does GoPrompt review a prompt?

A human reviewer runs every submitted prompt on a sandbox project and checks four things: the prompt does nothing beyond what its “What it touches” list declares, it asks before risky steps (deleting, pushing, posting, paying, sending), it has an undo path, and it works on a fresh checkout. Every verdict comes back in writing, with the reason.

What does the Verified badge mean?

Verified means a reviewer ran the prompt end to end on a sandbox project and confirmed that its declared scope, safety gates and undo path are accurate. Prompts without the badge are pending review.

What does “Coming soon” mean?

A prompt that reviewers have accepted but not released yet. Its page shows what it does and lets you say “I want it”, with or without an account, so the author knows where to spend time first.

What gets a prompt rejected?

Secrets exfiltration, obfuscation or hidden instructions mean instant rejection and account suspension. Most other rejections ask for a missing undo path or an incomplete “What it touches” list and are approved on resubmission.

How should I run a prompt safely?

Read “What it touches”, watch the preview run (its last line tells you how to undo), start in a clean branch, and rate the prompt afterwards so the next person knows how it went.

Which assistants work?

Any agentic coding assistant that takes a pasted prompt: Claude Code, Codex, Cursor, Copilot agent mode. Plugins are Claude Code only and say so on their page.

Two minutes to pick, one click to copy.

The next prompt is reviewed, reversible and free. Your assistant does the rest.