Selling app & game prompts
Apps & Games is a prompt type for prompts that build a playable game or a working app with a coding agent or a no-code builder. A buyer gets your prompt and your setup instructions, pastes them into the tool you used, and ends up with the thing your playthrough video shows. They do not get the app itself, your source code, or your hosted demo.
This guide covers what is different about listing one. Everything else — creating a seller account, review, pricing and payouts — works as described in Getting started as a seller.
Table of contents
- At a glance
- What buyers get
- Choosing the tool and model
- Category and tags
- Writing the template and instructions
- Examples: the playthrough videos
- Project details
- Review: what we check
- Pricing and promotions
- Where your listing appears
- Common questions
At a glance
| Apps & Games | |
|---|---|
| Tools | Claude Code, Codex, Cursor, Google Antigravity, Lovable, Replit Agent |
| Examples | 1 to 4 playthrough videos, MP4, MOV or WebM, up to 50MB each |
| Variables | Optional, up to 7 |
| Instructions | Required, at least 80 characters |
| Project details | Playable demo link, tech stack, build time — all optional |
| Price | Free, or $2.99 to $19.99, like any prompt |
| Promotion | Not eligible for Spotlight & Featured yet; site-wide promotions and subscriptions apply as normal |
What buyers get
- The full prompt template, exactly as you wrote it.
- Your usage instructions: where to paste it, what to run, what to expect.
- For paid prompts, the same bonus credits as any other prompt. App prompts can't be run on Graded Prompts itself — buyers run them in their own tool and account — so the credits are spent on runnable image and video prompts in their library.
- A perpetual personal and commercial licence to what they build with it.
Before buying, a buyer sees the title, description, playthrough videos, project details, tool and model, category and tags. The template and the instructions unlock on purchase. So the videos and the demo carry the sale, and the listing is honest when both show what the prompt actually produces.
Choosing the tool and model
Pick the tool you built the demo with. A listing is for one tool; if your prompt works in several, list it separately for each one you have tested.
- Claude Code, Codex, Cursor and Google Antigravity also ask which underlying model you ran the agent on. Choose the one you used for the build in your videos, not the one you would recommend.
- Lovable and Replit Agent choose the model themselves, so there is nothing to configure.
If the exact version you used is not in the list, pick the closest one and name the exact version in your instructions.
The tool and model appear on the listing as "Built with", so a buyer on a different plan or tool knows what to expect.
Category and tags
Pick the category by what the result is, not by the tool. Games: Arcade & Casual, Puzzle & Board, Action & Platformers, RPG & Adventure, Strategy & Simulation, Multiplayer & Party, Educational & Kids. Apps: Productivity & Tools, Business & SaaS, E-commerce & Marketplaces, Landing Pages & Portfolios, AI Assistants & Chatbots, Lifestyle & Personal, Developer Tools.
Tags are one to three from the Apps & Games set, and they describe the build rather than the genre: 2D, 3D, Pixel Art, Retro, Browser Game, Single Player, Multiplayer, Mobile-Friendly, Frontend Only, Full-Stack, PWA, Dashboard, MVP, Authentication, Leaderboard. A buyer filtering on Frontend Only wants something they can host as static files, so only tag it when that is true.
Writing the template and instructions
The template is the full build prompt you pasted: the spec, the constraints, the structure. It must be 80 to 30,000 characters in the standard format; the JSON template format is not available for this type.
Fold your follow-ups into the template. Most of the extra prompts in a real build session were discoveries: a bug, a missing feature, a layout fix. Once you know them they belong in the spec, as phases: "Phase 1: scaffold and the core loop. Phase 2: scoring and levels. Phase 3: sound and polish. Stop after each phase and wait for me." A coding agent handles that in one message, the buyer copies one prompt, and the distilled version is what they are paying for, not a transcript of your session. Keep one prompt per listing: the buyer's copy button copies the whole template as a single message, so two prompts in one field get pasted together.
Variables are optional here, up to seven. Use [VARIABLE_NAME] only for
the parts a buyer should change, such as a game title, a colour scheme or a
difficulty setting. If your template has variables, each example will ask for
the values you used; if it has none, the examples are just the videos.
The usage instructions are required and matter more for this type than for any other, because they are the setup guide. Cover:
- Where the prompt goes: an empty folder, a new project, a chat in the builder, and any command to run first.
- What to install or configure before it works, and what the finished result looks like when it does.
- Roughly how long the agent takes, so the buyer doesn't stop it early.
- Only the follow-ups that have to wait for a result ("play it, then send:
…"). Number them, give each a one-line trigger, keep them to two or three,
and write them without variables (say "the game", not
[GAME_TITLE]): buyers copy these by hand from the instructions, with nothing filled in.
Lovable and Replit Agent build best from short incremental messages rather than one long spec, so listings for those tools lean on numbered follow-ups more. The same rule applies: the first message is the template, the rest are numbered in the instructions.
Instructions are 80 to 30,000 characters. A longer, concrete guide sells better than a short one.
Examples: the playthrough videos
Each example is a video of the built result: a playthrough of the game or a walkthrough of the app. One to four per listing.
- Show the result, not the agent. Menus, gameplay, the core flow, what happens on a win or an error. A screen recording of the agent typing is not an example.
- 30 to 90 seconds works best. The limit is 50MB per file, in MP4, MOV or WebM.
- A frame from your first video becomes the card thumbnail automatically as soon as the video is processed, so the listing has one while it is in draft and review. On approval a single video keeps that frame; two or more videos become a 2×2 collage. A custom thumbnail uploaded on the Publish step overrides either.
- Record in landscape if you can. Listing cards crop the thumbnail to a square from the centre, and the homepage Apps & Games row shows the full widescreen frame, so keep the title screen or key visual centred.
Reviewers watch every video and compare it with your demo, so the videos should show the same build the demo link opens.
Project details
These three fields are optional, but a listing with all three converts better. They appear in a "Project details" block on the listing.
Playable demo link. An https:// link where a buyer can play or try the
result: itch.io, Vercel, Netlify, GitHub Pages, Replit or your builder's own
hosting. You host it; we only link to it. Buyers see a "Play the demo" button,
and reviewers open the link before approving to check it matches the videos.
Keep it live for as long as the listing is.
Tech stack. Free text, up to 200 characters, for what the build runs on: "React + Phaser 3", "Unity C#", "Godot 4", "Next.js + Supabase".
Build time. How long it took from your first paste to the result in your video, including your own follow-up prompts and fixes, not just the agent's unattended run time. Pick the closest bucket:
| Bucket | Means |
|---|---|
| Under 15 minutes | One shot, little or no follow-up |
| 15 minutes to 1 hour | One sitting, a few follow-up prompts |
| 1 to 3 hours | An afternoon, real iteration |
| 3 to 8 hours | A full day |
| More than a day | Multi-session project |
Honest buckets help buyers pick the right prompt for the time they have and protect you from refund requests from someone who expected a one-shot build.
Review: what we check
Every App prompt is reviewed by a person before it goes live, like every other prompt. Unlike image and text prompts, reviewers don't rebuild the app themselves; the videos and the demo stand in for that. We check that:
- the videos show a working result that matches the title and description;
- the demo link opens and shows the same build as the videos;
- the instructions are enough for a buyer to reproduce the result;
- the tool and model are the ones the build actually used;
- the template is genuinely yours and complete.
A rejection comes with a reason. Fix the listing and resubmit; the common ones are a demo that doesn't open, videos that show the agent rather than the result, and instructions that skip the setup.
The Seller Agreement's content rules apply as they do to every prompt. Two come up with app and game prompts in particular: no copyrighted characters or trademarked games in the title, description or template ("a runner in the style of" is fine, a named franchise is not), and you must hold the rights to any asset pack, font or code the prompt links to. Also strip API keys, tokens and account details from the template and the instructions before submitting; every buyer gets a copy.
Pricing and promotions
App prompts are priced like any other prompt, free or $2.99 to $19.99, with the same revenue split. Site-wide promotions and seller subscriptions include them unless you opt the prompt out on the Publish step.
They are not eligible for Spotlight & Featured for now; those sections stay image and video only until the type has its own promotion surface.
Where your listing appears
A published App prompt is listed on the Apps & Games page, on its tool's page under Apps & Games (Claude Code, Lovable and so on), on its category page, in keyword search and on your seller profile. The homepage Apps & Games row draws from them once the catalogue holds enough App prompts to fill it. Spotlight & Featured are not open to the type yet.
Common questions
Can I list the same build for two tools? Yes, as two listings, each with videos from that tool.
Do I have to share the source code? No. The product is the prompt and the instructions. Sharing a repository is your choice.
My build took several prompts. Fold them into one phased template where the agent can run them unattended, and keep only the follow-ups that depend on a result as numbered steps in the instructions. If the build needs a project file or a config first, say so in the instructions and keep the template as the one prompt.
Can buyers run it on Graded Prompts? No. They paste it into their own tool and account. Bonus credits from a paid App prompt are spent on runnable prompts instead.
Can an App prompt go in a bundle? Yes. Any published prompt with a price can; free ones cannot, whatever the type.
Can I change the demo link after publishing? Post-publish edits cover the title, description and thumbnail. For a new demo link, email sellers@gradedprompts.com and we will update it for you.