How we review kits

"Reviewed by Robot Academy" means a person ran the kit. Here's exactly what that involves, so creators know what to expect and buyers know what the badge is worth.

1. A clean session

We open the agent the kit is built for (Claude Code, Codex or Cursor) in an empty folder, with no project context, and paste the builder prompt as one message. We don't help it. If it asks a clarifying question we answer "use your judgement and keep it self-contained", which is exactly what the lesson tells buyers to say.

2. The demo, side by side

We open the kit's demo next to what the agent built and compare. The question is not whether the build is impressive. It's whether it is the same thing: the features the demo shows, the look, the behaviour. Small differences are fine; agents don't produce identical output twice. A feature in the demo that the prompt doesn't reliably produce is not fine.

3. The lesson's follow-up prompts

Steps 3 and 4 of every lesson ask the buyer to paste a follow-up prompt: change the data block, then add their own touch. We run both. They should change only what they say they change. "Edit only the WAVES constant" has to mean that.

4. The things that get a no

Some kits are declined before we run anything:

• Bare prompts with no demo, live or video.
• Anything that asks buyers to install prerequisites or run scripts outside the agent. That's how malicious "skills" have spread on other platforms, and we won't list it.
• AGENTS.md or config files on their own.
• Reskins of another creator's kit.
• Skills that phone home or read files outside the project.

5. Notes, or a yes

Most kits get one round of notes. The common ones: the demo shows more than the prompt builds; the data block isn't named, so the lesson can't point at it; the first run needs an API key the kit didn't mention. Creators fix, resubmit, and usually go live on the second pass.

The whole thing takes two to four days. We answer every application, including the ones we decline.