Inspiration

I’m George Manty, and I enjoy creating prompts for AI to get it to do what I want. When I saw that Bolt.new was having a one shot prompt bonus competition in it's hackathon, I brainstormed some ideas for that. My first ideas were game related, because I enjoy playing games. That lead me to think about meta moments in games and entertainment which I really enjoy. Which then lead me to this idea. What if I created a prompt that taught people prompting concepts by going over the techniques it used to create the prompt tutorial. A "meta prompt". I liked the idea so much I just had to do it. That’s where the idea for Meta Prompt was born, a meta-learning tool that reveals the exact prompt that built it.

What it does

Meta Prompt is an interactive tutorial web app built entirely from a single one-shot prompt in Bolt.new. It teaches users the basics of writing good prompts with a special emphasis on using Bolt.new to generate complete tutorial applications. The app displays its own prompt, explains how each section maps to code, and provides a prompt building tool that creates prompts you can cut and paste into Bolt.new to create your own tutorials.

How we built it

The entire project was generated from one carefully crafted prompt inside Bolt.new. The prompt specifies:

  • File structure (≤ 20 files)
  • Next.js 13 App Router architecture
  • Tailwind CSS for styling
  • Interactive components like syntax highlighting, progress tracking, and a live prompt builder
  • Exact hero text and UI requirements
  • Guardrails for file limits and fallback strategies

I refined the prompt over multiple iterations to fit Bolt.new’s context window while delivering a rich, fully interactive meta-tutorial.

Challenges we ran into

  • Context Limit: Keeping the prompt concise enough to fit within Bolt.new’s one-shot limits.
  • One-shot Constraints: Ensuring all required features and UI elements could be generated without relying on follow-up prompts.
  • Precision: Avoiding vague instructions in the prompt that might lead the AI to produce unintended code or extra files.
  • Compatibility: Tailoring the prompt to work with Bolt.new’s WebContainers and avoiding unsupported packages and features that might be too complicated to get working in one shot.

Accomplishments that we're proud of

  • Successfully building a self-revealing meta tutorial in a single Bolt.new prompt.
  • Teaching prompt engineering through an app that demonstrates its own creation process.
  • Staying within the strict one-shot constraints while achieving interactive functionality.
  • Highlighting the power of specificity and guardrails in prompt writing.

What we learned

  • The biggest thing I learned was that it's a much better to tell AI what you want than what you don't want. Anytime I put something into the prompt about what I didn't want, it was less successful than being specific about what I did want. I added that into the final prompt as one of the important prompting techniques to keep in mind.
  • Bolt.new’s one-shot capabilities are impressive.
  • Building educational tools with Bolt.new is actually pretty easy once you learn how to properly format a prompt for Bolt.new.

What's next for Meta Prompt

  • Leave it online for people to learn from and use
  • Try creating better meta prompts once context limits increase in Bolt.new to see what is possible
  • Explore using the site as a teaching resource in workshops or online courses for prompt engineering and AI development.

Built With

  • next.js
Share this project:

Updates