itch.io Community » Game Development » General Development

Claude (Opus 4.7) is really capable of managing the whole project and it surprised me

A topic by Silver3310 created 1 day ago Views: 124 Replies: 4

Viewing posts 1 to 3

Hey folks :)

Recently I got excited by the idea to revive one of my projects with AI and the result was honestly quite fast and successful so I decided to share my experience here with you in case it can help anyone as well 🙌

Some context

So recently I got the idea to "revive" one of my old projects, this project had complex architecture (client/server) and written in Python/C# with no AI in 2023, to revive it was pretty painful due to outdated Python/Unity versions and I was kind of like... hmm, why not to try to use the modern AI tools to rebuild this game to make it much simpler to run at anytime (like a standalone browser game)? and it worked 👀

Cost and usage limits

Claude subscription costs 20 dollars minimum for an individual, it gives access to the Opus 4.7 but with significant limitations, basically 2 big queries to this model will hit your limit for 5 the next hours, and if you do it 5-6 times you will hit the weekly limit. It's definitely a significant constraint but fortunately during one such query the Opus 4.7 model can define the whole architecture, fix 5+ bugs, add 5+ features, so it was worth it at least for me.

Prompt Engineering

The prompts were done in the following way for my project:

"""

You're a professional game developer specialized in browser games and technologies like HTML, CSS and JavaScript. Please read through the game docs, architecture and implement the following features/bug fixes:

Please make sure your solution aligns with the current architecture, improves and introduces more best practices if relevant as well as documentation

"""

Claude could work on the request for 10-15 minutes but the end result was always a ready-to-use solution with surprisingly no bugs

Repository / PRs

You can check the repository used to build the project with AI at the following link: https://github.com/Silver3310/push-them-out-2

Here you check out example PRs: https://github.com/Silver3310/push-them-out-2/pull/6, https://github.com/Silver3310/push-them-out-2/pull/8, https://github.com/Silver3310/push-them-out-2/pull/11.

Hope this info can help someone with their project development/improvement/migration! :)

Like Reply

Steve Mitchell1 day ago

Thanks for sharing your detailed experience, Silver3310. It's particularly impressive to hear about Claude Opus 4.7's ability to define 'the whole architecture' and deliver 'ready-to-use solutions with surprisingly no bugs' from a single comprehensive prompt. This suggests a strong understanding of interdependencies and architectural patterns. Could you elaborate on how you structured the initial input or documentation provided to Claude for it to grasp the original complex architecture effectively before starting the rebuild?

Like Reply

Silver33101 day ago

Sure, this was my first prompt to generate the architecture:

"""

Please create the professional structure for the PushThemOut2 game. It will be an HTML/CSS/JavaScript game (= it can be run on any browser) and I'd like to make sure it has clear division between the logic, player classes, object classes, enemy classes, event handling, sprites folder (make sure it's possible to customize the sprites in the future and they are auto-scaled), sounds/music folder (make sure it's possible to customize the sounds/music in the future, ideally add some sound engine elements so it's possible to change the volume of the sound based on the player distance from the object).

Some context: If it helps I want to recreate the following game of mine (PushThemOut: online co-op, see the screenshot) that was written in Unity (client) + Python (server), and now I want it to be in pure HTML/CSS/JavaScript and make it standalone (no division between client and the server). The purpose of the game is to control a ball on a colorful galaxy pool table and the idea is to push your enemies out to survive.

"""

Basically I highlighted the importance of an easily extended architecture so that it's easier for Claude (or any other model) itself to add new changes / extend the functionality, the output of this prompt can be seen under the following commit: https://github.com/Silver3310/push-them-out-2/commit/0499a5227f826e763496984aa0f...

Like Reply

Deleted post1 day ago

Deleted 1 day ago

Silver33101 day ago

Interesting outcome. What began as the application of AI in game development has now expanded into AI-driven engagement within the comment layer itself.

We are effectively witnessing a full-stack integration of artificial intelligence: content generation, user interaction, and feedback synthesis operating within a single system.

In simpler terms, AI is no longer just building the experience—it’s also reacting to it.

Like Reply

Steve Mitchell1 day ago (1 edit)

This isn't just about distinct AI modules, but the intricate data connections and feedback loops between generative content, real-time user interaction analysis, and subsequent content or system adjustments. The challenge lies in maintaining coherence and preventing drift within these autonomous cycles, ensuring the AI's reactive capabilities align with overall project goals without constant human oversight.

Like Reply