Wow! Recently, an Opus 5 game prompt went viral on X!
Millions are watching online. With this cheat code in hand, AI games start to replicate infinitely...
The most impressive part is that a netizen named Anshu, working alone for 24 hours, built a clone of "Outer Wilds" from scratch!
The logic behind the Prompt is also quite simple:
The main Agent breaks down the task, sub-Agents handle the work, plus an independent Agent acts as a supervisor to "PUA" them.
Basically, if the visuals aren't as good as similar games, they get automatically sent back for revisions. The AI only gets to clock out once the quality meets the standard.
The results are fantastic—
The original creator, Matt Shumer, directly crafted a "Call of Duty" Opus 5 version, with all textures, animations, and sounds AI-generated.
And! He didn't use any external assets!
Working Non-stop for 24 Hours
This internet-famous Prompt was officially named "The Challenge Loop" by its author, Matt Shumer:
I need you to create a first-person shooter game with visual quality on par with the latest "Call of Duty" series... You need to iteratively process each content item in a loop, while assigning dedicated sub-Agents to perform visual verification, ensuring the overall visuals meet 3A game standards... The system must use blind side-by-side comparison to judge which work has superior visuals.
In simple terms, the main Agent, after receiving the task objective, will break it down into several parts on its own. Then, it will assign a specialized production Agent for each part.
The first half of the instruction is almost identical to the common multi-Agent game development process. The key is the newly added acceptance phase:
The judge Agent will inspect the results and compare them with a chosen real-world reference. If the generated result is inferior to a real 3A game, it needs continuous revision and optimization.
Of course, it's impossible for it to truly match a 3A masterpiece yet. At this point, the user needs to stop it in time, using this prompt only as an extreme pressure tactic to push Opus 5's autonomous planning and long-range iterative capabilities.
That's exactly how netizen Anshu's 24-hour space exploration game, "The Long Silence", came about:
The specific production process is divided into several stages:
Step 1: Instruct Opus 5 to create a space exploration game using Three.js.
The user only needs to provide basic goals, such as the player being able to walk around and pilot a spaceship in the game, the visuals shouldn't look too plastic, and it should run as stably as possible in a browser at 60 FPS.
Everything else—the game's world view, technical architecture—is left for Opus 5 to decide.
Regarding assets, after installing the Blender MCP, Claude Code will find tools on its own for modeling.
Step 2: Once the first playable version is ready, give Opus 5 a continuous 24-hour goal to comprehensively improve the visual effects.
At this point, multiple sub-Agents start working in parallel to boost efficiency; the judge Agent will compare game screenshots item by item with 3A space games like "Starfield." During this period, Opus 5 is not allowed to secretly lower its judgment standards.
Author Anshu reminds us to also give Opus 5 a carrot appropriately here:
You have the ability to achieve this goal—do your best, don't give up. (doge)
Step 3: Manual optimization + generating skill.
Anshu would check the progress via phone and, when the model spent too much time on planetary scenes, would request it to adjust its priorities.
After stopping the run, he used other Claude instances to fix rendering issues, clean up the code, and complete the deployment.
Finally, he asked Opus 5 to summarize its experience into a skill and put it on GitHub.
According to the open-source repository, Opus 5 even created its own set of acceptance tools during the review process, including 17 mandatory verification steps. The tool scripts would also automatically capture different scene comparisons and frame rate statistics, as well as analyze exposure data.
BTW, the game can now be played directly on the web (https://longsilence.anshu.dev/). I just tried it myself, and my review consists of only three words:
Mind! Blowing! Game!
Netizens Collectively Copy Homework and Build Upon It
From one to ten, ten to a hundred, this prompt directly fueled the Opus 5 game craze.
Open your eyes, and it's time to play!
Netizen Ryan Campbell used this method to iterate on a go-kart game. Opus 5 continuously improved local rendering, mobile performance, controls, and camera. The finished product can also be played directly in a browser.
Netizen Yogi Suria directly used it to create Claudepunk 2077......
The same instructions migrated to GPT-5.6 Sol also work quite well, only slightly inferior to Opus 5~
Clearly, the existence of the judge Agent amplifies the model's utility, allowing different developers to create various types of game prototypes with it.
This also aligns with the ability emphasized by Anthropic (A社) when releasing Opus 5: the model is better suited for long-term, multi-step work, and is more willing to check results, identify errors, and iterate continuously.
From this perspective, the Challenge Loop prompt is more like applying a stricter correction mechanism to Opus 5's existing capabilities.
As for Matt Shumer, the inventor of this method, he is an AI entrepreneur and investor himself.
In February of this year, he went viral in the tech circle with a long article titled "Something Big Is Happening," accumulating over 80 million views.
The article argued that the public's perception of cutting-edge model capabilities severely lags behind the actual experience of AI practitioners.
And this Opus 5 game trend just provided Matt's judgment with additional samples.
Reference Links:
[1]https://x.com/anshuc/status/2081801979131818412
[2]https://x.com/mattshumer_/status/2081830214384886228
[3]https://x.com/mattshumer_/status/2081054356405731740
This article is from the WeChat public account "Qubit," author: Follows Cutting-Edge Tech
