Claude game-modding toolkit universal-modder nears 2,600 stars in four days, driven by a user-written log that includes failures

About 2,600 stars in four days for a toolkit that hands game modding to Claude

One repository on GitHub stood out this week. universal-modder is a set of skills and tools that lets coding agents such as Claude build mods for PC games. Mods are add-on data or code that change how a game works. The project was created by Rehan Sheikh (rehan-remade on GitHub), who has published logs of agents working on Terraria, Age of Empires II and a GTA V × Minecraft crossover.

The key point is that the agent can handle the whole process. It first identifies which engine a game runs on. It then analyzes the code, generates assets such as images, 3D models and audio through fal, tests the mod inside the running game and finally produces a showcase video. The toolkit includes guides for 12 engines and works with Claude Code as well as Codex, Cursor, Gemini CLI, Copilot and OpenCode. It is packaged as Agent Skills, a format for giving agents procedures and know-how, so in effect it hands an agent a modder's entire workflow.

The growth figures are clear. The repository was created on September 30. A write-up published that day put its star count at roughly 635 to 648, and by October 3–4 it had reached 2,634. It has 230 forks. It also appeared on GitHub's daily trending list on October 3 with 2,248 stars and has since passed 2.6k. All of that came about four days after the repository was created.

The growth came from users adding "here's what happened with my game"

Projects like this often go viral on a flashy demo video and then fade. Visuals like a GTA V–Minecraft crossover mod do grab attention. But the more interesting story was in the Issues tab, where users file reports and requests.

About 30 issues and pull requests arrived right after launch. Many of them were knowledge notes with "kb:" in the title. KB stands for knowledge base: a record of "here's what I tried with this game, and here's what happened." The issue list shows more than 10 KB case reports filed in just three days, from October 1 to 3.

The reports cover a wide range: Victoria 2, Dragon's Dogma 2, Darkest Dungeon, a total conversion that brings Elden Ring into CS2, Borderlands 2, a port of WoW gear models to Minecraft, and a port of Halo 3 weapons and vehicles to Minecraft. In Half Sword, one user created a weapon with 9.9 times the normal attack power. The games span many genres and eras, and each user is contributing their own results.

This matters because the toolkit has 12 guides, and there are far more games than that. No single author can research them all. When users add "here's what happened with my game," the next person (and their agent) to work on that game has an easier time. More than the star count, this steady stream of contributions is what is building out the tool's actual content.

Failures were recorded too

The most encouraging part is that the contributed knowledge is not limited to success stories. Accounts of what did not work are mixed in.

Issue #2 is one example. It describes an attempt to use ReShade, a widely used tool for adding post-processing effects to game graphics, in Black Myth: Wukong. Injection, meaning hooking into the running game to insert processing, worked. But the depth buffer, which records how far each point on screen is from the camera, came back as a flat 1.0 everywhere, making it unusable. On October 1, this was classified as a "design limitation." Recording what failed and why is some of the most useful information there is, because it keeps the next person from hitting the same wall.

Bugs in the tool itself were also found early. They included a failure when backing up files dated before 1980 on Windows, a UTF-8 encoding problem, and a path-separator problem (Windows uses a different character than other operating systems to join folder names). These were reported and fixed until all 27 tests passed. Within days of launch, a loop of users reporting bugs and the author fixing them was already working.

Some problems remain open. Issue #13, filed on October 3, reports that installing the Gemini CLI extension failed with a "gemini-extension.json not found" error. Issue #5, filed on October 2, criticized the quality of the implementation itself. Issue #16 asks for Mac support and #6 for mobile (APK) support; neither is supported yet. The tool lists six supported agents, but they may not all work equally smoothly, and that is worth keeping in mind.

Verifying in the running game is the core feature, but watch costs and scope

Media coverage has been positive. Write-ups including one from bitcoinversus.tech reportedly called the step where the agent launches the game, takes screenshots and checks that the mod actually works the biggest advance. The reasoning is simple: AI-written code often looks plausible but fails when run. An agent that runs, captures and verifies its work is far more trustworthy than one that turns out large amounts of code that only looks right.

Cost is a different matter. According to VGTimes' beginner's guide, asset generation through fal is billed per use, and the guide states explicitly that 3D and video are especially expensive. It recommends checking prices first with "um fal price" before running anything. The recommended showcase video length is reportedly 20 to 45 seconds, the plugin is at v0.2.0, and it requires at least 4GB of RAM.

The guide suggests Terraria, Stardew Valley, Minecraft and Unity-based games as starting points. All of them have well-established mod loaders, the frameworks that load mods into a game. It is safer to start where others have already built up knowledge than to begin with an obscure big-budget title.

The tool's scope is also clearly limited. It supports only offline single-player games. It is designed to refuse requests to bypass anti-cheat systems in online games and to refuse multiplayer cheats. Running it requires a fal API key, Python 3.10 or later and FFmpeg, the standard tool for converting video and audio. Notably, the coverage itself points out that "not every viral crossover mod came from this tool." Readers should not assume every crossover mod circulating on social media was made with it.

Our take: a working video is not the same as a playable mod, which is why the log matters

After looking into the project, one conclusion stands out: producing a showcase video and producing a mod that is actually playable are two different things. The agent will produce a 20- to 45-second video at the end, but that only records the moment the mod worked. A video alone cannot show whether it holds up over long play sessions. We also searched YouTube for demo videos posted within three days of launch and found none, so no video is embedded in this article.

That is why the KB entries piling up in the Issues tab are so valuable. When a dead end like the Black Myth: Wukong depth problem is on record, the next user's agent can avoid it from the start. Successes like the 9.9x weapon in Half Sword sit in the same log as unglamorous failures like the Windows date bug. Behind the figure of about 2,600 stars in four days, what is really growing may be this shared, collectively written notebook. That is the main takeaway.

It is worth remembering that the project is only four days old. Open issues include criticism of its code quality (#5), a failed Gemini CLI install (#13) and the lack of Mac and mobile support (#16, #6). Anyone trying it should pick a game with a mature mod loader, check asset generation prices first and start small. Then they can add their results to the KB, including what did not work. That may be the best way to help the tool grow.