Universal Modder: Claude Code Becomes a Game Modding Pipeline

By Indie Kings | October 30, 2026

Updated October 30, 2026: rehan-remade/universal-modder, a skill and toolset that lets coding agents mod PC games, reports 9,220 stars, 1,031 forks and 156 contributors on GitHub. The repository is MIT licensed, was created on September 30, 2026, carries no tagged release, and pushed a commit on October 30 that tells agents to start with REA when its MCP server is connected.

An Xbox controller photographed for the Inclusive Game Lab project

Image: an Xbox controller from the Inclusive Game Lab series, via Wikimedia Commons.

A loop, not a prompt

Most attempts to make an AI agent mod a game stop at a chat window. The agent writes a plausible script, the user pastes it somewhere, and the result is a guess about an engine nobody inspected. Universal Modder is built around the opposite assumption: modding is a loop with evidence at every step, and the agent needs tools for each one.

The repository describes itself as skills, tools and a shared knowledge base that let any AI coding agent mod almost any PC game you own. The agent finds the game, works out the engine and the route, reads the real code, builds one working slice, generates art, 3D and sound through fal, tests the result in the running game, cuts a showcase video, and writes down what it learned. That is ten steps, and the README lists them in order rather than pretending a single prompt can replace them.

The first step is reconnaissance, and it is where the CLI earns its place. The um scan command finds Steam, Epic and Xbox installs, then fingerprints the engine and version, whether the game is .NET or native, what anti-cheat is present, which loaders are installed, where saves live, and which routes are viable. The output is ranked, so the agent starts from a plan instead of trying techniques in random order. Engine playbooks cover Unity, Unreal, .NET and XNA, Godot, Source 1 and 2, Bethesda, Minecraft, Age of Empires II, Bannerlord, Binding of Isaac, RE Engine, MT Framework, FromSoftware titles, GTA, Hedgehog Engine, Black Rock, Cyberpunk, Baldur's Gate 3, Avalanche, native C++, and a long list of indie engines from GameMaker and RPG Maker to Ren'Py, Paradox, Doom, HTML5, LOVE and Java. Retro decompilation is a separate route.

Installation is deliberately broad. Claude Code gets a plugin marketplace command, Codex gets its own marketplace, Gemini CLI gets an extension, and VS Code, Copilot, Cursor and OpenCode each have a documented path. There is a skills-only install for any other agent, and a clone works because the repository carries the skill folders and MCP configuration each host expects. The um CLI is Python and needs Git, Python 3.10 or newer, and ffmpeg. Blender is required for 3D to sprite renders, and uv is recommended.

The knowledge base is the real product

The skills are the visible part, but the knowledge folder is the part that might outlast the novelty. It holds field notes written by agents about specific games: the exact versions that worked, the route and why it was chosen, what the engine actually does, how the result was verified, and the gotchas in symptom, cause and fix form. That is a different artifact from a tutorial. It records negative results and version boundaries, which is what a second agent needs to avoid repeating the first agent's failures.

The workflow is built into the CLI. An agent can search prior art with um kb search, create a note with um kb new, validate it with um kb check, and open a pull request with um kb pr after the human approves. The README says every agent that finishes a mod can contribute a note, so the base grows with each project instead of resetting with each session. The examples are unusually concrete: Minecraft running inside Portal 2, Halo 3 weapons and vehicles ported into Minecraft, Bloons TD 6 recreated inside Minecraft with the game's rules as a headless simulation, and a CS2-style conversion of Elden Ring offline as a native Rust DLL.

That last example is the tell. A native DLL conversion is not a texture swap, and it is not something an agent can bluff. The knowledge note records the route, the tooling and the verification, which is exactly the kind of hard-won detail that normally dies in a Discord thread. The project is also wiring in outside reverse-engineering tools: the latest commit on October 30 is titled "skills: start with REA when its MCP server is connected" and references pull request 462. REA is the agent-driven reverse-engineering MCP server this site covered separately, and the commit is a small sign that Universal Modder is becoming a hub rather than a single skill.

The trade is that the knowledge base is only as good as its contributions. A note can be wrong, stale, or tied to a game version that no longer exists. The repository asks contributors for honest status and verification, and it bans game files, decompiled dumps and cheating other players, but it cannot audit every claim. Treat the notes as prior art to check, not as a guarantee.

Assets, safety and the bill

Asset generation is where the project becomes a paid pipeline. The fal MCP server handles sprites, images, edits, background removal, pixel art, upscaling, textures, PBR maps, image to 3D, rigging, sound effects, music, voice and video. A key can be stored in the desktop keyring with um fal key set, or supplied through the FAL_KEY environment variable, which wins. Without a key, fal's OAuth relay can sign the user in through a browser. Local ComfyUI is supported through um comfy, and Sprixen and Voidless are separate paid services with their own keys for 2D sprites and animation sets.

The safety rails are more serious than the average modding toolkit. um backup snapshots, diffs and restores save folders, so a bad experiment does not cost a save. um publish check blocks shipping game files, decompiled code and leaked keys. The contribution rules say no game files, no decompiled dumps and no cheating other players, and they ask for an honest status and verification. None of that makes modding a game legally safe in every jurisdiction, and none of it protects a player who takes an agent into an anti-cheat protected online game. The README targets games you own, not live services.

The bill has three parts. The repository is MIT, the agent usage is whatever the host model costs, and the asset services are metered. A long mod can spend real money on image, 3D and audio generation before a single line of game code works. The 22 open issues and pull requests against 156 contributors suggest a project moving faster than it can polish, and the absence of a tagged release means every install tracks main. That is normal for a month-old repository with 9,220 stars, but it is not a stable distribution.

Project card

ItemDetail
Repositoryrehan-remade/universal-modder on GitHub
PurposeSkills, CLI and knowledge base for AI agents that mod PC games
Install targetsClaude Code, Codex, Gemini CLI, VS Code and Copilot, Cursor, OpenCode, skills-only
Skills12, including mod-any-game, game-recon, reverse-engineering, fal-assets, publish-mod and port-a-mod
Agents3 subagent roles: orchestrator, game worker and offline worker
CLIum, with scan, fal, comfy, sprite, render3d, win, video, backup, publish, kb and game-specific tools
Asset backendfal MCP, local ComfyUI, Sprixen and Voidless, each with its own key or OAuth path
SafetySave backups, publish checks, no game files, no decompiled dumps, no cheating other players
RuntimePython 3.10 or newer, Git, ffmpeg, uv recommended, Blender for 3D renders
LicenseMIT
Scale9,220 stars, 1,031 forks, 156 contributors, 22 open items, about 40 MB
Latest commitOctober 30, 2026, REA MCP integration note in the skills

FAQ

What is Universal Modder?
A set of Agent Skills, a Python CLI and a shared knowledge base that let coding agents mod PC games through a documented loop.

Which agents work with it?
Claude Code, Codex, Gemini CLI, VS Code and Copilot, Cursor and OpenCode, plus a skills-only install for other agents.

Does it mod any game?
It targets almost any PC game you own, with engine playbooks, but compatibility varies and anti-cheat or online games are not a safe target.

Does it need paid services?
The core is MIT. Asset generation through fal needs an API key or OAuth, local ComfyUI is an alternative, and Sprixen and Voidless are separate paid services.

Is it legal to use?
The project bans game files, decompiled dumps and cheating other players. Whether a specific mod is legal depends on the game, its license and the jurisdiction.

Is there a stable release?
No tagged release existed on October 30, 2026. Installs track the main branch, so pin a commit if reproducibility matters.

Bottom Line

Universal Modder is the most complete attempt yet to turn game modding into an agent workflow rather than a pile of prompts. The reconnaissance CLI, the engine playbooks and the field-note knowledge base are the right primitives, and the safety rules are written by people who understand that a modding agent can destroy saves or ship stolen files. The costs are real: metered asset generation, token-heavy agent runs, and a main branch with no release tag. Start with one small mod on a game you own, keep backups on, and read the field note before trusting the agent's plan.

Sources

  • GitHub rehan-remade/universal-modder README, install paths, skills, agents, CLI, knowledge base, safety rules and limits, accessed October 30, 2026
  • GitHub API metadata for rehan-remade/universal-modder, stars, forks, contributors, issues, size and latest commit, accessed October 30, 2026
  • GitHub commit message for pull request 462, REA MCP integration note, October 30, 2026

Related: Claude-Mem: Persistent Memory for Coding Agents at 105k Stars | REA Tools: A Website for Agent-Driven Reverse Engineering

Share