Universal Modder
Take an AI coding agent from PC game reconnaissance through mod building and live-game verification.
- Source repo
- rehan-remade/universal-modder
- Stars
- ★ 8.5k
- Last updated
- today
- License
- MIT
- Primary language
- Python
- FA score
- 49/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A Terraria modder wants to add weapons and enemies with tModLoader and create their sprites.
- Not a fit if
- Teams modifying online games protected by anti-cheat
- Users seeking to bypass DRM or ownership checks
- Source review
- 49/100 · Major gaps
What does this agent do, and when should you use it?
Universal Modder equips several AI coding agents with Agent Skills, a Python um CLI, fal MCP configuration, and a knowledge base for game-specific field notes. It can scan Steam, Epic, and Xbox installs, identify engines and versions, and guide reverse engineering, asset creation, testing, and packaging. Agents use um commands for game scans, generated assets, sprite processing, window automation, save backups, video, and knowledge-base tasks. Asset workflows can use fal or a local ComfyUI server for images; Sprixen is a separate paid API for 2D art. The README demonstrates mods for Terraria, Age of Empires II DE, and a Minecraft–GTA V mashup, with workflows that test mods in the running game and record what was learned.
After installing the relevant agent integration, Skills, fal MCP configuration, and um CLI, an agent can run um scan to locate a game and identify its engine, version, anti-cheat status, installed loaders, and save folders. It can search prior field notes with um kb, follow mod-any-game and related skills to inspect actual code, and build a working mod slice. Asset commands include um fal, um comfy, and um sprixen; um sprite processes sprite assets and um render3d renders 3D models. um backup snapshots and restores saves, while um win launches, captures, records, and drives the game window; the workflow calls for verification in the running game. For handoff, um publish check checks a package and um kb new, um kb check, and um kb pr support field-note contributions.
- A Terraria modder wants to add weapons and enemies with tModLoader and create their sprites.
- An Age of Empires II DE creator wants to add a civilization and render 3D units from the game's camera angle.
- A developer familiar with Fabric and ScriptHookV wants to connect Minecraft gameplay to GTA V.
- A player unsure of an installed game's engine, version, or modding route wants to start with
um scanand search existing field notes. - A contributor wants to document tested modding steps so later developers can reuse them and submit a knowledge-base PR.
How do you install or deploy this agent?
Choose an agent integration. Claude Code and Codex install through their plugin marketplace commands; the README also documents a Gemini CLI extension, VS Code / Copilot plugin installation, Cursor and OpenCode clone workflows, and Skills-only installation. Install the standalone um CLI with uv or pipx. The documented runtime prerequisites are Git, Python 3.10+, and ffmpeg; Blender is needed for 3D-to-sprite renders. fal asset generation needs a fal API key, and Sprixen 2D art commands use a separate Sprixen API key.
/plugin marketplace add rehan-remade/universal-modder
/plugin install universal-modder@universal-moddercodex plugin marketplace add rehan-remade/universal-modder
codex plugin add universal-modder@universal-modderuv tool install git+https://github.com/rehan-remade/universal-modderum fal key set
export FAL_KEY="$(secret-tool lookup service fal.ai key api)"How do you use this agent?
On a machine with the target game installed, start the configured coding agent and begin with um scan. Search for prior work using um kb search, then describe the mod you want the agent to make. The documented workflow proceeds through reconnaissance, save backup, code inspection, a working slice, asset creation, and verification in the running game. Use um --help for command-specific help; check packages before publishing and record reusable findings as a field note.
um scan
um kb search "grand theft auto"
um --helpWhat are this agent's strengths and limitations?
- Covers the modding workflow from game identification and engine-specific routes through reverse engineering, asset processing, live-game verification, video, and packaging.
- Combines
um scanwith field notes that record tested game versions, verification, and symptom–cause–fix gotchas. - Documents integrations for Claude Code, Codex, Gemini CLI, VS Code / Copilot, Cursor, and OpenCode.
- Requires local game files, Git, Python 3.10+, and ffmpeg; 3D-to-sprite rendering also needs Blender.
- fal asset generation needs a fal API key; Sprixen 2D art requires a separate key and paid service.
- It explicitly excludes injecting into anti-cheat protected online games and bypassing anti-cheat, DRM, or ownership checks.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Universal Modder This agent | 49 · Major gaps | Agent plugin / skill | ★ 8.5k | today | Python | Codex · Claude Code |
| Cheat Engine MCP Bridge | 59 · Major gaps | MCP serverFree + model costs | ★ 1.7k | 2mo ago | Lua | Codex · Claude.ai |
| Co-op Anything | 37 · Major gaps | Agent plugin / skillFree + model costs | ★ 693 | today | Python | Codex · Claude Code |
| Spring AI Alibaba | 53 · Major gaps | FrameworkFree + model costs | ★ 11k | 13d ago | Java | — |
How does FollowAgents rate this agent?
Why each dimension lost points
README and SECURITY.md state concrete safeguards: no injection into anti-cheat online games, no DRM/anti-cheat bypass, no shipping game files or decompiled code, save backups before touching saves, PID-only kills, and asking before driving input, installing loaders, or publishing. Those are real positives. However the permission surface is broad: um win drives real input, kills processes and writes the registry; plugin/MCP installs pull external code; the Minecraft x GTA example listens on 127.0.0.1:25599 without a token (author-declared design). Key handling has test evidence (keyring first, 0600 file / 0700 dir, secret-tool via stdin, masked status), but .env plaintext fallback and FAL_KEY passthrough to the MCP widen exposure. Dependencies are only pillow/numpy/pyyaml with no lockfile or hashes, and external tools (ffmpeg, Blender, ComfyUI, fal, Sprixen) are unpinned. Rollback exists via um backup snapshot/diff/restore but covers only save folders, not loader installs or registry edits. Attribution lists upstream projects in Credits but not per-file.
Tests cover encoding, key lookup and KB search with specific assertions, and CI runs pytest plus --help for every subcommand, indicating decent internal consistency. Dependency availability is only declared as Python>=3.10 plus external binaries with no version pinning; failure messages appear as SystemExit and hints in tests, but README does not systematically document error codes or diagnostics, so only a thin score.
Targets many coding agents (Claude Code, Codex, Cursor, Gemini CLI, Copilot, OpenCode) and many engines, with broad scenarios; capability boundaries are stated in Rules and safety.md. Trigger precision relies on skill descriptions and AGENTS.md routing, with no tests or conflict-handling notes for when skills should or should not fire, hence the deduction. Environment fit covers Windows native or WSL and Linux keyring, but macOS specifics are unstated.
README is well structured, the install table covers multiple agents, examples and a knowledge index are present, and the MIT license is complete. But there is no CHANGELOG and the version is only pyproject 0.2.0; known limitations are scattered rather than consolidated; maintenance responsibility points only to Discord and GitHub private reporting with no governance or SLA; naming stability couples plugin/marketplace names to the repo with no deprecation policy.
Examples (Terraria, AoE2, Minecraft x GTA) and the CLI subcommand table show usable assets and mods, and marginal value comes from chaining recon, reverse engineering, asset generation, testing and video into one loop. On cost, the three-agent team is self-described as several times the tokens, fal/Sprixen are paid services, and Blender/ffmpeg are heavy dependencies, so cost-benefit is only partly addressed.
README claims mostly trace to specific skill files and example directories, and the knowledge base asks for versions and verification methods, which is a traceability start. But broad claims (e.g. 'almost any PC game') lack independent corroboration, examples ship no reproducible test output, and fact and inference are mixed in the README (e.g. the 'September 2026 wave'), without clear separation. Static review cannot verify runtime results.
- um win can drive real input, kill processes by PID and write the registry, so a mistake has wide blast radius; first use in an isolated VM or sandbox.
- The Minecraft x GTA example's local WebSocket listens on 127.0.0.1:25599 without a token (author-declared design), reachable by other local processes.
- FAL_KEY is passed to the MCP server via environment and a plaintext .env fallback exists; prefer the keyring and avoid exporting it in shared shells.
- External dependencies (ffmpeg, Blender, ComfyUI, fal, Sprixen) are unpinned, so behavior may drift with upstream changes.
- Publisher identity is unverified by FollowAgents; review code and update paths before installing plugins or MCP servers.