This is my personal (mostly) "AI slop free" adaption.
2026 was a wild year for the Forge project as LLM contributions really started coming in at full speed. These certainly resulted in a lot of valueable features - e.g. online drafting, iOS support and stickers.
But a less capable coding agent in the hands of an overeager person can lead to a slowly enfolding disaster: duct tape logic, technical debt and regressions that often take experience to spot instead of getting blinded from false security via countless badly written tests. 💣 In the end you'd be stuck with a very bloated code base that basically requires using an agent to work with...
Pushing back against low effort, sometimes even redundant PR and cleaning up after them takes time that could be better spent elsewhere. Therefore this branch aims to be more conservative in keeping vibecode: implementations deemed too messy to recover from will simply be reverted here, either full PR / commit / parts as necessary.
Additionally semi-official/rare abilities that unnecessarily put a heavy strain on the engine may get reworked into slightly more hardcoded variants - e.g. just because the card Topsy Turvy is supported not all games should have to suffer extra checks for it.
While parity with upstream will be attempted to remain high, possibly and depending on the speed of slop merges especially some GUI features could be delayed on this branch while quality control is still in progress. That means this approach should be most useful for other downstream consumers who just need a reliable engine to build around - or if you simply want less risk of being a guinea pig for LLM hallucinations.
There will be no releases for now, though everyone is welcome to give feedback if they have a problem specific to this build.
Overview of important fork amendments (or future candidates)
| Slop suspect | Priority | Rationale |
|---|---|---|
| e6f4f67a | B | one of the first more complex parts was merged completely unreviewed. it had no regard for existing systems or other PR in progress, the first iteration broke netplay support and quickly caused memory problems for mobile users. when pointed out to the author he disappointingly responded rather aggressive and unwilling to reconsider such rush jobs |
| dc72214 | A | extreme test bloat + hides bugs |
| 9c77436 | A | adding overhead for artificial problems coming from wrong method usage |
| 1406f4e | A | incomplete and badly written fix at absolutely the wrong layer |
| 082e9ff | A | implements an unnecessary cache for tokens. at this point it seems like the author just isn't qualified enough or the LLM dumping has simply become too tempting to slow down |
| 5a7f95b | B | Once again touching way too many files for different reasons; only kept the typo fixes in scripts |
| excessive pattern of obscure named and often single-use static fields | C | each LLM session invents its own tweak layer, nobody guides it towards a shared architecture |
| anything relying on try-catch usage everywhere | B | author doesn't care why his code fails |
| Flat MC | C | some good ideas but also very unhealthy amounts of vibecode with little manual testing, which then lead to additional bandaids slapped on top - e.g. 8360afa but the nail in the coffin is it not even fully respecting the preference - e.g. discardTargetAI-> this leads to a mix that actually worsens AI behaviour while also making it harder to debug user reports some "highlights": - fragile tests due to relying on reflection (without a good reason) - rule-breaking script refactors |
| 1679040 | C | a good example that the code isn't even read once (at least that's my more flattering interpretation): over 1/3 (50 lines) completely redundant merged within hours |
Due to the many infected files from AI-related changes these will require a more careful "brain surgery" - probably only after that slop train hopefully slows down a bit. The current strategy is to keep a secondary forge-ai-stable module structure which can be swapped in-place (only irrelevant test classes with compile errors). This avoids some extra merge conflicts and bonus fixes no longer fitting into its parent have already been built into it.
Join the Forge community on Discord!
Forge is a dynamic and open-source Rules Engine tailored for Magic: The Gathering enthusiasts. Developed by a community of passionate programmers, Forge allows players to explore the rich universe of MTG through a flexible, engaging platform.
Note: Forge operates independently and is not affiliated with Wizards of the Coast.
- 🌐 Cross-Platform Support: Play on Windows, Mac, Linux, and Android.
- 🔧 Extensible Architecture: Built in Java, Forge encourages developers to contribute by adding features and cards.
- 🎮 Versatile Gameplay: Dive into single-player modes or challenge opponents online!
- Latest Releases: Download the latest version here.
- Snapshot Build: For the latest development version, grab the
forge-gui-desktoptarball from our Snapshot Build.- Tip: Extract to a new folder to prevent version conflicts.
- User Data Management: Previous players’ data is preserved during upgrades.
- Java Requirement: Ensure you have Java 17 or later installed.
- (Note: Android 11 is the minimum requirement with at least 6GB RAM to run smoothly. You need to enable "Install unknown apps" for Forge to initialize and update itself)
- Download the APK from the Snapshot Build. On the first launch, Forge will automatically download all necessary assets.
- Build the IPA according to Wiki
- No jailbreak needed, only developer mode and iOS 16-27
- Connect your device to a PC to self-sign and upload the app file, multiple tools exist e.g. Sideloadly
Forge offers various exciting gameplay options:
Embark on a thrilling single-player journey where you can:
- Explore an overworld map.
- Challenge diverse AI opponents.
- Collect cards and items to boost your abilities.
Engage in focused gameplay without the overworld exploration—perfect for quick sessions!
Test your skills against AI in multiple formats:
- Sealed
- Draft
- Commander
- Cube
For comprehensive gameplay instructions, visit our User Guide.
Need help? Join our vibrant Discord community!
- 📜 Read the #rules and explore the FAQ.
- ❓ Ask your questions in the #help channel for assistance.
We love community contributions! Interested in helping? Check out our Contributing Guidelines for details on how to get started.
Forge aims to deliver an immersive and customizable Magic: The Gathering experience for fans around the world.
| Metric | Count |
|---|---|
| ⭐ Stars: | |
| 🍴 Forks: | |
| 👥 Contributors: |
📄 License: GPL-3.0