Firzus/agent-skills / skills/game/unity/unity-core
unity-core
UnverifiedUse for any Unity task: writing, reviewing, or architecting Unity code, choosing between two Unity tools that do the same job, or when the user mentions Unity, CoreCLR, IL2CPP, Awaitable, Addressables, Content Directories, DOTS, ECS, Jobs, Burst, assembly definitions, Input System, audio, project structure, CI, builds, tests, or performance. Load it alongside the domain skills unity-ui, unity-editor-tools, unity-rendering, unity-multiplayer, unity-cli, and unity-play.
Unity Core
Ship on the default stack below. Each row names the one tool to reach for, so the choice is already made — spend the thinking on the feature instead.
This skill describes the latest Unity version the team uses, and nothing older. The Editor and desktop players run on CoreCLR; Android, iOS, and the Web build with IL2CPP. Code targets C# 9 and .NET Standard 2.1 until Unity ships .NET 10 and C# 14; runtime.md owns what the runtime changes for your code.
For what a build actually ships, read its release notes and Unity's Discussions posts: the manual still carries pages written for earlier versions.
The default stack
| Need | Reach for | Instead of |
|---|---|---|
| Any new UI | UI Toolkit — unity-ui | UGUI |
| World-space / diegetic UI | Panel Renderer | UGUI world canvas, render texture |
| Text | Advanced Text Generator | hand-rolled layout |
| Player-facing text in several languages | Localization module + Smart Strings, see unity-ui | the Localization package, hand-rolled string tables |
| Authoring UI for a custom tool | UI Toolkit — unity-editor-tools | Odin, IMGUI, a node canvas |
| Heterogeneous items in one authored list | [SerializeReference] + AdvancedDropdown | one list per type |
| Tool icons | Built-in Editor icons, then painter2D | shipping PNG variants per state |
| Async | Awaitable + CancellationToken | coroutines, raw Task |
| Static state across Play Mode entries | [AutoStaticsCleanup], see runtime.md | expecting a reload to reset it |
| Runtime loading | Addressables + AssetReference, every group on the Content Directory schema; post-install content goes through a project system until Unity ships remote Content Directories, see assets.md; settings needed before the first scene go through the Preloaded Assets | Resources/, Loadable<T> without Addressables |
| Input | Input System action maps | legacy Input Manager |
| Audio | built-in AudioMixer — workflow.md | FMOD/Wwise |
| Render pipeline | URP — unity-rendering | Built-In, HDRP |
| Custom render passes | Render Graph | ScriptableRenderPass.Execute |
| Lighting, static scenes | Adaptive Probe Volumes + baked lightmaps | Light Probe Groups |
| Light baking | Unity Compute Light Baker, set as the Default Light Baker | every other light-baking backend |
| Lighting, dynamic scenes | Surface Cache GI | baking a scene that moves |
| Draw-call reduction and occlusion | GPU Resident Drawer + GPU occlusion culling | static batching |
| Pooling | UnityEngine.Pool | hand-rolled pools |
| Object identity | EntityId (64-bit) | GetInstanceID(), int ids |
| Hot paths | Jobs + Burst | a full ECS rewrite |
| Netcode | Netcode for GameObjects — unity-multiplayer | Netcode for Entities |
| Authority | client-server | distributed authority |
| Sessions, matchmaking, relay | Multiplayer Services | Lobby + Matchmaker + Relay separately |
| Server builds | Dedicated Server + Multiplayer Roles | a hand-stripped client build |
| Testing multiple peers | Multiplayer Play Mode | several Editor installs |
| Per-platform build config | Build Profiles | hand-edited global Build Settings |
| Build-time scene processing | AssetPostprocessor.OnProcessScene | IProcessSceneWithReport, [PostProcessScene] |
| Tests | Unity Test Framework, edit-mode first | play-mode by default |
| Tests while the Editor is open | unity command run_tests — unity-cli | closing the Editor for unity test |
| Release builds | CoreCLR on desktop; IL2CPP on Android, iOS, and the Web | IL2CPP on desktop |
| Driving the Editor from an agent | Unity CLI — unity-cli | asking the user to click through the Editor |
Leave a default only where the linked reference gives that row an escape hatch. Say which row you left and what made it worth leaving.
Reference
| Topic | File |
|---|---|
CoreCLR runtime, static state, serialization, EntityId | runtime.md |
Composition, asmdefs, Awaitable, Jobs/Burst/ECS, profiling, allocations | architecture.md |
Exceptions, nullability, async void, the UnityEngine.Object null trap | code-standards.md |
| Folder layout, naming, GUID-safe renames | project-structure.md |
| Addressables, Content Directories, import settings, scenes, prefabs | assets.md |
| Input, audio, version control, CI, builds, testing | workflow.md |
| Device measurement, character animation, particles, managed code stripping, Android LTO, iOS memory limits | performance.md |
Domain skills
Each domain has its own skill, which applies this default stack and review pass. When the task touches a domain, load its skill; when that skill is not installed, say which one is missing and work from the rows above.
| Domain | Skill |
|---|---|
| UI Toolkit, USS versus CSS, design tokens, data binding, MVP, localization | unity-ui |
| Figma designs as UI Toolkit interfaces | figma-to-unity |
Custom Editor tools, inspectors, [SerializeReference] | unity-editor-tools |
| Pipeline, lighting, GPU-driven rendering | unity-rendering |
| Netcode, authority, sessions, dedicated server | unity-multiplayer |
| Unity CLI, driving the Editor from an agent, iterating while the user tests in the Editor | unity-cli |
| An agent playing the game in Play Mode, reproducing a gameplay bug by playing | unity-play |
| Performance review of a diff or folder: rule scan of code, Editor audit of assets and settings | unity-performance |
For engine-agnostic replication theory, use coop-session.
Review pass
Read changed code against every row of the default stack, then account for each line below — name the file and line where it holds, or fix it:
Several rows are conditional: they bind only when the change performs the operation they describe. "Not applicable, this change does not do that" is a complete answer, and the right one. Test the condition, never the presence of a token — and never infer the rule from how many neighbouring files show the same pattern, since copy-paste propagates a mistake as reliably as a convention.
- [ ] Statics reset through `[AutoStaticsCleanup]` or a lifecycle hook, so entering Play Mode twice behaves identically
- [ ] Every Awaitable carries a CancellationToken scoped to its component
- [ ] Every CancellationTokenSource this change creates, linked ones included, is disposed by its owner
- [ ] Assets load through Addressables, and every handle is released or bound to a lifetime token
- [ ] Update() free of allocation, GetComponent, Find*, and LINQ
- [ ] Runtime spawns come from a pool
- [ ] No runtime ParticleSystem.Simulate fast-forwards a Noise system by seconds
- [ ] Object identity flows through EntityId
- [ ] Player-facing strings come from Localization tables
- [ ] [SerializeField] on fields only ([field: SerializeField] for properties)
- [ ] Logic sits in plain C# in its own asmdef, covered by edit-mode tests
- [ ] Renamed assets moved with their .meta, via git mv
- [ ] Networked state gated on HasAuthority
- [ ] Editor code under an Editor asmdef, runtime types in runtime assemblies
- [ ] Any rename or move of a [SerializeReference] type in this change lands with [MovedFrom] naming its real former address (no former address, no attribute)
- [ ] Custom inspectors built on UI Toolkit, with IMGUIContainer only where named
When the user asks for something off the default stack, name the row, state what it costs, and follow their call.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |