The load-bearing design decisions of the SparkEngine C++23 game engine: the invariants that MUST hold, WHY each exists, what breaks if you violate it, and the OPEN known-weak-points. TRIGGER when you are about to add or wire a subsystem, touch the service locator / EngineContext, change ECS system ordering or the system-manager plumbing, add a global, change RHI backend selection, cross the DLL/game-module boundary, or you catch yourself asking "how is this engine supposed to fit together / where does X live / can I add a g_ global / why is my subsystem null in a module". DO NOT TRIGGER for pure build/CMake/CI failures (use the global cmake-msvc / ci-troubleshoot skills), pure formatting, or a self-contained bug fix that touches no cross-subsystem boundary and adds no new system.
SparkEngine subsystem-specific symptom lookup table: you observed a concrete wrong behavior in a SparkEngine subsystem (AI behavior trees, weapons, the ImGui editor, undo/redo, MMO modules, localization, cinematic sequencer, ECS system ticking, or client-side network prediction) and need to know where to look, what the correct code should look like, and a discriminating experiment to confirm the cause before changing anything. TRIGGER when the user says things like "cloned behavior tree does nothing", "weapon damage is attributed to entity 0", "editor crashes when I open a scene", "editor won't prompt to save", "MMO module uses the wrong NetworkManager", "localized string is garbage", "cutscene has no audio", "my ECS system never ticks / never runs", or "multiplayer is rubber-banding / desyncing". DO NOT TRIGGER for generic C++ crashes, undefined-behavior hunts, or compiler/linker error messages (use the global cpp-crash-triage and msvc-error-decoder skills), nor for a chronological "what changed and why" nar
SparkEngine concurrency and threading-discipline runbook: the real JobSystem / ThreadSafeQueue / StageBasedExecutor / ParallelPerception APIs, the ECS phase barriers, per-subsystem thread-safety rules, and a checklist for parallelizing a new system without data races. TRIGGER when: "parallelize this system", "is it safe to call from a worker thread", "add a job", "JobSystem", "Submit / ParallelFor", "which stage does my system run in", "data race / crash under TSan", "can I touch the ECS from a background thread", "thread-safe queue", "Jolt physics threads". DO NOT TRIGGER for: performance profiling / measuring / optimizing hot paths (use the global perf-hotpath skill), rendering-backend RHI work, or networking protocol logic.
Reference for SparkEngine's asset import/load pipeline in SparkEngine/Source/Graphics — how a source file (.obj/.fbx/.gltf/.glb, .png/.tga/.dds, .wav) becomes a runtime MeshAsset/TextureAsset/AudioAsset, what AssetMetadata tracks (guid, checksum, timestamps, fbx.* custom properties), and the non-obvious Windows/Linux platform-split (AssetPipelineWindows.cpp vs AssetPipelineLinux.cpp, D3D11 vs RHI/tinyobj/cgltf/stb_image). TRIGGER when: adding a new asset type or source format, editing AssetPipeline / AssetTypes / AssetMetadata / FBXImporter / ModelLoading, wondering "why is FBX loading Linux-only?", "why are there two .cpp files per asset class?", or "where does mesh loading actually happen?". DO NOT TRIGGER when: writing render-graph / RHI backend code (that is the RHI layer, not the asset layer — read SparkEngine/Source/Graphics/RHI directly), authoring shaders, or doing runtime ECS/gameplay work — use sparkengine-ecs-query-patterns for ECS instead. Phrases: "import an fbx", "add a texture format", "asset m
Catalog of every SparkEngine CMake build option (ENABLE_*/SPARK_*/BUILD_*), its real verified default, the C/C++ #define it produces, what code it gates, and the CMakePresets it is set by. TRIGGER when the user asks "what does ENABLE_VULKAN / ENABLE_EDITOR / ENABLE_DXR do", "how do I turn off networking / the editor / tests", "which preset builds shipping", "why is my #ifdef SPARK_* not firing", "what flag maps to which define", or "how do I add a new build toggle correctly". DO NOT TRIGGER for MSVC toolchain setup, vcvars, sccache, or build-error triage (use the global cmake-msvc skill instead) — this skill is only the project's feature-flag surface, not how to run the compiler.
Reference for SparkEngine's four hand-rolled memory allocators — FrameAllocator (per-frame bump), LockFreeRingAllocator (producer/consumer ring), HandleAllocator/VersionedHandle (generational handles), and RHI TransientBufferAllocator (GPU vertex/index suballocation). TRIGGER when you are about to write per-frame or transient allocations, ask "which allocator should I use", need the exact method signatures / thread-safety of one of these, wonder why the code uses naked buffers instead of new/delete, or need to prevent use-after-free on recycled slots/indices. DO NOT TRIGGER for hunting an actual leak or use-after-free bug (use the memory-hunt skill), for general std::unique_ptr ownership questions, or for GPU resource lifetime / ComPtr / RHI backend selection.
How to query and iterate the EnTT-based ECS in SparkEngine — building views over component combinations, iterating entities inside a system, the CoreComponents vs domain-header layout, the data-driven archetype/prefab spawn format, and the fixed Physics->Animation->AI->Audio->Lifecycle->Render execution order. TRIGGER when the user says "iterate entities with component X", "write a new ECS system", "loop over all entities that have", "GetEntitiesWith", "registry.view", "EnTT view or group", "how do I spawn a prefab/archetype", "what order do systems run in", "why does my system read a stale Transform", or "where do I add a new component struct". DO NOT TRIGGER for how to REGISTER a component with reflection/editor/save (see sparkengine-reflection-conventions), for AngelScript scripting (see sparkengine-hot-reload-seam), or for generic EnTT questions unrelated to this codebase's conventions.
Chronological case-file of SparkEngine's hardest, already-solved bugs — each with symptom, root cause, the exact fix commit, and current landed/open status. TRIGGER when you are about to re-investigate something that "feels like it was fixed before", when you say "has this ICF / type-id / C++23 gate / blue-screen / macOS build / camera race / AngelScript hot-reload thing happened before", when a Windows Release-only CI failure reappears, when you want the STORY and commit hash of a past incident, or when auditing which harden-fleet items are still deferred. DO NOT TRIGGER for a brand-new live symptom you just hit and want a fix recipe for — use the sibling sparkengine-debugging-playbook (symptom->fix triage) instead; this skill is a history book, not a first-responder.