프레임워크 내부

저수준 프레임워크 개발, 컴파일러 최적화, 커스텀 연산자 구현을 위한 에이전트 스킬을 탐색하세요.

python-patterns

Pythonic イディオム、PEP 8標準、型ヒント、堅牢で効率的かつ保守可能なPythonアプリケーションを構築するためのベストプラクティス。

pytorch-patterns

PyTorch深度学习模式与最佳实践，用于构建稳健、高效且可复现的训练流程、模型架构和数据加载。

regex-vs-llm-structured-text

选择在解析结构化文本时使用正则表达式还是大型语言模型的决策框架——从正则表达式开始，仅在低置信度的边缘情况下添加大型语言模型。

Patterns for robust error handling across TypeScript, Python, and Go. Covers typed errors, error boundaries, retries, circuit breakers, and user-facing error messages.

huggingface-accelerate

NousResearch/hermes-agent

Run PyTorch training across GPUs with minimal changes.

peft-fine-tuning

NousResearch/hermes-agent

Fine-tune large LLMs with LoRA on limited GPU memory.

NousResearch/hermes-agent

Unsloth: 2-5x faster LoRA/QLoRA fine-tuning, less VRAM.

darwinian-evolver

NousResearch/hermes-agent

Evolve prompts/regex/SQL/code with Imbue's evolution loop.

NousResearch/hermes-agent

Jailbreak LLMs: Parseltongue, GODMODE, ULTRAPLINIAN.

setup-ts-deep-modules

mattpocock/skills

Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files. User-invoked.

How to add or modify Next.js experimental feature flags end-to-end. Use when editing config-shared.ts, config-schema.ts, define-env-plugin.ts, next-server.ts, export/worker.ts, or module.compiled.js. Covers type declaration, zod schema, build-time injection, runtime env plumbing, and the decision between runtime env-var branching vs separate bundle variants.

wpf-to-winui3-migration

microsoft/PowerToys

Guide for migrating PowerToys modules from WPF to WinUI 3 (Windows App SDK). Use when asked to migrate WPF code, convert WPF XAML to WinUI, replace System.Windows namespaces with Microsoft.UI.Xaml, update Dispatcher to DispatcherQueue, replace DynamicResource with ThemeResource, migrate imaging APIs from System.Windows.Media.Imaging to Windows.Graphics.Imaging, convert WPF Window to WinUI Window, migrate .resx to .resw resources, migrate custom Observable/RelayCommand to CommunityToolkit.Mvvm source generators, handle WPF-UI (Lepo) to WinUI native control migration, or fix installer/build pipeline issues after migration. Keywords: WPF, WinUI, WinUI3, migration, porting, convert, namespace, XAML, Dispatcher, DispatcherQueue, imaging, BitmapImage, Window, ContentDialog, ThemeResource, DynamicResource, ResourceLoader, resw, resx, CommunityToolkit, ObservableProperty, WPF-UI, SizeToContent, AppWindow, SoftwareBitmap.

add-uint-support

pytorch/pytorch

Add unsigned integer (uint) type support to PyTorch operators by updating AT_DISPATCH macros. Use when adding support for uint16, uint32, uint64 types to operators, kernels, or when user mentions enabling unsigned types, barebones unsigned types, or uint support.

pytorch/pytorch

Convert PyTorch AT_DISPATCH macros to AT_DISPATCH_V2 format in ATen C++ code. Use when porting AT_DISPATCH_ALL_TYPES_AND*, AT_DISPATCH_FLOATING_TYPES*, or other dispatch macros to the new v2 API. For ATen kernel files, CUDA kernels, and native operator implementations.

cuda-index-width

pytorch/pytorch

Choose 32-bit vs 64-bit index math in PyTorch CUDA kernels. Use when fixing large-tensor indexing overflows, deciding whether to use int64_t, canUse32BitIndexMath, CUDA_KERNEL_LOOP_TYPE, or AT_DISPATCH_INDEX_TYPES, and when considering binary-size or performance impact of index-type templating.

pytorch/pytorch

Write Metal/MPS kernels for PyTorch operators. Use when adding MPS device support to operators, implementing Metal shaders, or porting CUDA kernels to Apple Silicon. Covers native_functions.yaml dispatch, host-side operators, and Metal kernel implementation.

pytorch/pytorch

Debug PyTorch 2 compiler stack failures including Dynamo graph breaks, Inductor codegen errors, AOTAutograd crashes, and accuracy mismatches. Use when encountering torch.compile errors, BackendCompilerFailed exceptions, recompilation issues, Triton kernel failures, FX graph problems, or when the user mentions debugging PT2, Dynamo, Inductor, or compiled model issues.

JuliusBrussee/caveman

Ultra-compressed communication mode. Cuts output tokens 65% (measured) by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

JuliusBrussee/caveman

Ultra-compressed communication mode. Cuts output tokens 65% (measured) by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

dedupe-issue-local

warpdotdev/warp

Repo-specific dedupe guidance for warp. Only the categories declared overridable by the core dedupe-issue skill may be specialized here.

analysis-api-mark-internal-apis

JetBrains/kotlin

Drive the per-module internal-API codebase test, then refine the suggested annotations down to `internal` (or up to `@KaImplementationDetail`) based on actual external usage

decompress-binary

ClickHouse/ClickHouse

Extract the inner ELF from a ClickHouse self-extracting `clickhouse` binary, including when its architecture differs from the host (e.g. to load an aarch64 CI core dump on an x86 workstation). Use when gdb/lldb needs real symbols from a downloaded CI/release binary, or when self-extraction by running the binary is not possible because of an architecture mismatch.

discourse-upcoming-changes-authoring

discourse/discourse

Use when modifying, debugging, or extending the upcoming changes framework code and system itself.

stage-tamagotchi-godot-csharp

Apply engine-local C# development method and code style only when working in `engines/stage-tamagotchi-godot`, including its `.cs` files, `.csproj`, engine-local `.editorconfig`, and Godot-specific C# structure decisions. Do not use for TypeScript, Electron, renderer code, shared workspace config, repo-wide C# conventions, or any file outside `engines/stage-tamagotchi-godot`.

hyperframes-registry

calesthio/OpenMontage

Install and wire registry blocks and components into HyperFrames compositions. Use when running hyperframes add, installing a block or component, wiring an installed item into index.html, or working with hyperframes.json. Covers the add command, install locations, block sub-composition wiring, component snippet merging, registry discovery, and authoring a new block or component to contribute upstream (idea → scaffold → validate → PR).

luongnv89/claude-howto

Tạo hoặc cập nhật các file CLAUDE.md theo các thực hành tốt nhất để onboarding tác nhân AI tối ưu

chakra-ui-migrate

chakra-ui/chakra-ui

Migrate Chakra UI projects from v2 to v3, covering package changes, codemods, provider setup, color mode, prop renaming, compound components, theming, recipes, and Next.js updates. Use this skill whenever a user is upgrading Chakra UI versions, encountering breaking changes after an upgrade, converting old v2 patterns (ColorModeScript, useColorModeValue, styleConfig, extendTheme, isDisabled, colorScheme, @chakra-ui/icons, framer-motion dependency), fixing compound component patterns, or asking about differences between Chakra UI v2 and v3 — even if they don't say "migrate" or "upgrade" explicitly.

spark-environment-setup

wshobson/agents

Set up a working ML training/inference environment on NVIDIA DGX Spark (GB10, aarch64, CUDA 13). Use when installing PyTorch/Unsloth/TRL/vLLM on DGX Spark, hitting libcudart or wheel-ABI errors on aarch64, or choosing between NGC containers and bare pip installs.

python-error-handling

wshobson/agents

Python error handling patterns including input validation, exception hierarchies, and partial failure handling. Use when implementing validation logic, designing exception strategies, handling batch processing failures, or building robust APIs.

create-implementation-plan

github/awesome-copilot

Create a new implementation plan file for new features, refactoring existing code or upgrading packages, design, architecture or infrastructure.

github/awesome-copilot

Get best practices for C# async programming

github/awesome-copilot

Ready-to-use prompts for comprehensive .NET framework upgrade analysis and execution

github/awesome-copilot

Get best practices for Entity Framework Core

multi-stage-dockerfile

github/awesome-copilot

Create optimized multi-stage Dockerfiles for any language or framework

github/awesome-copilot

CommunityToolkit.Mvvm (the MVVM Toolkit) core: source generators ([ObservableProperty], [RelayCommand], [NotifyPropertyChangedFor], [NotifyCanExecuteChangedFor], [NotifyDataErrorInfo]), base classes (ObservableObject / ObservableValidator / ObservableRecipient), commands (RelayCommand / AsyncRelayCommand), and validation. Companion skills: mvvm-toolkit-messenger for pub/sub, mvvm-toolkit-di for Microsoft.Extensions.DependencyInjection wiring. Works across WPF, WinUI 3, MAUI, Uno, and Avalonia.

refactor-method-complexity-reduce

github/awesome-copilot

Refactor given method `${input:methodName}` to reduce its cognitive complexity to `${input:complexityThreshold}` or below, by extracting helper methods.

semantic-kernel

github/awesome-copilot

Create, update, refactor, explain, or review Semantic Kernel solutions using shared guidance plus language-specific references for .NET and Python.

Yeachan-Heo/oh-my-codex

[OMX] Strict autonomous loop: $deep-interview -> $ralplan -> $ultragoal (+ $team if needed) -> $code-review -> $ultraqa

opentrons-integration

K-Dense-AI/scientific-agent-skills

Author, review, migrate, simulate, and troubleshoot official Opentrons Python Protocol API v2 protocols for Flex and OT-2 robots. Use for robot-specific liquid handling, deck and labware setup, pipettes, modules, runtime parameters, liquid classes, and Opentrons App analysis. Use pylabrobot instead when one workflow must support multiple robot vendors.

K-Dense-AI/scientific-agent-skills

Hardware-agnostic quantum ML framework with automatic differentiation. Use when training quantum circuits via gradients, building hybrid quantum-classical models, or needing device portability across IBM/Google/Rigetti/IonQ. Best for variational algorithms (VQE, QAOA), quantum neural networks, and integration with PyTorch or JAX. For hardware-specific optimizations use qiskit (IBM) or cirq (Google); for open quantum systems use qutip.

K-Dense-AI/scientific-agent-skills

Multi-objective optimization framework. NSGA-II, NSGA-III, MOEA/D, Pareto fronts, constraint handling, benchmarks (ZDT, DTLZ), for engineering design and optimization problems.

pytorch-lightning

K-Dense-AI/scientific-agent-skills

Deep learning framework (PyTorch Lightning / lightning package). Organize PyTorch code into LightningModules, configure Trainers for multi-GPU/TPU, implement data pipelines, callbacks, logging (W&B, TensorBoard, MLflow), distributed training (DDP, FSDP, DeepSpeed), for scalable neural network training.

K-Dense-AI/scientific-agent-skills

Build and troubleshoot TorchDrug 0.2.1 workflows for molecular graphs, property prediction, self-supervised pretraining, molecule generation, retrosynthesis, protein representation learning, and knowledge graph reasoning. Use when code imports torchdrug or needs its datasets, models, tasks, or Engine.

sgl-project/sglang

Step-by-step tutorial for adding a heavyweight AOT CUDA/C++ kernel to sgl-kernel (including tests & benchmarks)

huggingface-accelerate

davila7/claude-code-templates

Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device placement, mixed precision (FP16/BF16/FP8). Interactive config, single launch command. HuggingFace ecosystem standard.

pytorch-lightning

davila7/claude-code-templates

High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales from laptop to supercomputer with same code. Use when you want clean training loops with built-in best practices.

davila7/claude-code-templates

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

davila7/claude-code-templates

Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency. Use for production deployment on NVIDIA GPUs (A100/H100), when you need 10-100x faster inference than PyTorch, or for serving models with quantization (FP8/INT4), in-flight batching, and multi-GPU scaling.