菜单
Use when working with PyTorch APIs, tensors, autograd, torch.nn, torch.optim, torch.compile, export, distributed/RPC, CUDA/MPS/XPU backends, profiling, quantization, ONNX export, C++/LibTorch, or PyTorch runtime/debugging behavior.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Use when answering questions about Claude Code CLI features, configuration, hooks, skills, MCP, permissions, settings, sub-agents, teams, plugins, keybindings, IDE integration, CI/CD setup, troubleshooting, or any Claude Code usage topic.
Use when working with Apache Doris: table design, data models (Duplicate/Unique/Aggregate), partitioning, bucketing, SQL syntax, data import (Stream Load, Broker Load, INSERT INTO), data export, lakehouse (Hive/Iceberg/Hudi/Paimon catalogs), materialized views, query acceleration, inverted index, compute-storage decoupled mode, administration, or Doris ecosystem tools.
justfile (just) documentation — a command runner with make-inspired syntax. Covers recipes (parameters, dependencies, shebang/script recipes), justfile language (settings, strings, functions, constants, attributes, conditionals, backticks), modules/imports, variables, shell configuration, and CLI options. USE THIS SKILL WHEN the user asks about just, justfile syntax, just recipes, just modules, just functions, or just settings.
USE THIS SKILL WHEN working with Mastra (TypeScript AI agent framework): building agents, workflows, RAG pipelines, memory, voice, evals, MCP tools, observability/tracing, deploying to Mastra Cloud/Cloudflare/Vercel, or using Mastra client SDK. Triggers on: mastra, Mastra agent, Mastra workflow, createAgent, createWorkflow, mastra.run, MastraClient, @mastra/core.
Use when working with the OpenRouter API: model routing, provider selection, model variants (free/nitro/thinking), tool calling, structured outputs, prompt caching, OAuth, API keys, rate limits, streaming, embeddings, Claude Code integration, or any OpenRouter-specific feature.
Use when working with Ultralytics YOLO, including YOLO26/YOLO11/YOLOv8 models, detect/segment/classify/pose/OBB tasks, train/val/predict/export/track/benchmark modes, dataset YAML formats, HUB/platform workflows, integrations, solutions, or ultralytics Python API behavior.
| name | pytorch-docs |
| description | Use when working with PyTorch APIs, tensors, autograd, torch.nn, torch.optim, torch.compile, export, distributed/RPC, CUDA/MPS/XPU backends, profiling, quantization, ONNX export, C++/LibTorch, or PyTorch runtime/debugging behavior. |
Official PyTorch docs synced from pytorch/pytorch/docs.
Use this skill for PyTorch framework APIs, tensor semantics, autograd, neural network modules, compilation/export, distributed training, device backends, profiling/debugging, quantization, ONNX, and LibTorch/C++ docs.
references/ before giving version-sensitive PyTorch API, runtime behavior, environment variable, backend, distributed, compile/export, or C++/LibTorch guidance.torch.* API names, environment variables, dispatch behavior, compile/export limitations, or distributed configuration without checking references.rg -n "torch\.compile|torch.export|dynamo|inductor|AOT|graph break" skills/pytorch-docs/references
rg -n "autograd|tensor|nn\.Module|optimizer|DataLoader|distributed|RPC" skills/pytorch-docs/references
rg -n "CUDA|ROCm|HIP|MPS|XPU|environment variable|memory|profil" skills/pytorch-docs/references
rg -n "LibTorch|C\+\+|cpp|ONNX|quantization" skills/pytorch-docs/references
references/source/ — main Python PyTorch docs and notes: tensors, autograd, modules, optimizers, distributed, compile/export, profiling, backend/runtime behavior.references/source/notes/ — deeper technical notes and backend/runtime caveats.references/source/user_guide/ — user guides for compilation, export, distributed, and runtime workflows.references/cpp/ and references/libtorch.rst — C++/LibTorch API docs.rg terms.