GITHUB AGENT SKILLS / 技能收录
给你的 AI 助手
装上专业技能
我们从 GitHub 持续收录、审核并分类优质的 Agent Skills,按职业与使用场景整理,复制一行命令即可安装到 Claude Code、Cursor 等智能体。
按职业浏览 Agent Skills
为不同岗位挑选的高频技能
按使用场景浏览 Agent Skills
从你要完成的任务出发
按创作者浏览 Agent Skills
按 GitHub Stars 排序的优秀创作者
Microsoft
@microsoftopenclaw
@openclawAnthropic
@anthropicsOpenAI
@openaiThe Apache Software Foundation
@apacheGitHub
@githubAutoGPT
@Significant-GravitasGarry Tan
@garrytan全部 Skills 482556
markdown-manager
Expert persona for documentation and Markdown management. Use when writing project plans, PR reviews, or technical docs. Ensures consistency, structure, and canonical storage.
modal-serverless-gpu
Serverless GPU cloud platform for running ML workloads. Use when you need on-demand GPU access without infrastructure management, deploying ML models as APIs, or running batch jobs with automatic scaling.
import-profile
Weekly review script to check for updates in ~/.claude setup (settings.json, memory, goals, skills) and sync them into the Gemini environment. Use this when the user asks to "import profile" or run the weekly ~/.claude sync.
finish-worktree
Exits a git worktree, prepares a draft PR from the diff, applies project PR templates, and outputs copy-pasteable markdown. Use when a task in a worktree is complete.
mlflow
Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform
patch-my-hosts
Sync /etc/hosts with the latest upstream from someonewhocares.org while preserving your custom edits. Use when the user says "/patch-my-hosts", "update my hosts file", "refresh hosts blocklist", or when a session-start notice says a hosts update is pending.
ml-training-recipes
Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning neural networks, debugging loss spikes or OOM, choosing architectures, or optimizing GPU throughput.
llm-wikify
Set up an LLM-maintained wiki inside any repository, following Karpathy's "LLM Wiki" pattern — a persistent, interlinked markdown knowledge base the agent compiles from the repo's sources and keeps current, instead of re-deriving knowledge on every question. Scaffolds the three layers (raw sources, the wiki, the schema), the index and log, the pending-ingest queue, and optionally runs the first ingest. Use when the user says "/llm-wikify", "wikify this repo", "set up an llm wiki", "add a knowledge base to this repo", "build a wiki for this codebase", or asks for a Karpathy-style LLM wiki. Not for ingesting into a wiki that already exists — the wiki's own schema file owns that workflow.
browser-test-assist
Set up and run browser test sessions through the chrome-devtools MCP in an isolated test profile, and capture the context of a run (steps, console, network, screenshots, a11y snapshot, perf trace, Lighthouse) into a durable report. Use when the user says "/browser-test-assist", "browser test this", "repro this in the browser", "capture browser context", "check the page in Chrome", "run lighthouse on", "trace this page", "seed the test profile", or asks to verify a frontend change in a real browser. Not for driving the user's personal Chrome session.
ml-paper-writing
Write publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. For systems venues (OSDI, NSDI, ASPLOS, SOSP), use systems-paper-writing instead.
browse
Serve and view Claude artifacts (plans, reviews, goals, memory, pull-requests) in the browser with rendered markdown and live reload. Use this skill when the user says "/browse", "/view", "browse my artifacts", "browse my plans", "view my reviews", "open my goals", or wants to preview any file or directory from ~/.claude or ~/project-plans or ~/pr-reviews in a local web server. Also trigger when the user wants to serve a local directory or file for quick browser viewing.
miles-rl-training
Provides guidance for enterprise-grade RL training using miles, a production-ready fork of slime. Use when training large MoE models with FP8/INT4, needing train-inference alignment, or requiring speculative RL for maximum throughput.
training-llms-megatron
Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.
mamba-architecture
State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware design. Mamba-1 (d_state=16) and Mamba-2 (d_state=128, multi-head). Models 130M-2.8B on HuggingFace.
long-context
Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques. Use when processing long documents (32k-128k+ tokens), extending pre-trained models beyond original context limits, or implementing efficient positional encodings. Covers rotary embeddings, attention biases, interpolation methods, and extrapolation strategies for LLMs.
evaluating-llms-harness
Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.
llava
Large Language and Vision Assistant. Enables visual instruction tuning and image-based conversations. Combines CLIP vision encoder with Vicuna/LLaMA language models. Supports multi-turn image chat, visual question answering, and instruction following. Use for vision-language chatbots or image understanding tasks. Best for conversational image analysis.
debug-prod
暂无简介