Skills
97,222 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
Implement Terraform provider ephemeral resources with the Plugin Framework: the Open/Renew/Close lifecycle, ephemeral schema design, registration via EphemeralResources, renewal for expiring credentials, and how ephemeral values flow into write-only attributes and provider configuration. Use when adding an ephemeral resource, exposing secrets/tokens/certificates that must never persist in state or plan, deciding between an ephemeral resource and a data source, or wiring short-lived credentials from one provider into another.
Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.
Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques. Use when optimizing inference speed (1.5-3.6× speedup), reducing latency for real-time applications, or deploying models with limited compute. Covers draft models, tree-based attention, Jacobi iteration, parallel token generation, and production deployment strategies.
Build non-custodial wallets where end users retain control of their private keys via Circle's user-controlled wallets SDK. Supports Google, Apple, Facebook social login, email OTP, and PIN authentication with MPC-based key management. Covers wallet creation, token transfers, message signing, smart contract execution, and wallet management. Triggers on: user-controlled wallets, social login wallet, email OTP wallet, PIN wallet, w3s-pw-web-sdk, executeChallenge, MPC wallet, userToken, deviceToken, contract execution
Best practices for Zustand state management in React and TypeScript applications, covering store design, selectors, middleware, SSR, and testing. Use when creating or refactoring Zustand stores, deciding whether state belongs in Zustand vs component state vs a server-state library, optimizing selector re-renders, adding persist/devtools/immer middleware, or handling Zustand with SSR/React Server Components.
GDB debugger skill for C/C++ programs. Use when starting a GDB session, setting breakpoints, stepping through code, inspecting variables, debugging crashes, using reverse debugging (record/replay), remote debugging with gdbserver, or loading core dumps. Activates on queries about GDB commands, segfaults, hangs, watchpoints, conditional breakpoints, pretty-printers, Python GDB scripting, or multi-threaded debugging.
Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.
Migrate Terraform provider resources and data sources from Plugin SDKv2 to the Plugin Framework: muxing both plugins in one provider (terraform-plugin-mux, tf5to6server), per-resource migration workflow, SDKv2-to-Framework schema mapping (ForceNew, ValidateFunc, DiffSuppressFunc, Default, Timeouts, blocks), null-vs-zero-value behavioral traps, and state-compatibility verification. Use when converting or translating SDKv2 resources to the Framework, setting up a muxed provider server, deciding whether a resource should be migrated at all, or debugging plan diffs and state errors that appeared after a migration.
Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework. Use when conducting causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior.
Language-independent tokenizer treating text as raw Unicode. Supports BPE and Unigram algorithms. Fast (50k sentences/sec), lightweight (6MB memory), deterministic vocabulary. Used by T5, ALBERT, XLNet, mBART. Train on raw text without pre-tokenization. Use when you need multilingual support, CJK languages, or reproducible tokenization.
Serve a model with MAX's `max serve` command: set up the environment (pixi or uv with the max-nightly conda channel / nightly wheel index), point the server at a Hugging Face repo or local checkpoint, target a custom architecture with `--custom-architectures`, and pick the right serve flags for the model. Use this whenever the user wants to run, launch, start, or host a model on MAX, bring up an OpenAI-compatible endpoint, serve a custom/ported architecture, debug a `max serve` startup failure, or figure out which serve flags (devices, quantization-encoding, max-length, task, trust-remote-code) a given model needs, even if they don't say "max serve" by name.
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.
This skill is for enhancing the visual design and aesthetics of Next.js web applications. It provides modern UI components, design patterns, color palettes, animations, and layout templates. Use it to improve styling, create responsive designs, implement modern UI patterns, add animations, select color schemes, or build attractive frontend interfaces
Ships pull requests to open-source projects you don't maintain — CONTRIBUTING compliance, PR-size/scope checks, minimal-diff implementation, AI-disclosure PR descriptions, and rebase or bot-review replies. Use for 提 PR / 上游 PR / 贡献代码, “open a PR,” “fix this upstream,” “rebase against main,” or “respond to the bot review.” Not for reviewing PRs as maintainer (use github-review-pr) or general GitHub CRUD (use github-ops).
Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.
Framework for building LLM-powered applications with agents, chains, and RAG. Supports multiple providers (OpenAI, Anthropic, Google), 500+ integrations, ReAct agents, tool calling, memory management, and vector store retrieval. Use for building chatbots, question-answering systems, autonomous agents, or RAG applications. Best for rapid prototyping and production deployments.
Data framework for building LLM applications with RAG. Specializes in document ingestion (300+ connectors), indexing, and querying. Features vector indices, query engines, agents, and multi-modal support. Use for document Q&A, chatbots, knowledge retrieval, or building RAG pipelines. Best for data-centric LLM applications.
Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.
Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.
Sets up and uses Tailwind CSS v4 in Capacitor apps - CSS-first config (`@import "tailwindcss"`, `@theme`, `@utility`, `@custom-variant`, `@source`, `@plugin`), `@tailwindcss/vite` / `@tailwindcss/postcss`, upgrading from v3 `tailwind.config.js` with `npx @tailwindcss/upgrade`, WebView browser floor (Safari 16.4+ / Chrome 111+ vs Capacitor iOS 15/16 targets), safe-area utilities built on `env(safe-area-inset-*)`, platform variants (`ios:`/`android:` via a root class), class-based dark mode, `pointer-coarse:` touch targets, and hover-on-touch behavior. Use when installing Tailwind in a Capacitor project, styles do not apply after upgrading to v4, colors look wrong on older iPhones, or building mobile layouts with Tailwind. Do not use for Konsta UI components (konsta-ui), Ionic theming (ionic-design), native system bar / edge-to-edge config (safe-area-handling), or keyboard overlap (capacitor-keyboard).
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.
Comprehensive SAP ABAP CDS (Core Data Services) reference for data modeling, view development, and semantic enrichment. Use when creating CDS views or view entities, defining data models with annotations, working with associations and cardinality, implementing input parameters, using built-in functions, writing CASE expressions, implementing access control with DCL, handling CURR/QUAN data types, troubleshooting CDS errors, querying CDS views from ABAP, or displaying data with SALV IDA. Covers ABAP 7.4+ through ABAP Cloud.
Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.
Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.
Review an existing game codebase for architecture, performance, and best practices. Use when the user says "review my game", "code review", "check my game architecture", "is my game well structured", or "audit my game code". Do NOT use for making changes — this is read-only analysis. Use improve-game to implement fixes.
Skyline WXSS style support skill. Provides information on the CSS properties, values, and limitations supported by Skyline. Trigger keywords: WXSS, CSS support, style compatibility, property support, unsupported properties, display grid, position sticky, overflow, filter, gradient, linear-gradient, radial-gradient, conic-gradient, border-style, white-space, font-weight, text-decoration, text-overflow, aspect-ratio, will-change, calc, env, mask, mask-image, background-image, transition, animation, transform, font-feature-settings.
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.
Reduce LLM size and accelerate inference using pruning techniques like Wanda and SparseGPT. Use when compressing models without retraining, achieving 50% sparsity with minimal accuracy loss, or enabling faster inference on hardware accelerators. Covers unstructured pruning, structured pruning, N:M sparsity, magnitude pruning, and one-shot methods.
Rust language reference. Also known as: rustlang, rs (extension), Rust-lang. TRPL (The Rust Programming Language), Rust by Example, Reference, Cargo Book, rustc Book, rustdoc Book, Rustonomicon (Unsafe Rust), Edition Guide, Standard Library, Error Index. Ownership, lifetimes, traits, async / await, error handling, macros.
Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.
Write, rewrite, review, and organize developer-facing documentation for web software projects. Use when creating or improving README files, docs homepages, quickstarts, tutorials, how-to guides, API/reference pages, conceptual explanations, migration guides, or troubleshooting content for frontend, backend, full-stack, SDK, API, or framework-based web products. This skill applies strong information architecture, task-first page structure, clear voice, runnable examples, version and prerequisite hygiene, accessibility rules, and docs-as-code maintenance habits. Do not use it for marketing copy, legal text, or non-technical customer-support articles.
Builds human-in-the-loop (HITL) Airflow workflows: approval gates, form input, and human-driven branching. Use when a DAG needs a human in the loop—an approval or rejection step, sign-off before a task runs, a decision or approval UI, branching on a human choice, or collecting form input mid-run; also for mentions of ApprovalOperator, HITLOperator, HITLBranchOperator, HITLEntryOperator, or HITLTrigger. Requires Airflow 3.1+. Not for AI/LLM task calls (see migrating-ai-sdk-to-common-ai).
Provides guidance for interpreting and manipulating neural network internals using nnsight with optional NDIF remote execution. Use when needing to run interpretability experiments on massive models (70B+) without local GPU resources, or when working with any PyTorch architecture.
CloudBase Node SDK auth guide for server-side identity, user lookup, and custom login tickets. This skill should be used when Node.js code must read caller identity, inspect end users, or bridge an existing user system into CloudBase; not when configuring providers or building client login UI.
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.