Skills
100,928 packaged capabilities and playbooks for agents. Install any of them with one command.
All
When the user wants to check Local Services Ads rankings, see who's ranking in LSA results, monitor LSA competitive landscape, or track LSA ranking changes. Trigger on "LSA rankings," "Local Services Ads," "Google Guaranteed," "Google Screened," "who's ranking in LSAs," or "LSA competitors."
The lyt series skill for follow-up questioning on cross-border e-commerce problems. Handles only business problems related to cross-border e-commerce, TikTok Shop, stores, products, product cards, short videos, livestreaming, ads, GMV Max, data, content, creators, logistics, supply chain and compliance; reject non-business and non-e-commerce questions. When the user raises vague questions such as "why no orders", "why no traffic", "why is the ad not spending", "why the sudden drop", or "why does a viral video get no orders", first examine the concept, implicit assumptions, causal relationships, factual premises and information sufficiency, output only the foremost blocking layer, then ask for the minimum evidence. Triggers: /lyt-problem-clarifier, /问题追问, /追问定位, /不出单追问
Switch/reveal images through an organic ink-splatter-style mask (ink-splatter/blob, `mask-image` radial-gradient driven by scroll progress) or a pinned split-screen sticky layout (image fixed on the left, text scrolling on the right, swapping images through the mask as the active panel changes). The original verbatim versions (Ink Transition, Pinned Split-Screen…) are in the repo Rheinmir/uiux-asset; assets/ is a light vanilla CSS mask version, no WebGL. Call when the user says "mask reveal", "ink transition", "chuyển ảnh qua mask", "split-screen sticky pin", "blob reveal effect", "vết mực loang", "pin ảnh cố định cuộn chữ", or /mask-reveal-transition. DIFFERENT from `skills/scroll-effects` (simple opacity/translate reveals, NO shaped mask) and `skills/blur` (heavy GPU WebGL shader for full-screen image transitions — this is light CSS `mask-image`, no Three.js/canvas needed).
A skill for judging the separation of responsibilities among agent / workflow / tool / memory / state, and the boundary between what is left to the LLM and what is deterministic code, when designing, implementing, or refactoring AI services that use Mastra as their runtime. Use it for "design it with Mastra", "create a workflow", "what should state hold", or "do we need memory or a workspace", and apply single-agent first and minimal workflow. For checking, validating, or migrating to the current API specification, use mastra-framework-guide.
Optimize RF PCB dimensions for bandwidth, return loss, or area via patternsearch and surrogateopt with constraints. TRIGGER: user asks to optimize an RF PCB component for performance (bandwidth, return loss, insertion loss, area) or apply constraints to a design. Invoke BEFORE writing optimization code — RF PCB Toolbox has a built-in optimize() function that differs from generic fmincon/ga approaches. SKIP: designing a component from scratch without an optimization objective (use the specific matlab-design-pcb-* skill), EM analysis without optimization (use matlab-analyze-em), material/stackup setup only (use matlab-manage-pcb-material).
Generate and run MATLAB unit tests using matlab.unittest and matlab.uitest. Parameterized tests, fixtures, mocking, coverage analysis, CI/CD with buildtool, app testing with gestures. Use when creating tests, writing test classes, running test suites, checking coverage, testing apps, or validating MATLAB code.
Guidance for dependency injection in .NET MAUI apps — service registration, lifetime selection (Singleton/Transient/Scoped), constructor injection, automatic resolution via Shell navigation, explicit resolution patterns, platform-specific registrations, and testability best practices. USE FOR: "dependency injection", "DI registration", "AddSingleton", "AddTransient", "AddScoped", "service registration", "constructor injection", "IServiceProvider", "MauiProgram DI", "register services". DO NOT USE FOR: data binding (use maui-data-binding), Shell route setup (use maui-shell-navigation), or unit test mocking patterns (use maui-unit-testing).
Provides guidance for Google MediaPipe Pose Landmarker on web using @mediapipe/tasks-vision. Covers setup, landmark indices, running modes, and real-time video patterns. Use when working with MediaPipe, pose detection, body landmarks, or @mediapipe/tasks-vision.
Initialize the memory system in the current directory, generating CLAUDE.md (optionally AGENT.md for Cursor), MEMORY.md and the memory/ directory. Triggered when the user says "initialize memory", "set up memory", "memory init" or "/memory-init".
Use Mergify merge protections to control when PRs merge — PR dependencies (Depends-On header), delayed merges (Merge-After header), and scheduled freezes (CLI). ALWAYS use this skill when managing Mergify scheduled freezes, or when blocking a PR on another PR or delaying its merge with Mergify's Depends-On or Merge-After headers. Triggers on mergify freeze, Mergify scheduled freeze, Mergify merge freeze, Depends-On, Merge-After, Mergify merge protection.
Create evidence-based Meta ad creative briefs, copy angles, and testable asset concepts for Facebook and Instagram. Use when asked for Meta ad creative, Facebook ads copy, Instagram ad assets, UGC concepts, creative briefs, new ad concepts, or creative refreshes.
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.
A persistent, compounding knowledge base combining Karpathy's LLM Wiki pattern with a structured knowledge graph. Ingest documents once — the LLM writes wiki pages, extracts entities/relations into the graph, and stores raw content for evidence retrieval. Knowledge accumulates and cross-references; it is never re-derived from scratch.
Helps DevOps engineers configure mirrord Operator's Kafka queue splitting feature end-to-end. Generates the MirrordSplitConfig and MirrordPropertyList Kubernetes CRD YAMLs (the current resources; MirrordKafkaTopicsConsumer + MirrordKafkaClientConfig are deprecated but still supported), the matching mirrord.json split_queues section with message_filter and jq_filter, and Helm value guidance. Use this skill whenever the user mentions Kafka splitting with mirrord, MirrordSplitConfig, MirrordPropertyList, MirrordKafkaClientConfig, MirrordKafkaTopicsConsumer, Kafka queue/topic splitting, configuring mirrord with Kafka, Kafka Streams splitting, MSK IAM auth, or troubleshooting Kafka splitting sessions. Also trigger on split_queues with queue_type Kafka, or connecting mirrord to a Kafka cluster. This is a Team/Enterprise feature of mirrord.
Always invoke for training, validating, tuning, benchmarking, or claiming readiness of a predictive model. Covers leakage audits, spatial and grouped splits, metrics, reproducibility, and honest reporting. Invoke especially when spatial dependence, split design, or deployment geography is unknown; uncertainty is a reason to use this skill. Do not trigger for descriptive EDA or non-predictive statistical inference.
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. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.
Build end-to-end MLOps pipelines from data preparation through model training, validation, and production deployment. Use when creating ML pipelines, implementing MLOps practices, or automating model training and deployment workflows.
Provides shadcn/ui component library expertise for React applications with shadcn/cli v4, preset system, RTL support, unified radix-ui, and OKLCH theming. Use when implementing UI components, design systems, CLI workflows, or registry authoring with shadcn/ui.
Modal.com serverless compute, GPU, autoscaling, batching, and deployment knowledge. PROACTIVELY activate for: (1) Modal app/function definitions (modal.App, @app.function, @app.cls), (2) running ML/AI workloads serverlessly, (3) GPU configuration (T4, L4, A10G, A100, H100, H200), (4) Modal volumes and shared state, (5) Modal Image builder (apt, pip, run_commands), (6) Modal secrets and environment management, (7) web endpoints (FastAPI, asgi_app, wsgi_app), (8) cron schedules and continuous deployments, (9) Modal sandboxes, (10) cost optimization on Modal (autoscaling, spot, idle timeout), (11) Modal CLI (modal deploy, modal run, modal serve). Provides: app templates, GPU sizing matrix, Image builder recipes, secret patterns, web endpoint examples, and cost-optimization guidance.
Generate high-quality AI images from text prompts or transform existing images using ModelsLab's v7 API with 50,000+ models including FLUX, Realtime, and Community models. Supports text-to-image, image-to-image, inpainting, and ControlNet.
Cut a feature wishlist down to the smallest MVP that delivers value. Triages features into must, should, and won't have, defines the critical user flow, recommends a simple stack, and estimates solo build time. Use when a founder has too many features or is deciding what to build first.
MySQL relational database. Covers queries, indexes, and optimization. Use when working with MySQL databases. USE WHEN: user mentions "mysql", "mariadb", asks about "AUTO_INCREMENT", "ON DUPLICATE KEY UPDATE", "GROUP_CONCAT", "mysql specific syntax" DO NOT USE FOR: PostgreSQL - use `postgresql` instead, MongoDB - use `mongodb` instead, Oracle - use `oracle` instead, SQL Server - use `sqlserver` instead
GPU-accelerated data curation for LLM training. Supports text/image/video/audio. Features fuzzy deduplication (16× faster), quality filtering (30+ heuristics), semantic deduplication, PII redaction, NSFW detection. Scales across GPUs with RAPIDS. Use for preparing high-quality training datasets, cleaning web data, or deduplicating large corpora.
Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.
Use when the user explicitly asks for Nerd Loop or wants a cost-proportional, behavior-aware task-completion controller that selects the cheapest adequate loop profile and iterates toward an explicit Definition of Done through focused work, automatic verification, and bounded stopping rules.
Complete Next.js App Router fundamentals system (Next.js 15.5/16). PROACTIVELY activate for: (1) App directory structure and conventions, (2) Layouts and templates, (3) Loading and error boundaries, (4) Route groups and organization, (5) Parallel routes setup, (6) Metadata and SEO, (7) Server vs Client Components, (8) Navigation and View Transitions, (9) Async params pattern (Next.js 16 breaking change). Provides: File conventions, layout nesting, streaming UI, error handling, metadata templates, View Transitions API. Ensures correct App Router architecture with proper component boundaries.
Complete Next.js authentication system. PROACTIVELY activate for: (1) NextAuth.js (Auth.js) setup, (2) OAuth providers (GitHub, Google), (3) Credentials provider, (4) Session management (JWT/database), (5) Protected routes with middleware, (6) Role-based access control (RBAC), (7) Login/registration forms, (8) Authorization patterns, (9) Type augmentation for sessions. Provides: Auth.js configuration, middleware protection, session hooks, RBAC patterns, login forms. Ensures secure authentication with proper session handling.
Modal.com integration patterns for Next.js applications. PROACTIVELY activate for: (1) Next.js + Modal backend setup, (2) AI inference from Next.js (LLMs, image generation), (3) Video/audio processing backends, (4) Heavy compute offloading from Vercel, (5) GPU workloads for Next.js apps, (6) Webhook integration between Next.js and Modal, (7) File upload processing, (8) Background job processing, (9) Serverless AI API endpoints, (10) Next.js + Modal authentication patterns. Provides: Architecture patterns, API route integration, webhook handling, file upload workflows, CORS configuration, warm container patterns, streaming responses, and production-ready examples.
NVIDIA Sync desktop app. Manage remote connections to DGX Spark, DGX Station, and SSH Linux hosts. Features mDNS auto-discovery, manual IP, SSH config import, Tailscale, and automatic network switching. Includes Cluster Assistant, ConnectX-7 cluster topology, netplan validation, Resource Monitor (CPU / GPU / memory), and release notes. Launch VS Code, VS Code Insiders, Cursor, NVIDIA AI Workbench, and custom apps; supports port forwarding and SSH tunnels.
Upgrade an ObjectStack metadata project across a protocol major — run the deterministic conversion chain, then work the semantic residue the chain cannot express (intent choices, custom code on retired APIs, stale prose) to a decision with the project's owner, and finish with a green `validate` plus a human-readable upgrade report. Use when a project is on an older protocol major and must move to the current one, when `@objectstack/spec` was bumped across a major and metadata or code stopped parsing, when a parse or `tsc` error quotes a `[REMOVED]` prescription, or when asked to "upgrade to v17" / "升级到 v17" / "一键升级元数据项目". Do not use to author new metadata (the domain skills cover that), or to reconcile physical database drift (that is objectstack-platform's `os migrate plan` / `os migrate apply`).
Observability as an engineering discipline — wide events / canonical log lines, OpenTelemetry instrumentation (traces, metrics, context propagation, sampling, Collector), SLIs/SLOs/error budgets, symptom-based alerting with burn rates, telemetry hygiene, and testing instrumentation as behavior. Use when calls between services cannot be correlated, another team cannot line their side of a call up with ours or find the request it came from, a request or trace id has to cross a service hop, a failure cannot be debugged from the logs, or when instrumenting a service, designing SLOs or alerts, choosing what to log/trace/measure, investigating production unknowns, or reviewing telemetry cost and cardinality. For log transport and shape (stdout, JSON, levels, timestamps) see twelve-factor; for CI failure diagnosis see ci-debugging; for where instrumentation lives in ports-and-adapters codebases see hexagonal-architecture; for environment drift see production-parity-skill-builder; for HTTP error response shape see api-design.
Deploy self-hosted OAuth2 and OIDC servers with OpenAuth, createSubjects schemas, prebuilt PasswordUI/CodeUI, and cross-framework client verification. Use when building universal auth for Cloudflare Workers, Node.js, Bun, and AWS Lambda with @openauthjs/openauth, social logins, and secure session rotation.
Use when the user asks to review code/a PR with opencode or with a non-Claude model (e.g., "review with opencode", "review using gpt-5.6-sol", "/opencode-review"). Runs a fresh opencode session in a subagent, tells it to use the review-code skill, and relays its review.
Guide to using the OpenDesign component library. Use this skill when you need to quickly build pages with the OpenDesign Vue component library. Supports all OpenDesign components (46), including common UI components such as buttons, forms, tables, dialogs, cards, icons, sliders, step bars, and toasts. Use cases: (1) building Vue pages with OpenDesign components, (2) looking up component usage and property descriptions, (3) getting component code examples
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.