Skills
96,981 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
Use this skill for reinforcement learning tasks including training RL agents (PPO, SAC, DQN, TD3, DDPG, A2C, etc.), creating custom Gym environments, implementing callbacks for monitoring and control, using vectorized environments for parallel training, and integrating with deep RL workflows. This skill should be used when users request RL algorithm implementation, agent training, environment design, or RL experimentation
Iteratively build multi-page websites using Stitch. Reads next-prompt.md (the baton), generates the next page with Stitch MCP, integrates it into the site, then updates next-prompt.md to continue the loop. Works with stitch-design-md and stitch-ui-prompt-architect for consistent multi-page output.
Build, review, debug, configure, migrate, or plan Storybook 10 UI workshops with current docs. Use for storybook, create storybook, CSF3, CSF Next, Meta, StoryObj, definePreview, preview.meta, args, argTypes, decorators, parameters, tags, autodocs, play functions, storybook/test, @storybook/react-vite, @storybook/nextjs-vite, @storybook/addon-docs, @storybook/addon-a11y, @storybook/addon-vitest, portable stories, composeStories, main.ts, preview.ts, ESM config, Chromatic, @storybook/addon-mcp, componentsManifest, storybook skills/tools agent CLI, and upgrades from Storybook 8/9 (10.6.x current).
Expert knowledge for Supabase database, RLS, and backend patternsUse when "supabase, row level security, rls, postgres, database policy, supabase storage, supabase realtime, supabase, postgres, rls, database, backend, storage, realtime" mentioned.
Write, review, or improve SwiftUI code following best practices for state management, view composition, performance, macOS-specific APIs, and iOS 26+ Liquid Glass adoption. Use when building new SwiftUI features, refactoring existing views, reviewing code quality, or adopting modern SwiftUI patterns.
Implement Calendar, DatePicker, DateRangePicker, DateTimePicker, and TimePicker components using Syncfusion JavaScript (EJ2) calendars library. Covers date/time selection, formatting, validation, localization, masking, accessibility, and advanced customization patterns.
Comprehensive guide for implementing Syncfusion EJ2 JavaScript Notification components including Message, Skeleton, and Toast. Use this when displaying contextual messages with severity indicators (Normal, Success, Info, Warning, Error) and display variants (Text, Outlined, Filled); creating animated loading placeholders with shapes (Circle, Square, Rectangle, Text) and shimmer effects (Wave, Pulse, Fade); configuring toast notifications with positioning, animations, templates, and dismissal patterns; or customizing notification appearance and accessibility.
TanStack Query (React Query) v5 best practices for data fetching, caching, mutations, and server state management. Use when building data-driven React applications, setting up query configurations, implementing mutations/optimistic updates, configuring caching strategies, integrating with SSR, or fixing v4→v5 migration errors.
Enforce portable TanStack Start app architecture: file-route trees, page folders with index.tsx, route colocation (-components/-hooks/-lib), reuse-gated src/modules, naming, and import boundaries. Use when scaffolding a Start app or feature, adding pages/layouts, reviewing or reorganizing route/module layout, or when the user asks for tanstack-start-architecture / Start house style. Optional with-* overlays for generated OpenAPI clients, Better Auth client, TanStack Query, route gates, TanStack Form, env, client authorization, UI package, Effect, and Vitest.
Cosmos-Embed1 video-text embedding for text-to-video retrieval, video-to-video search, semantic deduplication, and fine-tuning. Use when the user asks to "fine-tune Cosmos-Embed1", "run cosmos-embed inference", "export Cosmos-Embed1", "embed videos", or "search videos with text".
Fine-tune any HuggingFace CV / VLM / LLM model on local NVIDIA GPUs inside an NGC PyTorch container. Use when the user wants to fine-tune a HuggingFace model (full or LoRA), train a vision / VLM / LLM model end-to-end, generate a reproducible HF training pipeline, smoke-test a HuggingFace model locally before scale-up, push a fine-tuned model to the HF Hub with a model card, or emit a self-contained rerun skill for an existing HuggingFace finetune. Supports image classification, object detection, semantic / instance / panoptic segmentation, depth estimation, image-text-to-text VLM (SFT / LoRA), and LLM SFT / DPO / GRPO. Six-step workflow: inspect and qualify, hardware and NGC image, research, generate and smoke, train + eval + infer, push and emit rerun skill.
Integrate a HuggingFace Computer Vision model into the NVIDIA TAO Toolkit ecosystem (tao-core config, tao-pytorch trainer, tao-deploy TensorRT pipeline). Use when the user asks to “integrate a HuggingFace model into TAO,” “add an HF model to TAO Toolkit,” “wire a HuggingFace ViT/DETR/ SegFormer into tao-pytorch,” “build a TAO trainer + deploy pipeline for an HF CV model,” or pastes a HuggingFace model URL/ID and wants it turned into a TAO model. Covers the full 7-phase loop: prerequisites check, HuggingFace inspection and validation, codebase exploration, tao-core configuration and native trainer implementation, ONNX export plus TensorRT deploy integration, packaging and L0 testing, container-based end-to-end validation, and (conditional) accuracy/latency tuning. Supports classification, object detection, semantic / instance / panoptic segmentation, zero-shot detection, and depth estimation.
Run the canonical NVIDIA AOI three-phase training pipeline — Phase 1 AutoML baseline (HPO), Phase 2 DEFT loop (RCA → SDG → mining → plain-train retrain), Phase 3 AutoML refinement on the DEFT-augmented dataset. This is the default entry point for any "run the AOI workflow", "fine-tune my PCB AOI model end-to-end", "improve my AOI ChangeNet model", or "AOI workflow with AutoML" request — route here instead of tao-run-deft-aoi directly unless the user explicitly asks for the DEFT loop ONLY (e.g. "run JUST the DEFT loop", "skip AutoML, only DEFT"). Also handles the same three-phase pattern for non-AOI DEFT applications — AutoML baseline then DEFT loop warm-started from AutoML's winning HPs then post-DEFT AutoML refinement on the iteration-augmented dataset. Trigger phrases include "run the AOI workflow", "AOI end-to-end", "AutoML + DEFT", "AutoML then DEFT", "tune hyperparameters then DEFT", "DEFT with AutoML at both ends", "warm-start DEFT", "improve my AOI model".
Start, query, and stop a network-specific TAO inference microservice ({network_arch}-inference-microservice) by delegating container execution to the appropriate platform skill. Handles container image resolution, job-payload JSON construction, and the service registry. Use when the user wants to run inference on a TAO model checkpoint using a microservice container, deploy a TAO inference endpoint, or stop a running inference container.
Tavily AI search API for LLM applications: web search, content extraction, site crawling, mapping, and research. Use when integrating real-time web search into LLM apps, extracting content from URLs, crawling sites, or building RAG pipelines with web data. Keywords: Tavily, AI search, RAG, web search API, LLM search, extract, crawl, map, research, tavily-python.
Use when establishing TDD as a discipline — phrases like "red green refactor", "TDD cycle", "test first", "should I do TDD here", "TDD discipline", "what's the TDD loop", "when should I refactor". Defines the canonical TDD cycle (red → green → refactor), what each phase actually requires, the honest scope of when TDD applies, and the cases where it shouldn't.
TeamLead orchestration — use when a Human wants coordinated implementation/delegation for product code, tests, config, infra, features, bug fixes, refactoring, or behavior/UI/API changes. Do not trigger for read-only inspection/status/explanation/lightweight summary/no-edit review/research unless the Human explicitly requests TeamLead orchestration or delegation would clearly help. Work that creates or modifies behavior must pass the Research Gate (spawn research agents immediately after classification) and always invoke `superpowers:brainstorming` first, then the Lead performs the Spec/Plan gates before delegating; the Lead must not write product code themselves. Every spawn must pick a spawn mode (one-shot Agent vs teammate) and specify the model per the Model Policy (Fable only for thinking/planning work, Opus for Dev/QA/research). Execute in batches per the Commit Plan, run tests per the fast/heavy policy (full suite + E2E only once at the end), Human checkpoints only at 3 points, and every prompt includes the repo's Convention Anchor.
Use personal preferences and project evidence to make trade-offs on tech stack, frameworks, tools, data formats or repository boundaries, and set up the Python toolchain (uv, ruff, ty, pyproject, prek/CI). Use when the main outcome is "what to choose, why it fits, when to deviate from the baseline", or when the user asks to create/migrate Python engineering configuration. Do not trigger merely because the request contains "architecture" or "refactoring": the project's next step, module/interface/state boundaries and multi-phase rollout are the responsibility of pilot; this skill only handles the technical choices that genuinely exist within them.
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.