Skills
97,905 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
Build, review, debug, migrate, or plan UI animation with Motion for React (motion/react, motion@13). Use for React/Next/Vite: motion.div, AnimatePresence, variants, gestures, drag, layout/layoutId, LazyMotion, useScroll, useAnimate, MotionValues, animateView, AnimateView (motion/react-animate-view, React 19.3), arc(), spring bounce, Reorder grids, reducedMotion, framer-motion→motion, Motion 13 isValidProp, CSS/WAAPI/View Transitions, microinteractions, page transitions, accessibility, and performance-safe animation. Do not use for Vue, vanilla-only Motion, React Native, or non-interactive film/video unless adapting it to product UI.
Provides nanoGPT, Karpathy's minimal PyTorch GPT implementation (model.py and train.py), with workflows for training character-level Shakespeare on CPU, reproducing GPT-2 124M on OpenWebText with multi-GPU torchrun, fine-tuning pretrained GPT-2 checkpoints, and training on custom text. Use when learning how GPT and transformer blocks work from scratch. Use when running a small training experiment on CPU or a single GPU. Use when modifying a transformer variant in plain PyTorch. Use when preparing character-level or BPE binary datasets. Use when troubleshooting out-of-memory or slow training in nanoGPT. Not for production or large-scale distributed training; use HuggingFace Transformers or Megatron-LM instead.
How to write NetBox custom scripts for automation, data validation, and bulk operations. Use when building scripts that run inside NetBox — covers the Script class, form variables, ORM access, logging, transactions, scheduling, and common patterns.
Require `neverthrow`-based return types in TypeScript and JavaScript code whenever the surrounding technology allows it. Use when creating, refactoring, reviewing, or extending standalone functions, exported module functions, class methods, object methods, service methods, repository methods, and similar APIs that should expose explicit success and failure result types in their signatures. Prefer `Result<T, E>` for synchronous code and `ResultAsync<T, E>` for asynchronous code. Only skip a `neverthrow` return type when a framework, library, runtime interface, or externally imposed contract is incompatible and requires a different return shape.
Use for Next.js 16 Cache Components and runtime rendering work involving `use cache`, `cacheLife`, `cacheTag`, `updateTag`, Suspense, runtime APIs, Partial Prerendering, remote cache, metadata/viewport runtime access, or migration from older route segment caching configs.
React and Next.js performance optimization from Vercel Engineering. Use when building React components, optimizing performance, eliminating waterfalls, reducing bundle size, reviewing code for performance issues, or implementing server/client-side optimizations.
Install and compose Nexus UI components for AI chat UIs — prompt input, model selector, suggestions, attachments, message, thread, citation, reasoning, text shimmer, image, feedback bar, toaster, chain of thought, tool, questions, and AI SDK patterns. Activates for @nexus-ui registry usage or Nexus UI source under components/nexus-ui.
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.
(NS) Write and edit human-facing project documentation — README, docs/ guides, runbooks, and contributor-facing markdown. Use whenever the user asks for documentation, README updates, how-to guides, or docs/ content — even if they say "explain how to use" instead of "write docs". Stack-agnostic; follow project layout from AGENTS.md. Do NOT use for code comments, API docblocks in source, requirements.md, or SDD version artifacts (use PM skills).
(NS) Distinctive production UI — layout, typography, motion, and component polish that avoids generic AI-slop aesthetics. Use whenever the user builds or refines pages, components, dashboards, forms, or design-brief work, or asks for better UI/UX — even if they do not say "design". Load docs/context/design-brief.md when present. Do NOT use for backend-only work, requirements writing, or full SDD orchestration (use ns-spec-driven).
(NS) Root-cause diagnosis from errors, logs, stack traces, failing tests, or CI — entry priority 4 when the user wants diagnosis WITHOUT implement ("why does X break?"). Use proactively on unclear test/CI failures if they have not asked to fix yet. Do NOT use for implement/fix (ns-code-coder), version scope (ns-spec-driven), GitLab ISSUE_URL execution, or pre-merge review (ns-code-reviewer).
Use when the user asks to "plan & implement", "plan this work", "research & build", "plan auth/API/work", or needs a multi-step pipeline from understanding through implementation. Flow is Understand → Research → Plan → Implement → Verify. For design documents or technical proposals without implementation, use octocode-rfc-generator instead.
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.
Use the Opper CLI (`opper` command, npm package `@opperai/cli`) for terminal work with Opper: signing in, calling functions, inspecting traces, tracking usage and costs, generating images, installing bundled skills, configuring AI code editors, and launching coding agents (Claude Code, OpenCode, Codex, Hermes, Pi) with their inference routed through Opper. Use this skill whenever the user mentions the `opper` command, the Opper CLI, opper-ai/cli, `@opperai/cli`, `opper launch`, or explicitly wants shell commands, editor configuration, private local key delivery, or agent launch workflows. Prefer opper-mcp for agent-assisted setup, platform management, and model tests. For exact arguments, always run `opper <subcommand> --help`; the help output is the authoritative reference.
Operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, skill sharing, worktree comments, and Orca's embedded browser through the `orca` CLI. Use when the user says "$orca-cli", "Orca worktree", "child worktree", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "handoff" / "handover" / "give this to another agent", "Orca browser", "orca artifacts", or "share skills". Prefer it over raw git worktree, ad hoc PTYs, or Computer Use when Orca state is involved. Use Computer Use only when a visible window needs GUI control that a CLI, filesystem, or API cannot do.
Operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, skill sharing, worktree comments, and Orca's embedded browser through the `orca` CLI. Use when the user says "$orca-cli", "Orca worktree", "child worktree", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "handoff" / "handover" / "give this to another agent", "Orca browser", "orca artifacts", or "share skills". Prefer it over raw git worktree, ad hoc PTYs, or Computer Use when Orca state is involved. Use Computer Use only when a visible window needs GUI control that a CLI, filesystem, or API cannot do.
Generates guaranteed-valid structured output from LLMs with Outlines (dottxt.ai), constraining token sampling via finite state machines for JSON schemas, Pydantic models, regex, choice lists, and integer/float types, on Transformers, llama.cpp, vLLM, and limited OpenAI backends. Use when you need output that always parses as valid JSON or matches a regex. Use when extracting typed data into Pydantic models from a local model. Use when classifying text into a fixed set of categories. Use when running batch structured generation on vLLM or Transformers. Use when controlling token sampling at the grammar level. Not for API models needing automatic retries (use Instructor).
Load PROACTIVELY when task involves payments, billing, or subscriptions. Use when user says "add payments", "integrate Stripe", "set up subscriptions", "add a checkout flow", or "handle billing webhooks". Covers Stripe, LemonSqueezy, and Paddle integration, checkout sessions, subscription lifecycle management, webhook verification and handling, customer portal, metered billing, refunds, and PCI compliance considerations.
A skill that generates prototype UI based on Mock data without a backend. Vue 3 + TypeScript + NuxtUI v4. It accumulates per-task screens in the modules-task folder of a separate ui-proto repository (e.g. peach-ui-proto-backoffice). Triggered by keywords such as "프로토타입 만들어줘", "Mock 화면", "proto UI", "기획 화면 빠르게", "ui-proto 작업", "기획자 검토용 화면", "태스크 폴더 추가". If a Spec document is provided as input, it automatically switches to team mode (Ralph loop 5/10 times); if a planner works directly without a Spec, it runs in solo mode. If real API integration is needed, use peach-gen-ui.
Open or update a GitHub PR from the current branch, pushing normally when needed. Use when the user invokes pr-branch or requests PR creation or refresh. Follow one value-first template for any project size: explain invisible engineering to non-technical readers, then record technical changes and verification. Default to English unless a language is explicitly specified.
Principal-level architecture, system design, trade-off analysis, ADRs, Clean Architecture, GoF, production-ready microservices, Clean Code, algorithms, domain modeling/DDD, distributed data, API contracts, quality attributes/ATAM, SRE/resilience, secure architecture, legacy evolution, C4/arc42 documentation, and engineering ownership. Use for architecture/code reviews, tech strategy, pattern audits, coding standards, algorithm or data-structure choice, Big O/performance, databases, consistency, events, scalability, SLO/reliability, threat modeling, modernization/migration, bounded contexts, aggregates, context mapping, API versioning/deprecation, team topology, platform engineering, multi-service design, or /principal-software-architect. Produces evidence-based findings and only the specialized baselines material to the decision.
Full-stack development and architecture design assistant, covering scenarios such as "development, implementation, writing code, continuing development, bug fixing, code refactoring, architecture design, technical solution evaluation/code review", suitable for full-stack R&D tasks in new projects and existing code repositories.
Use when the user needs to look up a TOTVS Protheus native function's syntax/parameters/return value, the SX data dictionary structure (SX1-SX9, SIX), REST API endpoint patterns (FWRest, WsRestFul), MV_* system parameter behavior, SX3 field validation, or whether a function is restricted/internal-only. Also triggers on Portuguese phrasing like "funcao nativa", "dicionario de dados", "parametro MV_", "campo SX3", "API REST protheus", "essa funcao e restrita".
High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales from laptop to supercomputer with same code. Use when you want clean training loops with built-in best practices.
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.
Build/deploy an internal Miovision site on Quick — a folder of static files served at <site>.quick.mio.vision, with a zero-config global `quick.*` API for identity, db, realtime, AI, Snowflake, per-user SaaS integrations (Jira/Confluence/Slack/Drive/Salesforce), file uploads, and multi-site deploy. Use for internal Miovision web site/apps/tools/dashboards, or when the user mentions Quick.
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.