Skills
97,725 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.
Automatically activates when over-engineering patterns are detected: (1) Modifying code or files the user did not explicitly ask to change (2) Creating new abstraction layers (class, interface, factory, wrapper) without being asked (3) Adding comments, documentation, JSDoc, or type annotations without being asked (4) Introducing new dependencies without being asked (5) Rewriting entire files instead of making minimal edits (6) Diff scope significantly exceeding the user's request (7) User signals like "too much", "don't change that", "only change X", "keep it simple", "stop" (8) Adding error handling, validation, or defensive code for scenarios that cannot occur (9) Generating tests, configuration scaffolding, or documentation without being asked
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.
Provides file organization conventions for React and Next.js projects. Use when creating new files, components, hooks, utilities, or services. Triggers on questions like "where should this go?", "where do I put this?", or when deciding between colocating vs grouping files.
Retrofitting OpenTelemetry into an existing, uninstrumented application. Trigger phrases: "migrate existing app to OTel", "add OpenTelemetry to existing project", "retrofit OTel into my codebase", "thread context through my code", "context propagation", "bridge Prometheus metrics to OTel", "logging bridge", "migrate logging to OTel", "slog bridge", "logback bridge", "verify my instrumentation", "traces are disconnected", "orphaned spans", "migrate to OpenTelemetry", "OTel migration plan", "how do I sequence an OTel migration", "add tracing to existing code", "refactor for context propagation", "Fiber context gotcha", "keep existing logging working with OTel", "add OTel without breaking Prometheus", "bridge existing metrics", "coexist with existing monitoring". For greenfield instrumentation use otel-instrumentation; for Beelines use beeline-migration.
Build React web UIs with PanelUI's web collection — adding its components from the shadcn registry or from npm, colouring them from the theme, and composing them into framed panels. Applies when working in a React project that has shadcn/ui set up with a @panelui registry or has panelui-web installed, when the user mentions PanelUI on the web, or when they ask for a dotted globe, a KPI card, a kanban board, a node graph, an activity grid or a feedback dialog in React.
Guide for running multiple parallel Claude Code sessions using cw (Claude Worktree Manager). Use when the user wants to parallelize work across multiple Claude instances, manage git worktrees for concurrent AI coding tasks, or merge results back together. Triggers on tasks involving parallel Claude sessions, worktree management, or splitting work across multiple agents.
Explore a large codebase in parallel by launching multiple explore subagents that each investigate a different area simultaneously. Use when onboarding onto a new project, understanding architecture, or investigating a cross-cutting concern.
Systematic performance optimization and regression debugging for OneKey mobile app (iOS). Use when: (1) Fixing performance regressions - when metrics like tokensStartMs, tokensSpanMs, or functionCallCount have regressed and need to be brought back to normal levels, (2) Improving baseline performance - when there's a need to optimize cold start time or reduce function call overhead, (3) User requests performance optimization/improvement/debugging for the app's startup or home screen refresh flow.
Build 2D games with the Phaser 3 framework. Covers scene lifecycle, sprites, physics (Arcade/Matter), tilemaps, animations, input handling, and game architecture. Trigger: "create phaser game", "add phaser scene", "phaser sprite", "phaser physics", "game development with phaser".
Expert in PlantUML diagramming language. Automatically invoke when users need to visualize processes, systems, architectures, workflows, data structures, or relationships between components. This includes when users describe how things connect, request flow documentation, explain system interactions, show lifecycle states, or need to communicate design visually - even if they don't explicitly say "diagram", "PlantUML", or "UML". Covers sequence, class, activity, use case, state, component, object, and flowchart diagrams.
Create premium, Awwwards-quality website designs as React (.jsx) components that look like they were built by a top-tier agency charging $50k+ per project. Use this skill whenever the user asks for a website, landing page, portfolio, or web component that should look expensive, premium, luxury, editorial, high-end, or agency-quality. Also trigger when the user mentions Awwwards, FWA, award-winning design, or references brands like Apple, Aesop, Bottega Veneta, Stripe, or any luxury/fashion/design-forward brand. Trigger when the user says things like "make it look professional", "make it look expensive", "I want it to look really good", "high quality design", "not generic", or expresses dissatisfaction with typical AI-generated website aesthetics. Do NOT trigger for dashboards, admin panels, internal tools, or functional UI where aesthetics are secondary to utility — the standard frontend-design skill handles those.
Apply when designing types, reviewing a function signature, or writing code in any statically-typed language. Make illegal states unrepresentable, brand semantic primitives, parse external data at boundaries, exhaust variants, derive from authoritative schemas.
Use when handling BK-CI project creation, project properties, member permissions, project tags, project sharding, and project resource boundaries. Prefer this when the user mentions project management, the meaning of projectId, project configuration, member management, or project migration.
Use when the user needs to understand, diagnose, or prevent Protheus record locks and deadlocks -- covering RecLock/MsUnlock/SoftLock/LockByName semantics, BeginTran/EndTran transaction scoping, lock leak and deadlock diagnostics (TopMemoStatus, MonitorActiveLocks, DBAccess monitor, SQL Server/PostgreSQL/Oracle lock queries), and prevention patterns (BEGIN SEQUENCE/RECOVER, lock ordering). Also triggers on Portuguese phrasing like "registro travado", "trava de registro", "deadlock", "erro de lock", "travamento no Protheus", "quem esta travando o registro", or "RecLock nao libera".
Use this skill when creating QML view files, designing QML component hierarchies, building layouts, styling QML controls, creating reusable QML components, implementing QML navigation / page switching, or working with QML resources. Covers QML file structure, component patterns, Material/Controls styling, resource management, and common QML idioms for desktop applications.
LLM workloads on open-source Ray (pinned to 2.57) — OpenAI-compatible serving with ray.serve.llm (vLLM-backed LLMConfig + build_openai_app) and batch inference with ray.data.llm (build_processor). Corrects the stale defaults a model produces — the archived ray-llm repo and its YAML configs, hand-rolled vLLM engines inside plain Serve deployments, the removed build_llm_processor name, deprecated boolean stage flags, top-level LLMServer/LLMRouter imports, free-form accelerator strings, one-deployment-per-LoRA-adapter designs — with the 2.57 idioms (stage configs, placement_group_config over hand-rolled PGs, deployment_config autoscaling, dynamic LoRA multiplexing, prefix-cache-affinity routing, the full OpenAI endpoint surface). Use when writing, reviewing, or productionizing LLM serving or batch inference on Ray. Classic-ML Ray (Train/Tune/Data/Serve/clusters) lives in the sibling ray skill.
Functional language runtime reverse engineering (Haskell/OCaml): closure/heap object models, calling conventions, and a dataflow-first strategy. Triggers: Haskell reverse engineering, OCaml reverse engineering, GHC RTS, thunk, STG, OCaml runtime, closure, functional language artifacts.
Java Card / SIM applet reverse engineering: parse CAP file components (12 components per the specification), reconstruct CAP bytecode (Java subset), analyze AID and installation parameters, and trace process(APDU) dispatch. Trigger terms: Java Card, javacard, CAP file, SIM card, USIM, applet, AID, bank card, JCVM
RISC-V architecture reverse engineering: RV32/RV64, compressed instructions (RVC), gp-relative addressing, ABI and ecall system call conventions, and toolchain fingerprinting. Triggers: RISC-V, riscv, RV32, RV64, RVC, compressed instructions, ecall, ESP32-C3, GD32V.
Advanced React patterns and conventions for data fetching, tables, forms, state machines, client state management, schema validation, and testing. Use when tackling complex React problems — not simple component questions, but multi-concern tasks like server-driven tables with filtering, multi-step wizards, eliminating useEffect, Suspense architecture, choosing between state management approaches, or designing data flow across server/client/URL/form state. Do not use for web-specific routing/SSR or React Native-specific navigation/performance.
Build, review, debug, migrate, or plan React applications and components using current React documentation. Use for React, React DOM, hooks, JSX, Server Components, Server Functions, Actions/forms, React Compiler, hydration, rendering roots, React 19/19.3 upgrades, View Transitions, Fragment refs, browser-only rendering, Trusted Types, or questions about modern React patterns and best practices.
Refactor NestJS/TypeScript code to improve maintainability, readability, and adherence to best practices. Identifies and fixes circular dependencies, god object services, fat controllers with business logic, deep nesting, and SRP violations. Applies NestJS patterns including proper module organization, provider scopes, custom decorators, guards, interceptors, pipes, DTOs with class-validator, exception filters, CQRS, repository pattern, and event-driven architecture. Transforms code into exemplary implementations following SOLID principles.
Refactor Spring Boot and Java code to improve maintainability, readability, and adherence to enterprise best practices. This skill transforms messy Spring Boot applications into clean, well-structured solutions following SOLID principles and Spring Boot 3.x conventions. It addresses fat controllers, improper transaction boundaries, field injection anti-patterns, and scattered configuration. Leverages Java 21+ features including record patterns, pattern matching for switch, virtual threads, and sequenced collections.
Performance best practices for Relay applications. Use when optimizing data fetching, reducing re-renders, configuring caching, or improving time to first meaningful paint. Covers query placement, @defer, pagination, fetch policies, garbage collection, fragment granularity, and server-side filtering. Companion to the relay-best-practices skill which covers correctness and architecture.
Summarises the delta between a tool's latest release and the last summary the user saw. Use when the user asks about what's new, the latest release, release notes, or the changelog of one of the supported tools: Claude Code, OpenCode, llama-swap, or llama.cpp.
Build production-grade frontend UI that does not look AI-made or borrowed: identity from the client's world, references, tokens, interaction, a11y/SEO, slop audit. Use when building any page, component or dashboard; when the owner says the result looks generic; or when a product needs a visual language of its own.
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.