Skills
98,067 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
Static-first Next.js 16 architecture patterns: cached shells with dynamic slots, provider islands, 'use cache' boundaries, and link preloading strategy. Use when building or refactoring Next.js routes to maximize static rendering, implementing 'use cache' with dynamic personalization, splitting entry vs static renderers, scoping client providers, or tuning prefetch behavior. Triggers on 'static shell', 'use cache pattern', 'dynamic slots', 'provider island', 'prefetch strategy', 'static first', 'cache boundary', 'route goes dynamic unexpectedly', or any Next.js architecture work involving mixed static/dynamic rendering.
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.
Structured diagnostic code library for JavaScript/TypeScript. Turns errors and other conditions into typed, machine-readable `Diagnostic` instances with stable codes, docs URLs, and actionable fields. Use this skill whenever the project imports `nostics`, or works with `defineDiagnostics`/`defineProdDiagnostics`, the `Diagnostic` class, diagnostic code registries, or structured error handling. Also covers reporters (`createConsoleReporter`, `createFetchReporter` from nostics/reporters/fetch, `createFileReporter` from nostics/reporters/node, `createDevReporter` from nostics/reporters/dev), formatters (`formatDiagnostic`, `ansiFormatter`, `jsonFormatter`), and Vite plugins (`nosticsStrip` from @nostics/unplugin/strip-transform, `nosticsCollector` from @nostics/unplugin/dev-server-collector). Also use when migrating a library's existing `console.warn`/`console.error`/`warn()` helpers or thrown `Error`s to diagnostics: follow `references/migration.md`.
First principles of observability: wide events, high cardinality, the core analysis loop, events vs metrics vs logs, and how instrumentation connects to debugging. Trigger phrases: "what is observability", "why observability", "why Honeycomb", "events vs metrics vs logs", "why wide events", "what is high cardinality", "core analysis loop", "observability vs monitoring", "what is dimensionality", "explain observability".
Take over an approved Open Design artifact (game UI, VFX, skill, RIG, enemy/boss) into Darkbone Archer without degrading the runtime. Use whenever Codex must analyze, port, integrate, or review an Open Design standalone; apply a responsive Home, Map Select, Talent, Mask, Fusion, Victory, or settlement design; port an Open Design VFX/RIG lab into Phaser; preserve existing animations omitted by a mockup; or prove current-runtime vs standalone vs integrated-runtime fidelity before merge.
Detailed Open WebUI guide: architecture, authentication, functions, pipelines, API, RAG, scaling, debugging and hidden features. Use for any Open WebUI question — how it works, how to deploy it, configure auth (OAuth, LDAP, JWT), write a function or pipeline, connect a model (Ollama, OpenAI), set up RAG/knowledge bases, scale to production, debug a problem. Also use when writing code for Open WebUI: functions (filter, pipe, action), pipelines, configs, docker-compose. Triggers (RU): веб-интерфейс, чат, база знаний, модель.
OpenAI's traditional stateless APIs: Chat Completions (GPT-5, GPT-5-mini, GPT-5-nano, GPT-4o, GPT-4 Turbo), Embeddings (text-embedding-3-small/large), Images (DALL-E 3 generation, GPT-Image-1 editing), Audio (Whisper transcription and TTS), and Moderation. Covers the openai Node.js SDK and plain fetch for edge runtimes, SSE streaming, function calling and tools, structured outputs with JSON schemas, vision, GPT-5 reasoning_effort and verbosity, error handling, and rate limits. Use when integrating OpenAI into an application, streaming chat completions, generating embeddings for RAG, creating images, transcribing or synthesizing speech, moderating content, deploying OpenAI calls on Cloudflare Workers, or debugging 429 rate limits, 401 invalid API keys, function calling failures, streaming parse errors, embedding dimension mismatches, or token limit exceeded.
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.
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.
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.
Bring an open-source GitHub repository up to community-standards and supply-chain hygiene baseline: scaffold the GitHub Community Standards files (README check, LICENSE, CODE_OF_CONDUCT, CONTRIBUTING, SECURITY, issue + PR templates), wire Dependabot version updates, the Dependency Review action, OpenSSF Scorecard, and CodeQL (when applicable), check Dependabot alerts and secret scanning, and add the OpenSSF Best Practices Badge link. Use when the user says "set up OSS hygiene", "make this repo open-source-ready", "add community health files", "wire dependabot", "add scorecard", "OSSF badge", "supply chain hardening", or invokes /oss-hygiene on a public repo. Idempotent — safe to re-run; diffs against templates and asks before overwriting. Distinct from /gh-bootstrap (merge-button + branch protection) and /safe-settings (org-scale settings as code): focuses on the contributor-facing and supply-chain surface, not merge or org policy. Run after /gh-bootstrap; before announcing the repo.
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).
Diagnose and fix the OpenViking memory plugin on this machine: the plugin install (enablement, hooks, MCP server), the client configuration (ovcli.conf / ov.conf / OPENVIKING_* env) and the connection to the OpenViking server (reachability, 401/403, /mcp). Use whenever memory "isn't working": no <openviking-context> block, empty recall, captures not landing, missing or failing MCP memory tools, 401/403, an offline statusline, right after installing/updating the plugin or switching servers/keys, or when the user asks for the plugin's status. Triggers: "memory not working", "check OpenViking", "plugin status", "记忆没生效", "插件状态", "连不上 OpenViking", "recall 为空", "401".
Use when integrating mediated ads into a Phaser 4 + Capacitor mobile project via AppLovin's official MAX Cordova native plugin; covers iOS/Android plugin integration, banners, interstitial and rewarded video, silent preloading of fullscreen ad placements and on-demand banner creation, per-placement independent load retries, video-unavailable notices, and privacy acceptance. Not for Phaser Web or mini-game ad implementations.
Build, configure, deploy, and troubleshoot durable blockchain indexers with the Subsquid Pipes SDK (EVM, Solana, Tron, Bitcoin, Hyperliquid) when Portal MCP or curl previews are insufficient for backfills, recurring syncs, joins, app-owned data, or production analytics.
Write the release notes for a Playwright release - the Node.js notes in docs/src/release-notes-js.md together with the Python, Java and .NET translations, and the GitHub markdown for each language's release. Use when asked to write, translate or publish release notes
Forces the laziest solution that actually works, simplest, shortest, most minimal. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependencies, one line before fifty. Supports intensity levels: lite, full (default), ultra. Use whenever the user says "ponytail", "be lazy", "lazy mode", "simplest solution", "minimal solution", "yagni", "do less", or "shortest path", and whenever they complain about over-engineering, bloat, boilerplate, or unnecessary dependencies.
Reverse-engineer the current state of analytics tracking from a codebase. Scans for SDK calls, identity management, and instrumentation patterns to produce a factual inventory — not recommendations. Outputs .telemetry/current-state.yaml and a timestamped audit report. Use when the user wants to know what's currently tracked, audit existing analytics, capture tracking reality before designing a new plan, 'what analytics do we have,' 'what events are tracked,' 'scan for tracking,' or 'tracking inventory.'
Prove it for debugging, performance investigation, bottleneck analysis, profiling, and optimization work. When reading code does not prove the cause, run a minimal reproduction, add temporary traces, measure bottlenecks, and verify the fix or improvement.
Scale PubNub applications for high-volume real-time events using channel groups, wildcard subscriptions, sharding, and large-event readiness. Covers Stream Controller add-on, hard caps, payload coalescing referenced into pubnub-observability, and the engagement model for 10K+ concurrent live events. Persistence/history is owned by pubnub-history.
Rebase the current Git checkout onto the requested remote base branch, or the repository's remote default branch when none is named, without losing local work. Always use this skill for an actual base-branch synchronization operation, including requests to pull or sync main, master, trunk, or the default branch; update a branch from its base; rebase onto a remote base; bring incoming base commits into the current checkout; or update a detached worktree from that base.
Publish local Git commits on a named branch without losing local or remote work. Always use this skill for an actual Git branch publication, including requests to push, publish, upload, or put the current or detached checkout on origin or another Git remote. Use it for detached-HEAD publication and post-rebase lease-protected updates; it pushes existing commits only and does not commit dirty changes, create a PR, or merge.
Python patterns for system reliability — background jobs and task queues (Celery, async), resilience and recovery (retries, backoff, timeouts, circuit breakers via tenacity), and observability (structured logging via structlog, metrics, distributed tracing, golden signals). USE WHEN building async workers, queueing tasks, handling transient network/IO failures, instrumenting Python services for production, designing retry policies, configuring logging or tracing, or any combination of these system-reliability concerns. NOT FOR language idioms or type hygiene (use `writing-python`) or project setup and dependency management (use `uv`).
Sets up static type checking for a Python package — choosing mypy, pyright, basedpyright or ty, strict configuration, per-module overrides, shipping py.typed — and fixes type-check errors. Use when the user says 'add type checking', 'set up mypy', 'pyright errors', 'make this package typed'. Not for lint or format failures, failing tests, runtime bugs, or the CI workflow that runs the checker.
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.
Comprehensive codebase quality audit with parallel agent orchestration, GitHub issue creation, automated PR generation per issue, and PM-prioritized recommendations. Use for code review, refactoring audits, technical debt analysis, module quality assessment, or codebase health checks.
Perform comprehensive code audits of Ruby on Rails applications based on thoughtbot best practices. Use this skill when the user requests a code audit, code review, quality assessment, or analysis of a Rails application. The skill analyzes the entire codebase focusing on testing practices (RSpec), security vulnerabilities, code design (skinny controllers, domain models, PORO with ActiveModel), Rails conventions, database optimization, and Ruby best practices. Outputs a detailed markdown audit report grouped by category (Testing, Security, Models, Controllers, Code Design, Views) with severity levels (Critical, High, Medium, Low) within each category.
Official Rails documentation. Use when asked about any Rails-specific topic including ActiveRecord, routing, controllers, views, mailers, jobs, Action Cable, Action Text, Active Storage, migrations, validations, callbacks, associations, caching, security, or internals.
Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.
Triage PR review comments by confidence score and act — fix the high-scoring ones, push back on the low-scoring ones, and ask about the borderline. Handles both inline review threads (anchored to diff lines) and PR-level review body comments (summaries submitted with reviews, like Age tables or Copilot overviews). First checks the PR's build and merge state and fixes those before processing comments. Use when the user says "respond to PR comments", "handle review feedback", "address PR reviews", "fix the build", "fix CI", "fix merge conflicts", or invokes /respond with a PR number. Also trigger when the user mentions a specific PR and wants to deal with reviewer suggestions — Copilot, human, or bot. Every reply this skill posts ends with an "agent on behalf of;" attribution line so reviewers can tell the comment came from an agent on behalf of the human, not the human directly. Do NOT use to generate a new review — that is /copilot-review's job.
Judges whether RHDH Jira work that already exists is ready to move forward — in RHIDP, RHDHPLAN, RHDHBUGS, and RHDHSUPP. Checks an issue, a JQL result, a sprint, or a backlog against the exit criteria for its status, then reports missing fields, hierarchy gaps, likely duplicates, unaddressed comments, stale work, and Feature Exploration readiness. Use for "is RHIDP-1234 ready", "refine this", "refine the backlog", "backlog hygiene", "what's missing on this epic", "run the Feature Exploration checklist", or "which of these are stale". Assesses existing work; it does not open new issues and does not build a sprint.
Build and revise Rive Luau scripts across Node, Layout, Converter, Path Effect, Transition Condition, Listener Action, Util, and Test protocols. Use when the user asks to write or modify Rive scripts, choose protocols, wire script inputs or data binding, debug runtime behavior, or plan unit tests for script logic.
Use when building or auditing Roblox monetization — Developer Products (Robux purchases via ProcessReceipt), Gamepasses, gem/currency bundles, or anything touching real Robux. Covers the idempotency + rollback pattern that prevents "charged but got nothing" and money-duplication bugs, plus DevProduct/Gamepass setup.
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.