React composition patterns that scale. Use when refactoring components with boolean prop proliferation, building flexible component libraries, or designing reusable APIs. Triggers on tasks involving compound components, render props, context providers, or component architecture. Includes React 19 API changes.

Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers. Also covers AWQ, GGUF, GPTQ and HQQ (guides in references/).

Guides the user through an interactive conversation to define their research project, then generates research_brief.json and tasks.json. Use when starting a new project, when no research_brief.json exists, when the user wants to start from a specific pipeline stage, or when the user wants to redefine their research pipeline.

Search published venue papers (IEEE, ACM, Springer, etc.) via Semantic Scholar API. Complements /aris-arxiv (preprints) with citation counts, venue metadata, and TLDR. Use when user says "search semantic scholar", "find IEEE papers", "find journal papers", "venue papers", "citation search", or wants published literature beyond arXiv preprints.

Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.

Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency. Use for production deployment on NVIDIA GPUs (A100/H100), when you need 10-100x faster inference than PyTorch, or for serving models with quantization (FP8/INT4), in-flight batching, and multi-GPU scaling.

WeChat Mini Program development skill for building, debugging, previewing, testing, publishing, and optimizing mini program projects (小程序开发、调试、预览、发布). Covers project structure and config (`project.config.json`, `appid`, `miniprogramRoot`, `tabBar`, routing/navigation, icon assets), WeChat Developer Tools Nightly workflows (`wechatide` CLI, WeChat IDE Skills/MCP), `miniprogram-ci` preview/upload, console/network debugging, message push (消息推送) and customer-service auto-reply (客服消息), mini program SEO / search indexing (小程序搜索优化、页面收录、搜索推广、mpcrawler), and CloudBase integration (`wx.cloud`, 腾讯云开发, 云开发) when explicitly used. Use when users create, develop, modify, debug, preview, deploy, publish, or promote WeChat Mini Programs. NOT for Web frontend (use web-development), pure backend services (use cloudrun-development / cloud-functions), or UI-design-only tasks (use ui-design).

Fast path for a minimal CloudBase Web + database demo (最小前后端 / 最小可用 fullstack / Lovable-like BaaS). Defaults to @cloudbase/js-sdk client CRUD (NoSQL app.database / PG app.rdb), MCP-only schema, preview-first, and forbids cloud functions unless secrets, cron/background jobs, or logic that security rules/RLS cannot express. Use for 搭一套 demo、留言板、Todo、Notes、Kanban, or when users say 带云函数+云数据库 but only need CRUD. NOT for production multi-service backends, CloudRun, WeChat Mini Programs, or tasks that truly need server secrets.

Use this skill for WeChat Mini Program AI via wx.cloud.extend.AI (小程序, wx.cloud apps). Covers generateText and streamText with callbacks (onText, onEvent, onFinish); streamText needs a data wrapper, generateText returns the raw response. Also covers multimodal image input (image_url content blocks, OCR and form extraction) — vision-capable models only. Models via wx.cloud.extend.AI.createModel with groups cloudbase (main managed, also 小程序成长计划 with `hy4-preview` / `hy3` / `deepseek-v4.1-flash`), the legacy hunyuan-v3 / hunyuan-exp, or custom-*; model id goes in the `model` field. MUST run two-step preflight before code — see body. NOT for browser/Web (use ai-model-web), Node.js backend (use ai-model-nodejs), or image generation (use ai-model-nodejs).

Unified CloudBase execution guide for all-in-one skill installs. Use this first for CloudBase app tasks, especially existing apps with TODOs, fixed pages, or active handlers. Routes PostgreSQL / CloudBase PG / app.rdb() / queryPgDatabase / managePgDatabase work away from legacy NoSQL and old auth patterns.

Use when working with camera, photos, audio, haptics, ShazamKit, on-device music analysis, face grouping and video highlights, screen recording, casting to non-AirPlay routes, SharePlay playback, the user's Apple Music library, lock-screen metadata, or CarPlay app design, templates, and navigation. Covers AVCaptureSession, AVFoundation, Core Haptics, MediaPlayer, MusicKit, NowPlaying, ScreenCaptureKit, DockKit.

Agentlas 시스템 운용 절차 — hep-network 편성, cargo/Agent Cloud·marketplace 활용, workforce 로스터 재사용, 메모리 계약. One(개인 에이전트)이 Agentlas를 조작할 때 이 절차를 따른다.

Install or verify the oma CLI and its runtimes (bun, uv, serena) before any oma-* skill, agent, or workflow runs an `oma` command. Use when `oma` is missing from PATH, when a skill says to run `oma ...` and the shell reports "command not found", or on the first task in a fresh Cursor or Grok Bot workspace.

Investigate source-build failures in local dotnet/dotnet VMR builds and `SB_*` CI legs. USE FOR VMR source-build issues involving prebuilt packages, source-build/source-only failures, source-build-assets (SBRP), package poisoning, repo build ordering, binary detection/removal, or failures limited to `SB_*` jobs. DO NOT USE FOR regular CI test failures, codeflow staleness, dependency flow tracing, crash dumps, or general NuGet package management unrelated to source-build.

Create or update a pull request using the NuGet.Client PR template. Use when asked to: create PR, open PR, push and create PR, submit PR, open pull request, send changes for review, update PR description, fix PR body, edit PR description, update the PR.

Generate a project-local skill that formally verifies the code where a bug costs the most and tests reach the least, in any language, with whichever verification tool fits. Use for /create-formal-verification-skill, "model-check this race", "prove this invariant", or when concurrency, retries, state machines, parsers, or unsafe code need more than tests.

Ship user-facing PocketHost features with a blog post instead of semver release notes. Use when landing feat/enh work, finishing a customer-visible change, or when the user asks to announce a feature. Pair with blog-voice for writing style.

Identifies high-quality leads for your product or service by analyzing your business, searching for target companies, and providing actionable contact strategies. Perfect for sales, business development, and marketing professionals.

Create a content brief for ONE planned post: target and secondary keywords, search intent, template choice, competitor gaps from the live SERP, sourced statistics to include, section plan, internal-link targets, and distribution ideas. Use when the user says "content brief", "blog brief", "SEO brief", "article brief". Not a skeleton only (blog-outline), a multi-post plan (blog-strategy, blog-calendar, blog-cluster), or the article itself (blog-write).

Maple's OpenTelemetry conventions: custom span attribute keys (`maple.*` vendor namespace, `query.context`, `db.query.*`, `result.*`, `cache.*`, `tenant.*`), Title Case status codes (`Ok`/`Error`/`Unset`), resource attribute dual-emit (`deployment.environment` + `deployment.environment.name`), span kinds, Tinybird MV pre-extracted columns, loop-prevention filters, and sampling. Use whenever writing or reviewing instrumentation code in any language (TypeScript, Rust, Python) in this repo: adding `setAttribute`/`setAttributes`/`record`/`#[instrument(fields(...))]` calls, setting span status, configuring an OTLP exporter, defining a new resource attribute, or wiring a new query through `WarehouseQueryService.compiledQuery()`.