Cursor 必备 MCP 组合推荐

排查页面加载缓慢、资源过大或渲染延迟，落实“优化 CSS 文件体积”时使用。在 DevTools、Lighthouse 或真实数据中确认实际瓶颈后再提改动建议。

审查样式表、组件样式与响应式行为，落实“用 CSS containment 限制重绘范围”这一规则时使用。提出修复前要在各断点和交互状态下检查实际渲染布局。

实现动画渐变、涉及自定义属性取值的复杂 CSS transition，或构建类型化设计令牌系统（希望自定义属性误用能直接报错）时使用。

审查与“在真实移动设备和视口上测试”相关的 CI 覆盖、自动化检查或测试策略时使用。重点看规则是否被持续验证，而不只是写在文档里。

在新建 TypeScript 项目、审计已有 tsconfig.json，或审查存在空引用错误或隐式 any 类型的代码时使用。

Use this skill when designing, writing, or minimizing regression tests (e.g., mjsunit) for V8 bug fixes

Workflow for automatically fixing errors or addressing comments on a Gerrit CL requested by the user.

Guides the methodology for investigating V8 concepts, codebase structure, performance, memory, and architecture. MANDATORY to use whenever triaging bugs, investigating crashes, reading C++ files, analyzing compiler optimizations (Ignition, Maglev, TurboFan, Turboshaft), or doing static research on V8 subsystems. Essential for locating high-level design documents in the local docs/ directory before reading C++ code. Do not use for raw build commands or Git workflows.

Manages git task isolation and environment setup in V8. Use at the start of any new task or bug fix. Do not use for general C++ editing.

Managing and querying Perfetto traces using the trace_processor MCP server.

Handles installation and configuration of clangd-mcp for V8. Use when clangd-mcp is missing or needs configuration for the current workspace. Do not use for general C++ editing.

Guides the initial analysis and impact assessment of a V8 security report, strictly excluding implementation or fixing.

遵循最佳实践与安全标准，为 Deployment、Service、ConfigMap 与 Secret 生成生产就绪的 Kubernetes manifest。在生成 Kubernetes YAML 清单、创建 K8s 资源或实施生产级 Kubernetes 配置时使用。

使用 ArgoCD 与 Flux 实现 GitOps 工作流，达成自动化、声明式的 Kubernetes 部署与持续调谐。在落地 GitOps 实践、自动化 Kubernetes 部署或配置声明式基础设施管理时使用。

当 HyperFrames 合成需要推近、拉出、缩放、改构图、Ken Burns 效果、运镜、视觉匹配/甩镜交接，或其他 seek-safe 的 2D/3D 关键帧时使用；也用于 GSAP、CSS keyframes、Anime.js、WAAPI、FLIP、路径、遮罩、SVG morph/draw、文字拖尾、3D 纵深，或 `hyperframes keyframes` 诊断。不要用于宽泛的场景策略、品牌设计、素材获取、字幕或一般视频规划。

Explains why a process is running and what started it, using the witr CLI. Use when a port is already in use (EADDRINUSE), a dev server or other process was left running, an unknown or stuck process needs explaining, or you need to know which service, container, shell, supervisor or scheduled job started a process, holds a port or holds a file lock. Use it instead of chaining lsof, ps, ss, netstat and docker ps.

用于修改或审阅 packages/qwik 下的 Qwik 核心包代码，尤其是响应式原语、信号、VNode、cursor 行为、QRL，以及面向 optimizer 的运行时行为，或 Qwik 核心测试。

Statically review MCP configuration for mutable package references before approval or CI, without executing discovered MCP servers.

Statically review MCP configuration for mutable package references before approval or CI, without executing discovered MCP servers. Use when reviewing .mcp.json, Cursor, VS Code, Claude, Windsurf, or other repo-scoped MCP config.

Statically review MCP configuration for mutable package references before approval or CI, without executing discovered MCP servers. Use when reviewing .mcp.json, Cursor, VS Code, Claude, Windsurf, or other repo-scoped MCP config.

自动化 Unreal Engine 编辑器的场景、Actor 与渲染操作。

Load the body before the first implementation edit for a bug fix or behavior change, or when the user or governing specification makes reusable tests a deliverable. It keeps the smallest focused test with the delivered project and updates coverage when behavior changes. Do not load for greenfield work unless reusable tests are explicitly required, behavior-preserving refactors, verification-only work, plan/review/explanation/documentation/formatting/generated-output-only work, explicit no-change or no-tool work, or when the user declines tests.

On an explicit planning request, always call `read_skill` for this skill before answering. Research first, then return one concise inline plan; do not create a plan file unless the user explicitly asks or a governing workflow requires one. Start and end the reply by saying this is not a special mode and `go` executes the plan; do not implement in the planning turn. For ordinary implement, build, fix, debug, or refactor requests, work directly. When an explicit plan divides work into separate diffs, commits, or PRs and the user asks to publish, preserve that split and tell the user before deviating.

Clarify a material user-owned product decision before substantive implementation commitments. Use only when request and named/governing evidence still leave a choice that would materially change scope, platform support, user-facing surface, data/API contract, domain values, project placement, or cross-system architecture; also use when a truncated request omits such a choice. On the initial turn, inspect only named or governing requirements evidence, then ask directly without loading: prefer structured input; ask one blocking outcome question or at most three coupled questions; wait. After the answer, load the body before write or delegation. Do not use for implement/build/fix/debug/refactor work when expected behavior is settled, plan/design deliverables owned by Plan, complexity/greenfield/integration alone, agent-owned technical choices, locally discoverable facts, reversible defaults, permission to begin, or stop/no-tool turns. Never re-ask supplied facts or turn clarification into refusal.

Mandatory preflight for web frontend visual design. If available, call read_skill for bundled:taste before the first frontend create or edit tool call whenever creating or visually styling a human-viewed web page, component, app, dashboard, or interactive report. Do not load it for non-visual frontend logic or machine-only HTML. If unavailable, continue without warning. Negative constraints only; user requests and other explicit styles still win.

Setting up or installing into a Python environment. One rule applies whether or not you read the body - the environment belongs with the project, so create it inside the project directory (.venv) or let uv manage it, and never in a scratch directory like /tmp and never by forcing an install into the system interpreter. Load the body before creating a virtualenv, choosing an installer, or writing run instructions for a Python project.

Remove a personal fact, preference, relationship detail, topic, or prior event from Muse's active memory and stop existing copies or automations from bringing it back. Use for explicit requests such as 'forget that', 'don't remember this about me', or 'remove that from your memory'. Do not use when 'forget it' merely means cancel the current task.

Build, revise, or manipulate a fixed-layout PDF (report, guide, one-pager, printable document). Use whenever a build task's artifact kind is pdf, a document build's output format is pdf, or the task reads, merges, splits, crops, or fills an existing PDF, including fillable AcroForms. Covers authoring the print-CSS HTML source, rendering, the geometry and validation gates, existing-PDF manipulation and form filling, and delivery under workspace/your_files.

Search and read Granola meeting notes and transcripts through Granola's OAuth-backed MCP server.

Inspect database-backed Muse records for diagnosis and cross-table tracing when purpose-built product tools do not expose the needed state.

Read, create, edit, and deploy web apps and projects in Lovable. Use to turn a product brief into a working website or prototype, inspect existing project code, update pages and features, and publish changes through Lovable's official MCP server.

Inspect and manage the user's Vercel websites, web apps, projects, and deployments. Use to read build and runtime logs, investigate production errors and failed deployments, check deployment status, and deploy projects through Vercel's official MCP server.

Create, read, edit, or manipulate Word documents (.docx) and Word templates (.dotx). Use whenever a build task's artifact kind is document with the default docx output, or the task mentions a Word doc, .docx, or .dotx, extracts or reorganizes content from one, inserts or replaces images, does find-and-replace in one, or works with tracked changes (redlines) or comments. Covers python-docx generation, raw OOXML editing of existing files, document structure and formatting, and render verification. Not for PDFs, spreadsheets, or Google Docs.

为生产级 UI 工作确定一套 ECC 专属的前端设计方向。当构建或改进网站、仪表盘、应用、组件、落地页、可视化工具，或任何需要更强产品化设计判断的 Web UI 时使用。

获取、分析、复现并精简 GitHub issue 的复现步骤。仅在用户要求验证某个 GitHub issue 是否可复现、精简其复现用例、分析 bug 报告，或为可复现性对 issue 做审查/分诊时使用。不适用于没有对应 GitHub issue 的独立复现请求。

查询 PyTorch CI、GitHub Actions、HUD、Grafana 与基础设施指标。当用户询问 CI 耗时、任务失败、排队时长、工作流趋势、runner 健康度、看板数据或 PyTorch 基础设施指标时使用。

复现并定位 PyTorch GitHub issue 报告的 bug，在本地工作区实现修复。当用户要求修复 PyTorch GitHub issue 时使用。

对 GitHub issue 进行分诊：路由到值班团队、打标签并关闭咨询类问题。在处理新的 PyTorch issue 或被要求分诊 issue 时使用。

调试 PyTorch 2 编译器栈故障，包括 Dynamo 图断裂、Inductor 代码生成错误、AOTAutograd 崩溃与精度不一致。在遇到 torch.compile 错误、BackendCompilerFailed 异常、重编译问题、Triton kernel 失败、FX 图问题，或用户提到调试 PT2、Dynamo、Inductor 或编译模型问题时使用。

Use when working on Trilium's published Docker images on Docker Hub or GHCR rather than on the Dockerfiles — "how many pulls does the image get?", "why is the Docker Hub repository so big?", "clean up old tags", "delete untagged images", "which tags do we publish?", the legacy repositories (triliumnext/notes, zadam/trilium) and their mirroring, the main-docker.yml copy-to-Docker-Hub step, mirror-legacy-docker.yml, Docker Hub categories, stars or the overview. Includes hub.mjs (stats, tag listing/deletion, the manifest inventory and a safe untagged-image purge, all dry-run by default) and the storage model that decides how a cleanup works. Do NOT use for building or running the image locally (see the Developer Guide's Docker page).

面向 web、移动端与桌面的 UI/UX 设计智能库。在设计、构建、审查或修复界面时使用，覆盖页面、组件、设计体系、无障碍、交互、响应式布局、字体排印、色彩、图表以及特定技术栈的 UI 实现。本地可检索数据：79 种风格（可检索，50 种在用）、192 套产品配色与推理画像、74 组字体搭配、119 条 UX 准则、105 个图标、17 个 GSAP 预设、25 种图表类型与 22 个技术栈。

加固代码以抵御漏洞。在审计输入处理逻辑、处理用户输入/认证/数据存储/外部集成，或检查登录流程是否符合 OWASP Top Ten 时使用。在构建任何接收不可信数据、管理用户会话或对接第三方服务的功能时使用。在审计依赖已知漏洞、分诊包管理器审计结果或评估新包供应链风险时使用。涉及个人数据或隐私合规（GDPR、CCPA）时使用。

构建生产级、可访问、响应式的用户界面。在构建或修改界面与页面、创建组件、实现布局、满足 WCAG 可访问性要求、管理状态时使用，或当产出需要达到生产质感而非 AI 生成感时使用。

准备生产上线。在准备部署到生产环境，或询问上线前需要就绪哪些事项时使用。需要上线前检查清单、搭建监控、规划分阶段发布或制定回滚策略时使用。

Comprehensive guide for creating software diagrams using Mermaid syntax. Use when users need to create, visualize, or document software through diagrams including class diagrams (domain modeling, object-oriented design), sequence diagrams (application flows, API interactions, code execution), flowcharts (processes, algorithms, user journeys), entity relationship diagrams (database schemas), C4 architecture diagrams (system context, containers, components), state diagrams, git graphs, pie charts, gantt charts, or any other diagram type. Triggers include requests to "diagram", "visualize", "model", "map out", "show the flow", or when explaining system architecture, database design, code structure, or user/application flows.

Triage GitHub issues, review contributor pull requests, diagnose CI, and prepare explicitly requested releases for VoiceStudio's Electron desktop app and Python backend. Use for repository maintenance work, not ordinary VoiceStudio audio generation.

当用户想规划、搭建或优化短信/MMS 营销时使用，包括欢迎流程、购物车挽回、购后触达、流失挽回、促销群发或验证码类事务短信。用户提到 SMS marketing、text message campaigns、SMS sequence、SMS automation、abandoned cart text、post-purchase SMS、Klaviyo SMS、Postscript、Attentive、Twilio、A2P 10DLC、TCPA、SMS compliance、short code、toll-free SMS、MMS campaign、“要不要做短信”或“SMS vs email”等场景时也适用。邮件序列见 emails，短信文案框架见 copywriting，采集手机号的弹窗见 popups。

Browser Use Cloud（托管版浏览器自动化 API）文档参考。凡涉及 Cloud REST API（v2、v3 或 v4）、browser-use-sdk（Python 或 TypeScript）、X-Browser-Use-API-Key 鉴权、cloud sessions、浏览器 profile、profile 同步、CDP WebSocket 连接、隐身浏览器、住宅代理、验证码处理、webhook、workspace、skill 市场、liveUrl 流、定价，以及集成模式（聊天界面、subagent、把浏览器工具接入已有 agent）时使用。n8n/Make/Zapier 集成、云上 Playwright/Puppeteer/Selenium、1Password 密钥库相关问题也适用。不适用于开源 Python 库（Agent、Browser、Tools 配置）——那些请用 open-source skill。