Audit entity data models by comparing PRD specifications against current codebase implementation. Produces a structured field-by-field review table that shows target design vs current state, with classification decisions and a migration change list. Use this skill when the user asks to: review a data model, audit entity fields against a PRD, check if backend fields match product spec, design a data model from a PRD, compare PRD fields with database schema, or when they mention "实体建模", "数据模型审核", "字段审查", "PRD 对齐", "模型设计". Also use when the user points at a PRD and says things like "check what's missing", "are the fields right", "review this entity".
PRD scope advisor — helps decide whether a feature should be its own PRD or merge into an existing one. Two modes: (1) Split — given a large PRD or feature idea, apply a 6-dimension test framework to decide where to draw boundaries; (2) Audit — review a set of existing PRDs for merge signals, scope creep, or broken user journeys. Use this skill when the user asks: "should this be a separate PRD", "how to split this PRD", "is this PRD too big", "PRD scope", "PRD granularity", "merge these PRDs", or mentions PRD 拆分, 需求拆分, PRD 粒度, 拆需求, 合并 PRD, 需求边界.
Deep software design reviewer based on "A Philosophy of Software Design" (John Ousterhout). Two modes: (1) Review — analyze existing code for complexity, shallow modules, information leakage, pass-through layers, error handling anti-patterns, and naming issues; (2) Design — guide architecture decisions using APoSD principles (should I merge or split? how deep should this interface be? where should this complexity live?). Use this skill when the user asks for: design review, architecture review, complexity analysis, module depth analysis, code structure critique, "is this too shallow", "should I combine these", interface design feedback, error handling review, or mentions APoSD / Ousterhout / deep modules / information hiding. Also use when the user is making structural decisions about packages, layers, or abstractions — even if they don't explicitly mention the book.
Naming consultant for variables, functions, modules, types, and files. Two modes: (1) Name — given a concept or responsibility, propose 3-5 candidate names with tradeoff analysis; (2) Rename — audit existing names in code for vagueness, inconsistency, or misleading semantics, then suggest improvements. Use this skill when the user asks: "how should I name this", "what should I call this", "is this a good name", "help me name", "rename suggestions", "naming review", or mentions 起名, 命名, 取名, 改名, 名字不好.
README 美化器。修复结构混乱、格式不统一的 README 文件，输出结构清晰、格式一致、视觉专业的版本。当用户请求美化 README、整理 README、README 排版、修复 README 格式时使用。触发词包括：美化 README、README 美化、整理 README、beautify README、fix README、README 排版、README 格式化。
React 组件架构诊断器。从 7 个维度（使用者 API、数据流、可测试性、可扩展性、性能、心智模型、边界契约）深度分析一个 React 组件的设计质量，每个维度 1-5 分评分并给出改进建议。当用户请求组件诊断、组件设计审查、架构分析、组件体检时使用。也适用于用户指向一个组件目录问"这个组件设计得怎么样"、"帮我分析一下这个组件"、"组件质量怎么样"等场景。触发词包括：组件诊断、component diagnosis、组件体检、架构审查、设计分析、component review、组件评审。
Generate high-quality visual cards as self-contained HTML. Use when the user asks to create a visual card, info card, 信息卡, design card, social card, or any visually rich HTML summary of content. Triggers on phrases like "make a visual card", "信息卡", "设计一张卡片", "create a visual card", "generate a visual card".