Azhi-ss

让 agent 用单文件 HTML 替代 markdown 输出空间性、并行性、可比较或可交互的内容。当用户请求"对比/计划/评审/报告/解释器/幻灯片/编辑器/可视化"或当输出含并排选项、流程图、时间线、代码 diff、可交互元素、需要"指着某处说就是它"的场景时使用。本 skill 是路由器：先用决策框架判断是否走 HTML，再按 9 大类调度对应的子 skill（01-explore / 02-review / 03-design / 04-proto / 05-diagram / 06-slides / 07-research / 08-report / 09-editor），并复用 ../templates/*.html 与 shared/base.css 输出真实可运行的产物。

子 skill — 设计系统与组件展示。当用户要从 repo 提取 design tokens 做团队 reference、或要把一个组件的所有 variant 平铺在一张 contact sheet 上 review 时使用。包含设计系统参考（design-tokens.html）和组件变体陈列两种范式。当父 skill 路由命中"设计系统/design tokens/组件变体/展示组件库"类请求时加载。

子 skill — 原型与可点击流。当用户要决定一个 transition 的 duration/easing（动画沙盒，animation.html）或要"走一遍"4-6 屏交互流验证手感时使用。包含动画沙盒和可点击流两种范式。当父 skill 路由命中"动画/原型/手感/clickable flow"类请求时加载。

子 skill — 插画与图表。当用户要博客/文档配图（SVG 插图集）、流程图、状态机、deploy pipeline（flowchart.html）时使用。所有图都用内联 SVG，不引外部库（mermaid/d3/excalidraw）。当父 skill 路由命中"图/diagram/flowchart/state machine/pipeline 可视化"类请求时加载。

子 skill — 探索与规划。用户在多种实现方案/设计风格之间未决，或方案选定后要写工程版 RFC 时使用。包含三方案对比（compare.html）、视觉设计方向、实现计划（plan.html）三种范式。当父 skill html-effectiveness 路由命中"对比/选型/RFC/计划"类请求时加载。

子 skill — 代码评审与理解。当用户要 PR 评审、PR 写作（作者侧）、或要理解陌生代码模块的请求路径/调用栈时使用。包含 PR 评审（diff.html）、PR 写作（pr.html）、模块图三种范式。当父 skill 路由命中"评审/PR/审阅/这段代码怎么工作"类请求时加载。

子 skill — 键盘控制幻灯片。当用户要"上去就能讲"的会议幻灯片、不想用 Keynote/PPT/Google Slides、希望单 HTML 文件左右键翻页时使用。当父 skill 路由命中"幻灯片/slide deck/做几张 slides/讲几页"类请求时加载。

子 skill — 研究与学习。当用户问"X 在这个 repo 里怎么工作"（特性解释器，含 TL;DR/折叠步骤/tab 代码/FAQ）或要解释抽象概念（concept.html，含交互式可视化）时使用。当父 skill 路由命中"解释/explain/教我/这是什么/learn"类请求时加载。

Guidelines and constraints for modifying the FastAPI api.py server, the BayesianOptimizer core logic, the physics-informed KnowledgeEngine prompting, or the VectorMemory embeddings/RAG layers. Use this skill whenever the user asks to modify python backend code under backend/optimization/ or backend/api.py.

Guidelines and constraints for modifying the React, Vite, Tailwind CSS 4.x, and Recharts frontend of BOagent. Use this skill whenever the user asks to modify App.tsx, BenchMode.tsx, OperationalMode.tsx, custom components in src/components/, or the styling theme in index.css.

Rules and protocols for running backend unit/integration tests (pytest) and frontend E2E tests (Playwright) in BOagent. Use this skill whenever the user asks to run test suites, check code coverage, add new test cases, mock API calls, or verify modifications to the codebase.

审查科学计算代码的数值稳定性、物理单位一致性和数学正确性

强制测试驱动开发循环 - 先写测试（RED）→ 实现功能（GREEN）→ 重构（REFACTOR）

Specialized expert for BOagent codebase. Analyzes Bayesian Optimization logic, materials science physical constraints (Band Alignment/Defects), and verifies PVK-LLM alignment. Use when auditing backend optimization logic, benchmark consistency, or frontend-backend API contracts.

Simplifies code for clarity. Use when refactoring code for clarity without changing behavior. Use when code works but is harder to read, maintain, or extend than it should be. Use when reviewing code that has accumulated unnecessary complexity.

Rules and instructions for local environment setup, dependency management, starting backend/frontend servers, compiling builds, and troubleshooting environment configurations in BOagent. Use this skill whenever the user asks to start/restart development servers, install packages, configure environment variables (.env files), or troubleshoot connection, CORS, and build issues.

Use this skill whenever the user wants help choosing an academic figure color palette, needs venue-specific or colorblind-safe design advice, wants a paper color scheme recommendation, wants to match a color scheme for extracted architecture diagrams, or says "学术配图配色", "论文配色方案", "色盲友好配色", "学术配色", "架构图配色", "academic color palette", "colorblind safe figure", "paper color scheme", "architecture diagram color matching", "Nature Blue", "单色系". Now 13 preset schemes including the new Nature Blue monochrome, plus monochrome-vs-polychrome philosophy.

Use this skill whenever the user wants detailed English prompts for AI image tools to produce top-conference-quality academic figures, needs prompts for framework diagrams, architecture diagrams, pipeline flowcharts, module detail diagrams, comparison figures, or data-pattern grids. Now also supports JSON structured figure specs for precise layout/text control. Trigger: "论文配图提示词", "生成论文配图", "学术论文生图", "架构图提示词", "框架图提示词", "顶会风格配图", "CVPR 风格图", "NeurIPS 风格图", "paper figure prompt", "academic diagram prompt", "fig JSON spec".

Use this skill whenever the user wants an end-to-end academic figure workflow, wants to go from a repository or paper to a figure prompt, is unsure which academic-figure skill to start with, or says "帮我从仓库到配图走一遍", "完整论文配图工作流", "academic figure workflow", "end-to-end figure pipeline", "from paper to figure prompt", or "which skill should I use first".

Use this skill whenever the user wants to extract architecture diagrams from academic papers, filter out invalid images, analyze the structure and components of diagrams, automatically match suitable color schemes, or says "提取论文架构图", "架构图分析", "从PDF中提取图表", "自动分析架构图", "architecture diagram extraction", "extract figures from pdf", "analyze architecture diagram".

Use this skill whenever the user wants to analyze an academic paper, identify figure-worthy content, plan which figures to generate, suggest figure types and count per section, or says "分析论文配图需求", "论文需要哪些图", "论文配图规划", "paper figure planning", "analyze paper for figures", or "which figures does my paper need".

Use this skill whenever the user wants modern ML or RL paper-style figure prompts matching recent ICLR, NeurIPS, or ICML 2024-2025 aesthetics, needs a soft pastel academic diagram style, or says "pastel风格论文配图", "现代ML论文配图", "modern ML figure prompt", "pastel academic figure", "ICLR 2024 风格图", or "NeurIPS 2025 风格图".

Use this skill whenever the user wants to analyze a deep learning or machine learning code repository, understand what it does, identify its architecture and tech stack, generate a quick understanding document for downstream figure planning, or says "分析代码仓库", "仓库分析", "repo analyzer", "analyze this repo", "理解这个代码库", "what does this repo do", or "code repository analysis".

Run claude-sci-switch release readiness checks while respecting the project iron laws.

Launch the claude-sci-switch Tauri desktop app safely for local UI inspection without touching real Claude Science state.

Generate visually unified image-based PPT/PPTX decks from articles, reports, papers, notes, or outlines.

面向 AI 辅助深度学习研究的文件化上下文交接协议。适用于创建或集成研究项目状态文件，让任何 AI / IDE 能通过 AGENT.md 和 .workflow/ 快速理解项目目标、计划、实验、事件和下一步行动。

Minimize Python/ML project dependencies to their smallest working set. Use this skill whenever the user asks to analyze, shrink, extract, or clean up dependencies for any Python or Machine Learning project — even if they just say "fix my requirements.txt", "what packages do I actually need", "extract this module from a big repo", or "set up a clean environment". Also use when migrating ML code between machines (e.g. GPU → CPU) or debugging ModuleNotFoundError in a new environment.