Changxing
@changxingConnectOnion creator. The actual Claude Code setup I live in every day — release automation, bilingual writing pipeline, social cross-posting, and a small army of subagents for code review, design, and SEO. Everything here is what I run, not what I think you should run.
This is the actual Claude Code setup I run on my machine — every skill, slash command, and subagent here is something I use weekly or more. Nothing aspirational, nothing demo-only.
It clusters around four workflows:
Shipping software. /release runs a 12-step PyPI release end to end (version bump → tests → build → upload → tag → docs site → announcement tweet). ship-feature is the higher-level orchestrator. migrate-to-ts ports one Python feature at a time to the TypeScript SDK. sync-docs and update-docs keep the docs site honest after every change.
Writing in two languages. A pipeline of small skills that compose: outline-plan → nonfiction-structure → nonfiction-scene → nonfiction-depth → wenbai-jiehe (文白结合) → ai-language-fix (strip AI-isms) → reader-feel (critical pass) → title-refine → translate-to-english. Each skill does one thing well; together they take a rough Chinese draft to a polished bilingual essay.
Cross-posting & engagement. tweet and post push to X and LinkedIn from a single command (X Premium, 4000-char threads). x-engage, x-reply, x-tone-check, and x-write are the toolkit for replying with the right voice. make-cover and make-poster generate the visuals — Bauhaus posters via nano-banana (Imagen 4 / gemini-3.1-flash-image-preview).
Specialist subagents. Six narrow-focus reviewers I summon for specific work: simplicity-engineer (kills over-engineering), ui-ux-design-auditor (catches UX issues), code-documentation-writer (10-principle docstrings), nextjs-landing-page-architect and landing-page-design-architect (paired build+design), and seo-optimization-analyst (Next.js-specific SEO).
The meta-skill that holds it all together is skill-sediment: every time I correct Claude in a way I'd want to apply again, that lesson goes back into the relevant skill file. The setup compounds because the skills compound.
Subscribe and you get the same tools, namespaced under changxing/ so they don't clash with anything you already have.
§ ISkills
26Reusable capabilities the agent can call on — each one a small unit of expertise.
AI语言特征修复 (user)
React and comment on LinkedIn feed. Picks appropriate reactions based on content, leaves short authentic comments.
Use browser agent to test frontend functionality and detect UI issues through automated testing and screenshot analysis
Generate images using Google Nano Banana (Imagen 4). Use when the user wants to create, generate, or produce images using AI.
Implement fix for a GitHub issue and create commits
Create implementation plan for a GitHub issue
Update PR based on review feedback
非虚构写作·思想深度 (user)
非虚构写作优化流程·编排器(调度其他 skill)(user)
非虚构写作·场景写作 (user)
非虚构写作·结构布局 (user)
大纲规划·根据读者痛点重设章节骨架,写到 frontmatter outline: 字段,结构层据此执行 (user)
Post content to both X (Twitter) and LinkedIn simultaneously. Handles text, files, Chinese input. Use when user wants to cross-post.
读者感觉·AI 创业者+作家视角批评,输出网友评论腔 (user)
Best practices for Remotion - Video creation in React
Ship a feature end-to-end — update tests, docs, docs-site, then release to PyPI. Use when user says "ship", "ship feature", "release", or asks to publish a new version.
把对话中验证过的修正和技巧沉淀进 skill 文件,防止下次重蹈覆辙。用于在改进某个 skill 后,把洞见固化到对应文件里。
文章标题优化流程 (user)
Post a tweet or thread to X. Handles content from text, files, or Chinese input. Use when user wants to post something to X/Twitter.
文白结合·逐段中文表达打磨 (user)
Raw X (Twitter) API layer - post, reply, search, get timeline. Called by other skills. Not for direct user invocation.
Find hot tweets (by topic search or home timeline) and reply to them with humor. Fully automatic posting.
Auto-generate a witty humor reply to a tweet, polish it, and post it. Fully automatic. Use when user wants to reply to a tweet with humor.
Check if a reply will make people laugh or angry. Soften with emoji or suggest angle change. Posts automatically after approval. Called by x-reply before posting. Not for direct user invocation.
Prepare and refine content for X - translate, fix AI language, format threads, generate short replies. Called by other skills. Not for direct user invocation.
§ IISlash commands
10Single-shot operations triggered by name. Predictable inputs, predictable outputs.
Analyze code structure and add documentation headers to each file
Analyze and iteratively refine website design until it meets professional standards
Generate a social media cover image with title overlay. Supports Red Note, LinkedIn, X, Instagram, and any custom size. Smart mode: pass a markdown file to auto-design from article content.
Generate a Bauhaus/Swiss style graphic art poster (no title). Pure visual background art for use as a cover base image or standalone poster.
Migrate one Python ConnectOnion feature to TypeScript version
Automate version release - update version, build, publish to PyPI, create GitHub release, update docs, and post announcement tweet
翻译中文文章为英文 (user)
AI-driven wiki content synthesis from docs/ with SEO optimization (comprehension-based, not copy-paste)
§ IIISubagents
6Specialised collaborators with their own scope, tools, and tone of voice.
Use this agent when you need to create or update documentation for code following best practices. This includes writing docstrings, API documentation, README sections, inline comments, or any form of code documentation. The agent follows the 10 core documentation principles to ensure high-quality, maintainable documentation. Examples:\n\n<example>\nContext: The user has just written a new function and wants to document it properly.\nuser: "I've created a new authentication function, please document it"\nassistant: "I'll use the code-documentation-writer agent to create comprehensive documentation for your authentication function following best practices."\n<commentary>\nSince the user needs documentation for their code, use the Task tool to launch the code-documentation-writer agent.\n</commentary>\n</example>\n\n<example>\nContext: The user has a module that lacks proper documentation.\nuser: "The payment processing module needs documentation"\nassistant: "Let me use the code-documentation-writer agent to document the payment processing module with clear explanations of the why, not just the what."\n<commentary>\nThe user needs documentation for an existing module, so use the code-documentation-writer agent to create structured, principle-based documentation.\n</commentary>\n</example>
Use this agent when you need to create comprehensive design plans for landing pages that maximize conversion rates and user engagement. This includes situations where you're starting a new landing page project, redesigning an existing page, or need expert guidance on layout, visual hierarchy, and conversion optimization strategies. The agent will provide detailed design specifications following the 12 core principles of effective landing page design.\n\nExamples:\n- <example>\n Context: User is building a new SaaS product landing page\n user: "I need to design a landing page for my new project management tool"\n assistant: "I'll use the landing-page-design-architect agent to create a comprehensive design plan for your project management tool landing page"\n <commentary>\n Since the user needs a landing page design, use the landing-page-design-architect agent to provide expert design guidance.\n </commentary>\n</example>\n- <example>\n Context: User wants to improve conversion rates on an existing page\n user: "My landing page isn't converting well, can you help me redesign it?"\n assistant: "Let me use the landing-page-design-architect agent to analyze your needs and create an optimized design plan"\n <commentary>\n The user needs landing page optimization, so the landing-page-design-architect agent should be used.\n </commentary>\n</example>
Use this agent when you need to create, set up, or organize a Next.js landing page project with Tailwind CSS. This includes initial project scaffolding, configuring the development environment, establishing project structure, setting up build processes, and preparing the project for team collaboration. The agent excels at using terminal commands for project setup and ensuring the project is ready for other developers to use immediately.\n\nExamples:\n- <example>\n Context: User wants to create a new landing page project\n user: "I need to set up a new landing page for our startup using Next.js"\n assistant: "I'll use the nextjs-landing-page-architect agent to set up a professional Next.js landing page project with Tailwind CSS"\n <commentary>\n Since the user needs a Next.js landing page setup, use the nextjs-landing-page-architect agent to handle the project initialization and configuration.\n </commentary>\n</example>\n- <example>\n Context: User needs help organizing an existing Next.js project\n user: "Can you help me reorganize my Next.js project structure to follow best practices?"\n assistant: "Let me use the nextjs-landing-page-architect agent to analyze and reorganize your project structure"\n <commentary>\n The user needs help with Next.js project organization, which is a core capability of the nextjs-landing-page-architect agent.\n </commentary>\n</example>
Use this agent when you need to analyze and provide recommendations for improving the search engine optimization (SEO) of a website or web application. This includes evaluating current SEO implementation, identifying missing optimizations, and providing actionable recommendations for improving search rankings, visibility, and organic traffic. The agent will analyze technical SEO, on-page SEO, content optimization, and Next.js-specific SEO features.\n\nExamples:\n<example>\nContext: The user wants to improve their website's search engine visibility and rankings.\nuser: "find what we should do to optimize the SEO for the website"\nassistant: "I'll use the seo-optimization-analyst agent to analyze the current SEO implementation and provide optimization recommendations."\n<commentary>\nSince the user is asking for SEO optimization recommendations, use the Task tool to launch the seo-optimization-analyst agent to perform a comprehensive SEO audit.\n</commentary>\n</example>\n<example>\nContext: After implementing new pages or features, the user wants to ensure SEO best practices are followed.\nuser: "Check if our new landing page follows SEO best practices"\nassistant: "Let me use the seo-optimization-analyst agent to review the landing page's SEO implementation."\n<commentary>\nThe user needs SEO validation for new content, so use the seo-optimization-analyst agent to audit the page.\n</commentary>\n</example>
Use this agent when you need to review code or features for over-engineering and complexity issues. Examples: <example>Context: User has just implemented a new feature with multiple abstraction layers. user: 'I just added a new caching system with factory patterns, strategy patterns, and dependency injection containers' assistant: 'Let me use the simplicity-engineer agent to review this implementation for potential over-engineering' <commentary>The user has implemented something that sounds complex, so use the simplicity-engineer to identify if the complexity is justified or if simpler approaches would work better.</commentary></example> <example>Context: User is considering architectural decisions for a new component. user: 'Should I create an abstract base class with multiple inheritance levels for my data processing pipeline?' assistant: 'I'll use the simplicity-engineer agent to evaluate this architectural approach' <commentary>The user is asking about potentially complex architecture, so the simplicity-engineer should review and suggest simpler alternatives if appropriate.</commentary></example>
Use this agent when you need to analyze and critique UI/UX implementations in code, identifying design problems and prioritizing fixes. This agent reviews component code, templates, and styling to find usability, accessibility, and design principle violations. <example>\nContext: The user wants to review newly implemented UI components for design issues.\nuser: "I just finished implementing the new dashboard components"\nassistant: "I'll use the ui-ux-design-auditor agent to analyze the design implementation and identify the top issues to fix"\n<commentary>\nSince new UI components were implemented, use the ui-ux-design-auditor to review the code and identify design problems.\n</commentary>\n</example>\n<example>\nContext: The user has created new page layouts and wants design feedback.\nuser: "The product listing page is complete with filters and sorting"\nassistant: "Let me launch the ui-ux-design-auditor agent to review the page design and find any UX issues"\n<commentary>\nThe user completed a new page implementation, so the ui-ux-design-auditor should analyze it for design problems.\n</commentary>\n</example>
§ IVPosts
3Notes, demos, and stray thoughts published from the workshop floor.
This is what's in my `~/.claude` right now, end to end — the skills I actually invoke, the slash commands I lean on, the subagents I summon. No "tutorials," no aspirational config. The thing I run.
Releasing used to be a checklist of 12 things — bump version, run tests, build wheel, upload to PyPI, tag, push, update changelog, regenerate docs site, post the announcement. Each step a 30-second decision.
Every conversation with Claude Code teaches me something — a sharper way to phrase a prompt, a workaround for a flaky API, a mistake I want it to never make again. But the lesson lives in the conversation, and the conversation ends.