# huashu-skills: A Curated Directory of 53 Open-Source Agent Skills > huashu-skills is a GitHub repository that indexes 53 open-source agent skills in the standard SKILL.md format, organized into three tiers: 18 flagship skills in independent repositories, 14 character-perspective skills also in independent repos, and 22 lightweight built-in skills hosted directly in this repository. All skills work with Claude Code, Codex, Kimi Code, and any agent that supports the SKILL.md format. **alchaincyf/huashu-skills** — 花叔全部开源 Agent Skills 总目录：16 旗舰 + 14 人物视角 + 22 内置共 52 个 skill，分层分类 + AI Agent 安装协议 + 机器可读 skills.json + 更新检查机制 - Repository: https://github.com/alchaincyf/huashu-skills - Stars: 1,620 · Forks: 249 - Language: Python - License: MIT - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/alchaincyf-huashu-skills ## What huashu-skills Is and Who It Is For huashu-skills solves a navigation problem. The author has published over 50 open-source agent skills across multiple GitHub repositories. Without a central index, finding the right skill requires knowing which repositories exist. This repository is that index. The directory is useful for two audiences. The first is individual developers using Claude Code, Codex, Kimi Code, or any agent that reads SKILL.md files, who want a curated set of skills for specific tasks without building them from scratch. The second is AI agents themselves: the repository provides a machine-readable skills.json file so an agent can read the full index programmatically and help its operator choose and install a skill based on a stated need. The 30-second quickstart in the README suggests sending a message to your AI agent with the README link and asking it to recommend or install a skill based on your requirements. The agent reads the README and skills.json, then proceeds. ## Three-Tier Structure: Flagship, Character, and Built-In The directory organizes skills into three tiers based on scope and location. Flagship skills are 18 complete systems in independent repositories, each focused on a specific domain. Examples include huashu-design for high-fidelity prototypes and animations, huashu-excel for data analysis and reconciliation, huashu-report for institutional research reports, and tramstop-skill for removing AI-sounding prose from text. Each flagship skill has its own repository and is installed independently. Character-perspective skills are 14 independent repositories, each distilling the thinking style of a real person (Steve Jobs, Charlie Munger, Richard Feynman, Naval Ravikant, Andrej Karpathy, and others) into a runnable SKILL.md. These are generated using the nuwa-skill (Nuwa) tool, which is itself a flagship skill that distills any person's decision heuristics and communication style into a skill file. Built-in skills are 22 lightweight content-creation and utility skills housed directly in the huashu-skills repository as subdirectories. These include huashu-md-to-pdf for converting Markdown to PDF, huashu-proofreading for lightweight text review, huashu-topic-gen for topic ideation, and huashu-skill-updater for checking whether installed skills have new versions. They are installed by copying the subdirectory rather than cloning a separate repository. ## Installing Skills: Three Methods Flagship and character-perspective skills install via a single git clone command. The pattern is the same for all independent-repo skills: ```bash git clone https://github.com/alchaincyf/huashu-design.git ~/.claude/skills/huashu-design ``` Replace the repository name with the name of the skill you want. The README notes that the skill takes effect after restarting the agent session. The npx method also works for skills that support it: ```bash npx skills add alchaincyf/huashu-design ``` For built-in skills located inside this repository, the installation is a directory copy rather than a clone. The dukou skill, which distributes content across X Articles and Bilibili, is a special case: it requires installing the skill from the skill/dukou/ subdirectory rather than from the top-level repository. The machine-readable skills.json at the repository root lists all 53 skills with metadata. The README states that AI agents should read skills.json first when processing the index. ## Flagship Skill Coverage by Category The flagship skills cover five broad categories. Design and visual work: huashu-design generates high-fidelity HTML prototypes, slides, and animations using 20 design philosophies and a five-dimension expert review process, with MP4 export. huashu-gpt-image builds prompts for GPT-image generation using real reference names instead of adjectives. huashu-icon-set generates style-consistent icon sets using a 4x4 grid batch approach, though it requires the Doubao environment's built-in image generation. huashu-slide-doubao and huashu-slide-codex are environment-specific variants for Doubao and Codex. Data and documents: huashu-excel covers the full data analysis workflow from dirty table inspection through cleaning, metric calculation, reconciliation, and report generation, using only openpyxl as a dependency. huashu-md-html converts any file format (PDF, DOCX, EPUB, web pages, audio) to clean Markdown and then to HTML or DOCX. Skill engineering: nuwa-skill distills any person's thinking patterns into a runnable skill. darwin-skill evaluates and improves existing skills through a test-retain-or-rollback loop with git tracking. freud-skill diagnoses prompt and skill identity conflicts based on Anthropic interpretability research and rewrites them. Computer control: huashu-mac-use lets an agent operate native macOS apps that have no API, using CDP, accessibility trees, coordinate input, and pixel-level verification, with a Swift kernel. ## Environment-Specific Skills and Compatibility Constraints Several flagship skills are designed for specific agent environments and will not work outside them. huashu-icon-set and huashu-slide-doubao explicitly require the Doubao environment because they rely on Doubao's built-in image generation API. huashu-slide-codex is the Codex equivalent of huashu-slide-doubao and targets Codex's image generation. huashu-mac-use is macOS-only because it depends on macOS-specific accessibility and scripting interfaces. The seedance-skill for AI video generation requires a scene.json file format and is oriented toward a specific video generation service. Before installing a flagship skill, reading its own README to confirm environment compatibility is the correct first step. The huashu-skills directory README lists environment notes in the skill tables, but the definitive requirements are in each skill's own repository. ## Limitations: What the Directory Does Not Provide huashu-skills is an index and a set of skill definitions. It is not a package registry with dependency resolution, version pinning, or conflict detection. Two skills installed to the same agent's skills directory could interact in unexpected ways if they define similar activation phrases, and the directory provides no tooling to detect or prevent this. Version management is manual. The huashu-skill-updater built-in skill checks whether installed skills have newer versions available, but updating requires running the install command again. There is no automated update mechanism. A comparable resource is the Claude Code skills ecosystem more broadly, where any developer can publish a SKILL.md-based skill to a public repository. The difference huashu-skills provides is curation and categorization: instead of searching GitHub for skills, you get an opinionated directory from a single author with a consistent design philosophy. The trade-off is that all 53 skills reflect one author's priorities and tool choices. ## Maintenance and Repository Structure The last push to huashu-skills was on 2026-09-22. The repository has no GitHub releases. The LICENSE file is MIT and a NOTICE file is also present. The top-level directory contains 22 subdirectories for built-in skills, a skills.json machine-readable index, and the README. Each built-in skill subdirectory contains its own SKILL.md and any supporting files. The assets/ directory holds images used in the README. The author's background listed in the README includes developing the Kitten Fill Light mobile app (a top paid App Store ranking in its category), authoring a book on DeepSeek, and coverage by CCTV Focus on China as a representative of the hand-built economy trend. The skills reflect a focus on content creation, distribution, data analysis, and AI-assisted tool building. ## Conclusion huashu-skills is the right starting point for a Claude Code or Codex user who wants ready-made skills for design, data analysis, writing, video, research reporting, or character-perspective reasoning. The directory is organized around real task categories and provides machine-readable access via skills.json, which is more useful than a plain README when you want your agent to help choose and install a skill. The limitation is that several flagship skills are environment-specific: some require Codex's built-in image generation, others the Doubao environment. Verify environment compatibility before installing skills that document platform restrictions. ## FAQ ### How do I install a skill from huashu-skills into Claude Code? For flagship and character-perspective skills, run git clone https://github.com/alchaincyf/.git ~/.claude/skills/ and restart your agent session. For built-in skills in this repository, copy the relevant subdirectory into your skills folder. The npx skills add alchaincyf/ command also works for supported skills. ### What is the skills.json file in huashu-skills? skills.json is a machine-readable index of all 53 skills in the directory. The README states that AI agents should read it first when processing the directory, since it provides structured metadata that is easier for agents to parse than the full README text. ### Do all huashu skills work in every agent environment? No. Several flagship skills are environment-specific: huashu-icon-set and huashu-slide-doubao require the Doubao environment's built-in image generation, huashu-slide-codex requires Codex, and huashu-mac-use is macOS-only. Each skill's own README documents its environment requirements. ## Sources - [alchaincyf/huashu-skills on GitHub](https://github.com/alchaincyf/huashu-skills) - [Issues](https://github.com/alchaincyf/huashu-skills/issues) - [README](https://github.com/alchaincyf/huashu-skills/blob/master/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/alchaincyf-huashu-skills