Illustrated Agent Skills: Companion Repository for a Practical Guide to AI Skill Writing

What This Repository Is and Who It Is For

The README describes this as the companion resource repository for the book 'Illustrated Skill: A Practical Guide to AI Productivity' by the author known as Baoyu (GitHub: JimLiu). The book was published by Turing Press and is available in print and as an ebook.

The repository holds content that is either too large for print or better suited to online reference: the full appendix, skill templates from chapter 5, complete functional skills, chapter code and data examples, and the image generation prompts used to produce every illustration in the book. Engineers interested in writing skills for Claude Code, Cursor, Codex, or similar AI agents will find the skill templates and the skills/ directory directly usable. Readers of the book use the appendix here as the definitive and up-to-date version, since platform-specific installation instructions and API endpoints change over time.

Repository Structure: What Lives Where

The top-level layout covers six main areas. The 附录.md file is the full appendix, kept online so that platform-specific content can be updated without reprinting. The 术语表.md file is a glossary of English technical terms and their Chinese translations used throughout the book. The skill-templates/ directory holds simplified chapter-5 skill templates. The skills/ directory holds complete, installable skills. The examples/ directory holds chapter code, data, and outputs. The campaign-data-2025-01.csv file is the sample dataset for chapter 7.

Chapter illustrations and their prompts are stored in a separate archive branch named archive-chapters-illutrations/chapters-illutrations, moved there because of file size. Each chapter directory under that branch contains an imgs/ subdirectory with the final images used in print, a prompts/ subdirectory with the generation prompt for each image (named to match the image file), a bak/ subdirectory with rejected alternatives where they exist, and a ref/ subdirectory with reference material where it exists.

The appendix has four sections: Appendix A is a terminology table, Appendix B is platform-specific installation guides covering Claude.ai, Claude Code, VS Code, and OpenCode, Appendix C lists Chinese-market model API platforms and deployment options including 百炼, Kimi, 智谱, DeepSeek, and 火山方舟, and Appendix D contains the complete versions of the writing workflow and interview workflow skills that the book discusses in abbreviated form.

Installing and Using the Skills

The skills in skills/ are designed to be installed directly into a supported agent. The README points to Appendix B for platform-specific installation steps. For the baoyu-skills collection, the README documents a direct install command:

npx skills add jimliu/baoyu-skills

An alternative for Codex or Claude Code plugin marketplaces:

/plugin marketplace add JimLiu/baoyu-skills

The README includes a caution worth noting: each active skill occupies context in the agent's context window. Installing all 20+ skills at once will slow the agent and increase the likelihood of confusion. The README's advice is to read the full list and install only the skills that match your current workflow.

For the baoyu-design skill, which generates UI mockups, prototypes, wireframes, and slides as self-contained HTML files, the installation path is the jimliu/baoyu-design repository. It is listed separately because its output format and use case differ from the content-generation skills.

The Skill Templates: Writing, Analysis, and Meetings

The skill-templates/ directory contains simplified versions of the skills introduced in chapter 5. These are intentionally reduced from the full versions in the appendix, making them easier to understand and adapt. The table in the README lists eight templates:

content-analyzer.md analyzes source material. outliner.md generates multiple outline options. writer.md handles prose generation. writing-style.md constrains the tone and voice. article-polish.md polishes finished drafts. article-illustrator.md is a simplified version of the illustration skill. meeting-analyzer.md processes meeting transcripts. meeting-minutes.md produces meeting minutes from analyzed transcripts.

The complete versions with tighter orchestration logic and more detailed instructions are in Appendix D. The README recommends starting from the templates and iterating toward the full versions after seeing where the simplified ones fall short.

Full Skills in skills/: From Illustration to Info Digest

The skills/ directory contains seven complete, installable skills. The book-illustrator skill is notable because it was used to produce every illustration in the book itself. It takes a full chapter manuscript, identifies all illustration positions in one pass, runs sub-agents in parallel to generate images, and writes the results back into the manuscript. It includes a references/design-system.md with a three-color publishing palette, contrast rules, and a list of AI-image cliches to avoid.

The adversarial-polish skill is listed with a specific warning in the README: it is intended for learning and research, not practical use. The skill depends on sub-agent context isolation, which the README states is not currently supported on any major platform, so actual results will not match the design intent. It is kept in the repository to illustrate "role isolation" and "adversarial reasoning" as design concepts.

The info-digest skill converts any input (news, long-form articles, reports, papers, policies, product materials, or URLs) into a concise Chinese-language digest. It is listed as immediately practical and does not have the adversarial-polish caveat.

Chapter 7 Example: Watching a Skill Evolve

The examples/chapter-07/ directory contains three versions of a data analysis skill: data-analysis-v1, data-analysis-v2, and data-analysis-v3. The README describes this as showing how a skill is "used out", refined through actual use rather than designed completely upfront. An output/ directory under each version shows the results that version produced.

The campaign-data-2025-01.csv file in the root of the repository is the sample dataset used in chapter 7. Readers can run the three skill versions against it to observe the differences in output directly.

This incremental example is one of the more concrete learning tools in the repository. Skill design is not described abstractly; the evolution from v1 to v3 shows specific decisions made to improve the skill, and the output directories let you compare results without running anything. A reader who studies the three versions side by side gets a view of the iteration process that written explanations alone cannot provide.

The last push to the main branch was on 2026-06-15. The repository has no GitHub releases. The license follows the book's standard terms; the README states that commercial use requires contacting the author.

One limitation is worth stating directly: the repository does not include documentation for each skill beyond what the README summarizes and what the book explains. Someone picking up skills/ without the book will be able to install and run them, but the reasoning behind design choices in, for example, the book-illustrator skill's design-system.md or the adversarial-polish skill's warning will be opaque. The README is explicit that the archive branch holds illustration prompts and images, but navigating the archive branch requires knowing the chapter structure of the book.

Editorial conclusion

This repository is useful for readers of the book and for engineers who want concrete, working skill examples for Claude Code, Cursor, Codex, or similar agents. The skill templates in skill-templates/ are ready to copy and adapt immediately. The full skills in skills/ include a book illustrator, content analyzer, interview analysis, outliner, and others. Anyone learning skill design can study the chapter-07 evolution from v1 to v3 to see how a data analysis skill is refined. The repository is not a standalone tool: it requires the book for full context on the design decisions behind each skill. The last push was on 2026-06-15, and the repository has no releases. The license is the book's standard terms, so commercial use requires contacting the author.

Frequently asked questions

Official sources

1. Issues
2. JimLiu/Illustrated-Agent-Skills on GitHub
3. README