Skill Recommendation: Kami Turns Content into Unified, Deliverable Visual Outputs

I recently came across Kami, which tw93 open-sourced, and my first reaction was: this is not a purely template-based project, but a very good case for observing the shape of Claude Skill.

Its tagline is: Good content deserves good paper. Translated literally, it means, “good content deserves good paper.” This sentence actually captures its positioning quite accurately: AI is already very good at generating text, summarizing materials, and rewriting structure, but when it comes to actual delivery, there is often one last missing step—typesetting, visual hierarchy, document tone, and readability.

That is exactly what Kami does.

It packages a document design system as a Skill, so when users issue natural-language requests, Claude Code, Claude Desktop, and even general Agents compatible with the ~/.agents/ directory automatically call the corresponding capabilities to organize content into pages that look more like “formal deliverables.”

1. What Kami Is

Kami is an open-source document design Skill from tw93. The project name Kami corresponds to the Japanese word for “paper.” From the project description, it appears to be the third tool in the Kaku · Waza · Kami trio:

• Kaku is responsible for writing code;
• Waza focuses on habits and practice;
• Kami handles document delivery.

In practical terms, Kami mainly targets these scenarios:

• resumes;
• one-pagers;
• long-form documents;
• formal letters;
• portfolios;
• slide decks;
• equity or research reports;
• changelogs;
• product landing pages.

These kinds of content can of course be produced directly by AI, but often they are only “text-correct.” Kami emphasizes that content should not only exist; it should be readable, presentable, and deliverable.

So it does not simply say, “please make the layout look nice.” Instead, it puts templates, fonts, visual rules, language flows, export methods, and usage instructions into one reusable Skill package.

2. Why It Works Well as a Skill Example

Anthropic’s official definition of Agent Skills is that a Skill is a modular capability for extending Claude; each Skill can package documentation, metadata, and optional scripts, templates, and other resources, and Claude automatically uses it in relevant tasks.

This is clearly different from traditional prompting.

A normal prompt usually only works within the current conversation. Today you might write, “please format this in a certain style,” and next time you need to paste it again. A Skill is more like an operational manual in the file system: it can have SKILL.md, it can have scripts/, and it can also include references/, templates, fonts, and other assets. Claude loads the relevant content only when needed, instead of forcing all information into context at once.

Kami fits this model nicely.

It is not merely a prompt saying “use warm paper color, blue highlights, serif fonts.” It stores those preferences as a design system:

• It describes the Skill’s purpose, trigger scenarios, and workflow in SKILL.md;
• It uses templates for different document types;
• It uses scripts to handle packaging, font fallback restoration, version checks, and other mechanical steps;
• It uses reference files to define brand materials, visual rules, and output constraints;
• It is triggered by natural language, so users do not need to remember complex commands.

That is why I find it particularly valuable: it demonstrates that a Skill is not just “a longer prompt,” but can gradually become a reusable workflow asset.

For a deeper look at the core difference between Skills and prompts, you can read the introductory piece here: Skill Special Topic: From Prompts to Reusable AI Workflows. For another example of packaging a workflow as a Skill, see Adding a Workflow to a Coding Agent with Superpowers.

3. What Kami Can Generate

Publicly available information shows that Kami currently supports a variety of document and page templates, including one-pagers, long-form documents, letters, portfolios, resumes, slide decks, equity reports, changelogs, and landing pages.

Its output is not about being “flashy,” but about a consistent paper-like and editorial feel. Several visual keywords repeatedly emphasized in the project copy include:

• warm-toned paper-like backgrounds;
• ink-blue accent color;
• serif-driven hierarchy;
• a rhythm closer to formal publications;
• CJK typography paths for Chinese, Japanese, and Korean.

This is especially meaningful in Chinese contexts. A major issue with many AI-generated Chinese documents is not content quality but typographic tone: unstable fonts, uncomfortable line height, rigid heading hierarchy, and awkwardness when a direct English template is applied to Chinese.

Kami uses dedicated font stacks and fallback chains for Chinese. In the public SKILL.md, you can see it sets different font strategies for Chinese, English, Japanese, and Korean, and also notes that Japanese and Korean paths are best-effort and should be visually reviewed before delivery.

This shows it does not treat “multilingual support” as a slogan; it actually considers fonts, spacing, punctuation rhythm, and export quality.

4. How to Install and Trigger

If using Claude Code, the README gives this installation method:

npx skills add tw93/kami -a claude-code -g -y

If using a general Agent, such as Codex, OpenCode, or other tools that read ~/.agents/, use:

npx skills add tw93/kami -a '*' -g -y

If using the Claude Code plugin marketplace and your version supports it, you can do this:

/plugin marketplace add tw93/kami
/plugin install kami@kami

If using Claude Desktop, you typically download kami.zip from GitHub Releases, then go to Customize > Skills, click + to create a Skill and upload the archive. Public documentation shows that the Claude Desktop package for Kami generally avoids including large CJK font files to prevent the Skill ZIP from becoming too big; missing fonts are then restored via scripts.

Triggering is also natural and does not require remembering slash commands. Users can simply say:

帮我做一份一页纸

or:

帮我把这份研究整理成长文档

You can also say:

build me a resume

design a slide deck for my talk

When Claude judges the request relevant, it will call this Skill automatically.

5. What the Relationship Is Between It and MCP

Strictly speaking, Kami itself is closer to a Skill than an MCP server.

MCP, or Model Context Protocol, is an open standard proposed by Anthropic to let AI applications connect to external systems in a unified way, such as local files, databases, search tools, business systems, and workflows. Official documentation often likens MCP to a USB-C interface for AI applications.

The focus is different between Skill and MCP:

• A Skill is more like “task documentation + templates + scripts + resource bundle”;
• MCP is more like “a protocol for connecting AI to external data sources and tools”;
• Skills are good for consolidating stable workflows;
• MCP is good for connecting external systems, reading/writing data, or calling tools.

For the Kami case, its core value is Skillization rather than MCPization. It encapsulates “how to make a polished document” as a reusable capability. In the future, if it needs to connect to Notion, Google Drive, CMS, or enterprise asset libraries, then it would be more suitable to combine it further with MCP.

In other words, Kami solves “how to turn content into a deliverable artifact”; MCP solves “how an Agent stably connects to the outside world.” They can complement each other, but they are not the same layer.

6. The Workflow Value I See in Kami

What I find most valuable about Kami is not that one particular template looks especially beautiful, but that it decomposes the AI workflow into more stable layers:

1. Turning Aesthetics into Rules

Many times people ask AI to make documents and the result is inconsistent because aesthetic requirements only stay at the verbal level.

For example:

Make it feel more premium

Make the layout cleaner

Like a formal report

Those expressions are too abstract, and the model may interpret them differently each time. Kami writes visual style into the Skill: background paper, accent color, font hierarchy, template structure, and language support become clearer constraints.

2. Making Repetitive Tasks into Templates

Resumes, one-pagers, long-form documents, portfolios, slide decks, landing pages—these all have stable structures at their core.

If you start prompting AI from scratch every time, that is wasteful. The value of a Skill is to fix these repeatable structures so users only need to provide content and goals instead of re-designing the process each time.

3. Writing Delivery Standards into the Tool

Many AI outputs can seem “usable,” but they are not yet suitable to truly share with others. For example:

• unclear heading hierarchy;
• uncomfortable Chinese typography;
• unstable white space;
• misalignment after exporting PDFs;
• slide decks lacking a consistent style;
• long documents missing reading rhythm.

Kami pushes these issues to the front within the Skill. So AI is no longer just writing content, but moving toward “deliverable quality.”

7. What I Learned for My Own Work

If I transfer this approach to my own content site or AI Workflow projects, Kami reminds me: do not only write prompts. Put repeatable work into a Skill.

For example, in writing articles, name-pattern case studies, AI tool introductions, and SEO content organization, I have many repetitive actions:

• standardizing frontmatter;
• unifying heading hierarchy;
• checking naked links in body text;
• collecting sources at the end;
• controlling SEO description length;
• preserving the original author’s voice;
• verifying time-sensitive information online;
• outputting Markdown that is ready to publish.

If these are written into prompts each time, they can work, but in the long run they become very repetitive. A better approach is to organize them into your own Skill so the Agent can automatically apply them whenever it sees this type of content.

That is also Kami’s biggest reminder to me: the key to AI workflows is not just model capability, but the distillation of human-verified judgment, formatting, and process into reusable assets.

8. Who This Is Suitable For

I think Kami is especially suitable for:

• people who frequently ask AI to generate resumes, reports, PPTs, or portfolios;
• people who want AI-generated content to become a formal deliverable;
• people using Claude Code or Claude Desktop;
• people wanting to learn the Claude Skill project structure;
• people creating personal branding, product intro pages, or project one-pagers;
• people who care about Chinese document typography quality.

But note that it is not a universal designer.

If you need a highly complex brand system, enterprise-level visual design, or strict compliance with a specific company’s VI guidelines, human review is still needed. A Skill can raise common delivery quality, but before final delivery you should still check content accuracy, visual quality, font licensing, exported files, and target-platform compatibility.

9. Conclusion

This project made me feel that the real value of Skills is gradually becoming clear.

In the past, we often discussed whether AI can write. The more important question in the next stage may be: after AI writes, can it consistently turn into a deliverable artifact?

Kami gives an answer: yes, but only if style, templates, scripts, fonts, trigger logic, and delivery standards are all distilled into a reusable structure.

That is precisely what makes Claude Skill worth attention. It is not just about writing a longer prompt, but about turning one person’s experience, one team’s standards, and even one product’s delivery method into a workflow that an Agent can invoke repeatedly.

For me, Kami is a strong reminder: a truly useful AI tool does not have to start out huge. Building one high-frequency, concrete, reusable action and making it stable, clear, and deliverable is already highly valuable.

References

• tw93/Kami GitHub Repository
• Kami Official Project Page
• Kami SKILL.md
• Anthropic Agent Skills Official Documentation
• Anthropic Skills Example Repository
• Anthropic: Introducing the Model Context Protocol
• Model Context Protocol Official Documentation: What is MCP?