Kami vs Taste Skill
Which should you use? A side-by-side look at what each one does, who it suits, what it costs, how hard it is to set up, and where each falls short.
Open source · MIT
Layout rules and templates that let AI agents export documents as PDF, PNG, or editable PowerPoint.
Open source · MIT
Agent skill that steers coding agents toward distinct web designs with reference images and browser checks.
The short answer
Choose Kami if you need…
- People who use Claude Code, Codex, or Cursor and want polished documents without manual layout work
- Writers or founders who need one-pagers, resumes, or formal letters in a consistent style
- Developers who want technical reports or changelogs rendered as clean PDFs
- Users who write in Chinese, Japanese, or Korean and need proper font handling
Choose Taste Skill if you need…
- Developers using Claude Code or Codex who want less generic frontend output
- Designers prototyping landing pages from a brief
- Teams building marketing pages that need real imagery with credits
- People who can wait roughly 30 to 45 minutes per build
Side-by-side
| Kami | Taste Skill | |
|---|---|---|
| Type | AI model / agent | AI model / agent |
| Price | Free and open source | Free and open source |
| Licence | MIT | MIT |
| GitHub stars | 12.2k | 108k |
| Setup difficulty | Beginner | Beginner |
| How you get it | Install with a package manager | Install with a package manager |
| Main language | HTML | JavaScript |
| Last code update | 5 days ago | 6 days ago |
What each one is
Kami
Kami is a set of layout rules and document templates for AI coding agents such as Claude Code and Codex. You ask the agent for a one-pager, letter, resume, report, or slide deck, and Kami helps it produce a styled file as PDF, PNG, or editable PowerPoint. The README describes eight built-in document templates plus a landing page system, with English, Chinese, Japanese, and Korean support.
Taste Skill
Taste Skill is an Agent Skill for coding agents such as Claude Code and Codex. It gives the agent an art direction, real photographs from open archives, and a way to check its own page in a real browser, so the output is less like a generic AI-made landing page. The README says it also covers dashboards and app screens since v2.1.
Best for
Kami
- People who use Claude Code, Codex, or Cursor and want polished documents without manual layout work
- Writers or founders who need one-pagers, resumes, or formal letters in a consistent style
- Developers who want technical reports or changelogs rendered as clean PDFs
- Users who write in Chinese, Japanese, or Korean and need proper font handling
Taste Skill
- Developers using Claude Code or Codex who want less generic frontend output
- Designers prototyping landing pages from a brief
- Teams building marketing pages that need real imagery with credits
- People who can wait roughly 30 to 45 minutes per build
Strengths
Kami
- Installs through a single documented command for several major agents
- Covers multiple languages with language-specific font fallbacks
- Optional brand profile file lets you reuse your color, name, and tone settings
Taste Skill
- Gives the agent a concrete design process: directions, a found idea, browser checks at three widths
- Uses real, credited photographs from open archives rather than placeholder art
- Installs with no API keys required, according to the README
Not ideal for
Kami
- People who need precise pixel-level control in a visual editor, since Kami works through an agent and templates rather than a drag-and-drop interface
- Windows users who do not want to use WSL, because the README says helper scripts are Bash and native Windows needs separate GTK/Pango setup
- Anyone who cannot run an AI agent, since the skill is designed to be triggered by agent requests
Taste Skill
- Users who want a quick one-line component: a full build takes about 30 to 45 minutes and a few hundred thousand tokens, per the README
- Users without Node 22+ or Chrome, Edge or Chromium, since the scripts need them
- Projects that cannot reach the internet, because finding pictures needs network access
Trade-offs
Kami
- Setup depends on external tools such as WeasyPrint, and the README says native Windows needs extra GTK/Pango runtime installation
- Some features, like syntax highlighting, degrade gracefully when optional packages such as Pygments are missing
- Output quality and layout depend on the AI agent driving it, which the README does not fully detail
Taste Skill
- Slow and token-heavy per build, by the README's own estimate
- Image sourcing depends on network access and open collections, and the README says replacing them with your own is recommended
- Its design rules reflect the author's taste, so output may still need human review
Getting started
Kami
Install the skill for Claude Code, Codex, and Cursor
npx skills add tw93/kami -a claude-code codex cursor -g -yUpdate the skill
npx skills update -g -yTaste Skill
Install the design-taste-frontend skill
npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend"Frequently asked questions
What is the difference between Kami and Taste Skill?
Kami: Layout rules and templates that let AI agents export documents as PDF, PNG, or editable PowerPoint. Taste Skill: Agent skill that steers coding agents toward distinct web designs with reference images and browser checks. Kami suits people who use claude code, codex, or cursor and want polished documents without manual layout work, while Taste Skill suits developers using claude code or codex who want less generic frontend output.
Are Kami and Taste Skill free?
Kami: Free and open source. Taste Skill: Free and open source.
Is Kami or Taste Skill more popular?
Taste Skill has more GitHub stars (108k) than Kami (12.2k). Popularity isn't the same as fit — check which one matches your use case above.
What are other alternatives to Kami and Taste Skill?
Other options people compare them with include Marp, diagram-design, Humanizer, Ponytail, Hermes Agent.
Keep comparing
Built from our catalogue entries for both products: open-source details come from each project's README (AI-assisted, refreshed weekly) and GitHub data; AI-tool details are editorial. Spotted something wrong? Let us know from either product's page.