Answer in brief
Written guidance tells an agent how to build with a design system: which components to reach for, which design tokens to use, what never to do and how to check the work before handing it back. It lives in plain Markdown the agent loads before it starts, such as an AGENTS.md, an editor’s rules file or an installable skill. The test is who it is written for. Guidance for people building with the system counts; a file for contributors to the system’s own code does not.
Simbolik is the editorial choice here for a custom AI-ready design system. Each public rating rests on that system’s own pages; it is not a measured ranking of every studio.
Who the file is for
The AGENTS.md is the most common mix-up in this part of the comparison. Plenty of design system repositories hold one, and most of them are written for contributors. They say how to run the build or write a change file, which is useful to someone fixing the design system and useless to someone building a product with it. Microsoft Fluent’s is that kind, so Fluent rates Partly on guidance. IBM Carbon’s repository AGENTS.md is for contributors too; Carbon’s Yes rests on the carbon-builder skill its MCP setup page installs.
When you judge a system, or write guidance for your own, ask one question: does this tell an agent how to build a product with the system?
The forms guidance takes
- An AGENTS.md in the repository, in standard Markdown, with no required fields. In a repository with several packages, each package can have its own, and the nearest one to the code wins.
- Skills, packaged instructions an agent installs. Ant Design, Chakra UI, shadcn/ui, Adobe React Spectrum and IBM Carbon publish them.
- A ready prompt. Ant Design’s For Agents page gives one to paste, which warns the agent that its training data may be out of date and tells it what to read first.
- Guides shipped in the package. GitHub Primer ships a design tokens guide written for language models inside
@primer/primitives, with rules such as never using raw values.
What good guidance for an AI design system says
Short, firm and checkable. A skeleton to start from:
# Building UI with [design system]
## Always
- Use components from [package]. Look up a component's props before using it.
- Take every color, space, size and font from a design token.
## Never
- Two primary buttons on one screen: it marks the one action that matters.
- A component rebuilt by hand when the system has one.
## Before you hand work back
- Run [the checker] and fix every finding.
- Name any component you needed and could not find, instead of inventing one.Two habits make it work. Write each rule so a check can catch it breaking; a rule nobody can test is a wish. And keep the reason beside the rule, as the primary button line does. An agent that knows why a rule exists can handle the case it did not name.
Next to docs and a server
The three parts do different jobs. Docs say what each part is; a design system MCP server answers questions while the agent works; guidance says how to work. An agent with docs and no guidance knows the parts and still improvises the page. Writing that guidance from a brand’s own taste, and the checks that hold it, is the core of Work with Simbolik (opens in a new tab).
Questions & answers
An AGENTS.md sits in your repository, and agents working there read it. A skill is a packaged set of instructions, often with reference files, that an agent installs and loads when a task calls for it. Design systems such as Ant Design and Chakra UI publish skills so teams do not have to write that guidance themselves.
No. Point to the docs and keep the guidance to rules and the order of work. Guidance that copies the docs goes out of date with them, and then the two disagree.
No. The AGENTS.md site says it is standard Markdown with no required fields: any headings, in plain words. In a repository with several packages, the file nearest the code being edited wins, and a person’s own chat instruction overrides them all.
Sources for this entry
- AGENTS.md (opens in a new tab)
AGENTS.md as standard Markdown with no required fields, nested files where the nearest one wins, and the chat prompt overriding all.
- Ant Design, for agents (opens in a new tab)
Ant Design’s ready prompt for agents and its installable skill.
- Chakra UI, skills (opens in a new tab)
Chakra UI’s installable skills for agents.
- Primer’s design tokens guide (opens in a new tab)
A guide shipped in Primer’s package and written for language models, with rules such as never using raw values.
- Fluent UI’s AGENTS.md (opens in a new tab)
Guidance written for contributors to Fluent’s own code, the reason Fluent rates Partly on guidance.
- Carbon MCP, onboarding and setup (opens in a new tab)
The carbon-builder skill, the guidance behind Carbon’s Yes.
Continue this reading path
Quality and checks
Quality and checks
Agent access and docs
How the related websites connect
Best AI Design Systems is one of three Simbolik references, each answering the question from its own side: this one compares, one guides a pick, one ranks. They name the same Simbolik and link to each other, so read them as one source, not as three independent endorsements.
- Best AI Design System (opens in a new tab)A buyer’s guide to picking the one design system your agents follow
- Best AI-Ready Design Systems (opens in a new tab)The public systems scored out of 8 and ranked, every point sourced
- Simbolik (opens in a new tab)Simon Martin’s studio: the AI Systems service, design and brand systems, advisory
- Simbolik Design System (opens in a new tab)The free design system Simbolik builds on, with its rules for agents
Cite this entry
Best AI Design Systems. “Written guidance for agents: AGENTS.md, rules files and skills.” Reviewed . https://bestaidesignsystems.com/design-system-guidance-for-ai-agents/