§0terms
Glossary
The words that this site uses, in plain language. On each page, the first use of a term has a dotted underline. Point at it, or move to it with the keyboard, to read the short text.
- agent
- An AI program that writes and changes code, for example Claude Code or Cursor. On this site, "agent" means a coding agent.
- AGENTS.md
- A file in a repository with the instructions for agents. Source
- alias
- A token whose value points at a different token. DTCG also calls it a reference. Source
- Basecoat
- The look of shadcn/ui as plain HTML and CSS, without React. Source
- CI
- Continuous integration: the checks that run on each change of the code.
- claim
- A number on this site, kept with its base, its source and its date in one registry. Source
- component registry
- A list of the components of a system that a tool or an agent can read. Source
- component token
- A token for one component only. It points at a semantic token. Source
- context
- One option of a modifier, for example index, kiln, light or dark. Source
- decision
- A written choice of the team, with an id, a reason and the check that enforces it. On this site, a row of the register. Source
- design token
- A named design value, for example a color or a spacing step, kept as data. DTCG 2025.10 gives the file format. Source
- deterministic
- Gives the same result each time, with no language model. Source
- drift
- A difference between the design system and what was built. Source
- DTCG
- The Design Tokens Community Group and its file format for design tokens, version 2025.10. A Final Community Group Report, not a W3C standard. Source
- Enforced by
- The field of a decision that names the check that proves it, or says "not enforced (gap)". Source
- hallucination
- Something that an agent invents, for example a component name that does not exist. Source
- lint
- A script that finds where code or content breaks a rule.
- llms.txt
- A short text file at the root of a site that tells agents what the site contains. Source
- MCP
- Model Context Protocol. The open protocol that agents use to read tools and data. Source
- modifier
- A named choice in the resolver with its options. This site has two modifiers: brand and mode. Source
- permutation
- One combination of the modifier options. This site has four. Source
- primitive token
- A token that holds a raw value and no intent, for example a gray of a given lightness. Source
- proposal
- A request for a new part or a new rule. An agent writes it. A person approves it. Source
- raw value
- A color, size or font written directly in the code, not taken from a token. Source
- resolver
- The part of a token set that selects the values for a context, for example a brand or a mode. Source
- semantic token
- A token that holds an intent, for example the color of a link. It points at a primitive token. Source
- shadcn/ui
- A set of open-source components that you copy into your code. Agents know it well. Source
- skill
- A file that tells an agent how to do one task. In my method, a skill applies a rule. It does not make one. Source
- tacit knowledge
- What a team knows but never wrote down.
- typed spec
- A structured description of a page that a script turns into HTML. The agent fills the spec. It does not write the HTML. Source
- vendored
- Copied into your repository and owned there, not installed as a package.