AI Layer

Orbit is built to be used by AI, not just by people. Its tokens, component descriptions, and documentation are written so tools like Claude can read the system and build against it directly — which is what lets a small team keep a growing product suite consistent. For why that matters strategically, see About Orbit.

What makes it machine-readable

• Semantic naming. Tokens and layer names describe intent, not appearance — color-action-primary, not blue-500. The same vocabulary appears in Figma, the docs, and the code, so a tool reading any one of them works from a shared language.
• Explicit component descriptions. Every component says what it is, when to use it, when not to (and what to use instead), and what related components exist. That’s what stops a tool from producing output that’s technically correct but contextually wrong.
• Machine-readable specs. A compact specs/design.md covers the token system, colour roles, type and spacing scales, and the accessibility baseline; an auto-generated llms.txt indexes every page. Both are plain text any capable assistant can load.

How to use it with your tools

Orbit is staff-only, served behind Cloudflare Access — so AI tools read the context from the repository, not the live URLs:

• Claude Code / Cursor / Copilot — clone Open-Point-Design/orbit and work inside it. The repo’s CLAUDE.md gives full context automatically; public/specs/design.md and public/llms.txt are there for lookups.
• Claude Project — upload the knowledge files from ai/knowledge/ and the instructions in ai/instructions/ to the project’s knowledge base.
• Any other assistant — paste public/specs/design.md and the UX Guidelines at the start of a session.

The live orbit.openpoint.com/… URLs need an interactive staff login, so a tool can’t fetch them — always point tools at the repo files.

Human-led, AI-assisted

We’re a small team, and like most people building software today, we use AI tools every day. They help us move faster — exploring ideas, drafting documentation, writing first-pass specs, and catching gaps we might otherwise miss.

But Orbit is the design system our product teams build on. Our products are used across many sectors — government, transport, utilities, and more — and, through Social Point, directly by the community members and citizens who take part in engagement. For all of them, consistency, accessibility, and accountability aren’t optional. So we’re clear about how AI fits in.

Everything in Orbit is designed and owned by our design team. AI assists — it never gets the final say. Designers are involved from start to finish, and stay responsible for:

• The ideas, direction, and design decisions
• What’s right for our products, our teams, and the people who use them — including the citizens and communities on the other side of the screen
• Accessibility, correctness, and fit with the rest of the system
• The review, QA, and sign-off — every change is reviewed by a designer against the pre-ship review rubric, and system-level calls sit with the Design System Guild

No component, guideline, or page ships as raw AI output. There’s always substantial design direction, judgement, and human approval behind it. AI helps us do more, and document it more thoroughly — it doesn’t replace the design thinking that accessible, high-stakes software demands for the organisations and communities who rely on it.

Design ideas. Design decisions. Design sign-off. AI just helps us build what we were going to build anyway — faster, and more completely.

— The Product Design team