Product Context

Design tokens tell an AI agent how your product looks. Context documents tell it what your product is. Upload brand voice guidelines, copy principles, product descriptions, or tone-of-voice docs and every Explorer generation will read them.

What to upload

Short, specific documents work best. Examples:

  • brand-voice.md — tone, personality, words you use and avoid.
  • product-one-pager.md — who you serve, what the product does, the three value props.
  • copy-guidelines.md — button verbs, sentence case vs title case, empty-state patterns.
  • terminology.md— domain language (e.g. “orders” vs “bookings”, “patients” vs “clients”).
Context docs are most useful for copy, not visuals. Variants generated with a brand-voice doc attached produce noticeably more on-brand microcopy in the first draft.

Limits

LimitValue
Max files per project3
Max size per file50 KB
Accepted formats.md, .txt

Oversized or surplus uploads return a friendly error naming the file and the limit (for example: Context file brand-voice.md is 82KB. Each file must be ≤ 50KB.).

How context reaches generations

On every Explorer prompt, Layout concatenates the context documents into a Product Context section of the prompt, sent alongside the design tokens. The same section is written into exported layout.md, so MCP-connected agents (Claude Code, Cursor, Copilot, Windsurf, Codex) also see it when they call get_design_system.

You don't need to reference the context docs in your prompt. They are always on.

Tips

  • Prefer plain prose and bullet lists. Agents handle structured markdown better than HTML or screenshots.
  • Keep each file under one screen of text. Three short focused docs beat one sprawling document.
  • If you have a style guide, excerpt the rules you actually want enforced. Full style guides often bury rules in examples.
  • Update docs in place and re-save. No re-extraction needed.

Related