mirror of
https://github.com/multica-ai/multica.git
synced 2026-08-06 19:00:01 +02:00
6f14c3004a
design.md had the same failure mode as product-overview: its most
prescriptive section restated code, and the code moved without it.
§3.2 declared "the entire project uses only three sizes — text-base /
text-sm / text-xs". Actual usage across packages/ui + packages/views:
text-sm 0, text-xs 0, text-base 0. The type system migrated to a
role-named scale (text-micro / text-caption / text-label / text-body /
text-body-lg / text-title-sm / text-title, 638 and 1131 uses for the two
commonest). tokens.css:63-85 now documents that scale, its rationale, and
the `text-sm -> text-body` migration in the file that defines it, where it
cannot drift — a strictly better version of what design.md was for.
§3.3 banned font-semibold and font-bold; the codebase has 116 and 5.
A ban nobody enforces is worse than no rule: an agent reads it, believes
it, and writes something inconsistent with 116 existing call sites.
Parts were still accurate — the interaction-state classes match button.tsx
verbatim, and the surface/radius tokens exist. But nothing referenced the
file, and the sections most likely to steer an agent's output were the
wrong ones.
Three rules that genuinely cannot be regenerated from tokens.css move into
CLAUDE.md UI Rules, which sets exactly this bar ("hard to infer from code
or easy to get wrong"): the authoritative --text-* scale pointer, the
active-must-survive-hover bug pattern, and spacing-before-divider.
docs/ is now just assets/ — the images README.md and README.zh-CN.md embed.
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>