Open-source tools.

One loop for AI‑native design systems: ds‑context prepares context, ui‑repair repairs code, ds‑eval measures whether the agent follows the system, prompt‑regress compares prompt versions. Specification and production checks follow.

Naive59.3
Fixture87.8

Runs the same UI tasks through model outputs and measures whether a coding agent follows a design system it did not invent.

select 38→91settings 44→93dialog 41→90delete 84→61empty 86→87
dialog 41→90empty 86→87delete 84→61select 38→91settings 44→93
settings 44→93delete 84→61empty 86→87select 38→91dialog 41→90
empty 86→87select 38→91settings 44→93dialog 41→90delete 84→61

Compares two prompt versions on one UI suite: what improved, what broke, instead of an average score.