design-is

design-is runs a structured design audit against Dieter Rams' ten 'Good design is...' principles and then hands the result off to a /make-plan prompt aimed at one of three outcomes: a new design, a refinement, or a full redesign. It operates as an orchestrator rather than an implementer, producing evidence-cited scores, a verdict, and a ready-to-run handoff prompt, but writing no implementation code itself.

Work proceeds in phases. Phase 0 locks scope: what is being audited, who the primary user is, the primary task, constraints, and any reference designs, all recorded in a scope file. Phase 1 fans out evidence-gathering subagents, each returning only facts with source citations and never scoring. A structural agent counts interactive elements, nesting depth, repeated patterns, and dead props; a visual agent reports spacing and type scales, color counts, contrast ratios, and state coverage; a copy-and-honesty agent flags inflated claims, dark patterns, and label-behavior mismatches; a weight-and-friction agent measures bytes, requests, and idle animations; an optional accessibility agent checks contrast, focus order, and keyboard reachability. Phase 2 keeps scoring with the orchestrator, which rates each principle 0-3 with one piece of evidence and a one-line justification, then reaches a NEW, REFINE, or REDESIGN verdict.

All output lands in a dated DESIGN-IS-<date>/ folder containing scope, consolidated evidence, a scorecard, the verdict with reasoning, and the copy-pasteable handoff prompt. A principle-to-subagent mapping tells the orchestrator which evidence feeds each score, with innovation and longevity reserved for orchestrator judgment. Use it when someone asks to 'audit this design', 'check this UI against Rams', or wants a critique that should lead to a plan. It is deliberately not for routine UI code review, which belongs to /review, for pure copy edits, or for pre-design ideation with no artifact yet, which should start with /make-plan directly. Findings without a source citation are rejected and the subagent redeployed, keeping every score anchored to concrete evidence.