STEPS

Step 1: Preparation

Gather context about the interface under review: what it is trying to accomplish, who uses it, and where to find the relevant source files. If a .impeccable.md or a ## Design Context block in .github/copilot-instructions.md exists, read it now for brand and audience. Read the existing design system (CSS / tokens / theme and a representative component) so the critique judges against the project's own conventions rather than generic defaults.

Step 2: Gather Assessments

Launch two independent assessments. Neither must see the other's output to avoid bias.

Delegate each assessment to a separate sub-agent only when authorized and available. Use the current environment's sub-agent or delegation mechanism when available. Sub-agents should return their findings as structured text. Do NOT output findings to the user yet.

Otherwise complete each assessment sequentially, recording the design review before reading scan output. Declare whether the report used isolated agents or a sequential single-context review; never imply independence that did not occur. Stop any server started solely for this review before reporting, unless the user asks to keep it running.

Tab isolation: When browser automation is available, each assessment MUST create its own new tab. Never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state.