Anyone who has asked a coding agent to build a landing page or UI has likely run into the same output regardless of which model did the work: a purple gradient background, three identical feature cards, a hero section that says something like 'revolutionize your workflow,' and a general visual sameness that instantly signals the page was AI-generated. Hallmark is a design skill, meant to be dropped into agent tools like Claude Code, Cursor, and Codex, that tackles this as a structural problem rather than something you can fix by writing a better prompt. Instead of just asking an agent to make it look good and hoping for the best, Hallmark has the agent pick a macrostructure for the layout, apply one of twenty distinct visual themes to avoid landing on the same handful of on-distribution defaults every model gravitates toward, and then run the result through fifty-seven automated slop-test checks plus a self-critique pass before it's ever handed back to the user. The specific checks target the exact tells that make AI-generated UI recognizable: generic marketing copy, repeated card layouts, overused gradient combinations, and other patterns so common across model outputs that they've become a genre unto themselves. The tool ships four distinct modes: building new UI from scratch, auditing existing code against these same anti-patterns, redesigning an interface with a different visual fingerprint while preserving its actual content and structure, and extracting a portable design specification from reference screenshots or URLs. What makes this worth paying attention to as a pattern rather than just one specific tool is the underlying idea: treating 'looks AI-generated' as a checklist of measurable, checkable failure modes rather than a vague aesthetic judgment, which is a technique that's straightforward to adapt into any team's own agent-driven UI workflow even without adopting this specific project.