WorkBuddy HTML Prototypes Look Too AI-Generated: Am I Using It Wrong?

I came across an article about Huashu Design today, describing how to turn Agents into design studios using HTML as the canvas: first find brand assets, then propose directions, and finally export prototypes and slides. After reading it, I got itching to try.

I have a small task: organizing internal backend PRDs, screenshots, and a few design-tokens.json files into a clickable HTML slide deck, while providing three visual directions.

I've just started seriously using WorkBuddy for design-related tasks. Previously, when I used it to organize materials, I was afraid it would mix structured fields with screenshots into a mess, so I created isolated workspaces and followed the Ask-Plan-Craft workflow. This time, I threw the files directly in, asking it to read tokens and generate pages. It did provide a few versions, but the color scheme looked like overnight latte—rounded cards, purple gradients, and emoji icons all over the place. Worse, in the export preview, titles overlapped buttons, and the text looked like a keyboard stepped on by a bug.

I tried having it list brand assets first, then compositions, then write HTML. I also tried giving it just one page.

But the output still wasn't worth looking at. Maybe I misunderstood. WorkBuddy's role here is to package materials into something handoff-ready for other Agents; it hasn't filled the gap in the design tool layer.

Has anyone else encountered this? I suspect it's missing the skill set or references directory from Huashu Design. WorkBuddy is good enough for PPTs and spreadsheets, but for HTML prototype aesthetics, I dare not rely on it for now.