WorkBuddy Expert Mode Prompts Lead to Logic-Lacking Demos. Anyone Else?

I think the real hassle is that when you write requirements briefly, AI paints you a pie in the sky until it disappears. I saw a post about PRDs turning into Prompts. Since I've been using WorkBuddy for a month and have run automated weekly reports for four weeks, I got itchy fingers. Listen to this: the requirement is simple—throw the community points system in there, let it generate a PRD first, then convert it into a clickable Demo.

I tried for two days. When I first started using AI-generated PRD tools, the results weren't quite right. I set up a fixed entry point in WorkBuddy, pasted some user quotes from Tencent Meeting transcripts and old workflows from WeCom docs, and asked it to output only goals, roles, fields, and exception flows. The first attempt gave me a decent-looking table. Points expiration and redemption failures were just titles, no rules.

Later, I clicked the button to generate prototypes and page flows, and it gave me a bunch of static descriptions. Field names didn't match my docs either—for example, user_id became "User ID," and point_balance was split into "Points Balance" and "Available Points." Even weirder, it prompted missing context, cut off halfway through generation, and exports lacked version comparison. I wrote a post last week about WorkBuddy auto-collecting review materials; back then I thought fixing the entry point was enough. Now I see maybe it isn't. If the Prompt is too short, it acts like a developer being rushed for requirements—it just parrots back, never asks follow-ups.

My current confusion is: For WorkBuddy's Product Manager expert mode, should I feed it a complete PRD first, or a brief requirement description and let it ask questions? Is it because I haven't configured field mappings, or does Demo generation require state machines and exception tables to be included? Has anyone encountered this? Did I mess up some setting?