user-feedback clusters a pile of customer feedback into themes — each with a frequency count, severity rating, trend direction, source distribution, and 2–3 representative quotes. It separates signal (actionable, recurring, cross-source) from noise (one-time complaints, vague frustrations, off-topic requests), and produces a prioritized list of recommended actions ranked by frequency × severity × trend.
The problem isn't collecting feedback — most product teams have more than they can read. The problem is that unstructured feedback doesn't tell you what to do next. A pile of Zendesk tickets, NPS verbatims, and app reviews looks like chaos until it's clustered. Running this skill on a real feedback corpus once shows you how much signal was sitting in plain sight. The noise call is equally instructive: the skill will tell you what not to act on, and why.
Day 7 follows the discovery plan because the PM now has a framework for what they're looking for. Running user-feedback without that frame produces themes; running it with that frame tests assumptions. Today's exercise does both — and the synthesis you produce here will feed the business case on Day 10.