review-extractorlisted

# Review extractor Turns a review page into a table of people worth contacting, with the reason attached to each row. ## Why the review text is the whole point Generic scrapers collect contacts. This collects **reasons**. A name and a job title tells you who to email; a review tells you what to say. If the review text does not survive into the output, this skill has failed regardless of how many rows it produced. ## Which reviews to take Default to **3-star reviews**, and 2- and 4-star if volume is thin. A 1-star reviewer has usually already churned — they are the competitor's post-mortem, not your prospect. A 5-star reviewer is happy and not going anywhere. A 3-star reviewer is still paying, still using it, quietly frustrated, and sitting in a contract that renews. They have written down, in public, the exact thing that would make them switch. Take 1-star reviews only if the user explicitly asks, and say why you are flagging it. Sort by **most recent** and prefer the last 90 days. A complaint from two years ago has usually been fixed or the person has left. ## Getting the input Two paths, both fine: **A URL** — fetch it if a fetch tool is available. If fetching fails or returns a JS shell with no review content, do not retry repeatedly. Say so plainly and ask the user to open the page, select all, and paste. That is a 10-second fix and much faster than fighting a renderer. **Pasted text** — this is the common case and works better than people expect. Ask for sele