doubao-sentiment-tracker

Trigger the skill with a sentiment research request and it runs a five-step workflow: broad web search, deep platform retrieval, content filtering, source tracing, and structured output. Every collected post includes a direct original link and sentiment classification.

The skill extracts clues like usernames, timestamps, and content fragments from secondary reports, then searches the original platform to locate the source post. Secondary sources are only used after exhausting tracing attempts and must be explicitly labeled with the reason.

Core coverage includes Weibo as the primary platform, plus Zhihu, Jike, Maimai, Bilibili, Douyin, and Douban. Twitter/X is searched only when the user explicitly requests it, using both Chinese and English keywords.

No, the skill is restricted to web and desktop clients and will not trigger on mobile. On mobile it suggests that the desktop or web version works better, then continues helping with model capabilities alone.

The skill requests user assistance to complete login through a browser control action rather than switching platforms or closing dialogs. It never substitutes a no-login platform or abandons the search early.

Results are delivered as a document with two parts: a positive/negative sentiment classification summary first, then a detailed per-platform list. Each item includes author, timestamp, summary, original post link, interaction metrics, and source type.