Multi-Platform Sentiment Monitor

Problem

When tracking how a brand, product, company, or topic is discussed across social media and news sites, manual search often misses platforms, mixes in stale posts, and makes it hard to reach a defensible sentiment or risk conclusion. This skill turns keyword-based public opinion monitoring into a repeatable workflow: scope confirmation, multi-platform search, post-search time validation, detail extraction, sentiment and risk tagging, and report generation. It fits tasks such as checking recent negative mentions for a brand, identifying where a topic is gaining traction, or setting up scheduled negative-risk alerts.

How It Works

The skill first confirms the keyword and time range. If the user does not specify a range, it defaults to the last 7 days and states that assumption. It then runs web_search in controlled batches across Weibo, Zhihu, Bilibili, Toutiao, NetEase, Tencent News, The Paper, Sina, Sohu, 36Kr, Huxiu, IT Home, People's Daily Online, Xinhua, Caixin, Xiaohongshu, WeChat Official Accounts, Baijiahao, and similar sources, with dedicated queries for negative and positive signals.

The key constraint is filtering results after retrieval: instead of relying on unreliable search-time filters, it checks dates in titles or snippets and discards items outside the requested window, so stale content does not fill the report. For high-value links, it uses web_fetch to extract full text, prioritizing official statements, authoritative media, and in-depth analysis. It then labels items as positive, negative, neutral, or mixed, assigns risk levels, and writes structured data according to report_schema.md. Finally, generate_report.py produces an HTML report with overview metrics, sentiment charts, platform ranking, detailed entries, risk alerts, and recommendations.

It also supports scheduled monitoring: it parses expressions like “every day at 7” or “weekdays at 9 and 18,” creates separate scheduled_task entries, and supports pausing, resuming, changing frequency, adding keywords, or running a task immediately.

Boundaries

The skill is best suited for public Chinese web and social media monitoring where content can be discovered via search. It does not guarantee complete coverage of every platform or access behind login walls. If no relevant content exists in the time window, the report should say so rather than backfilling old posts. Clear keywords and an explicit time range make the output far more useful.