Content summary skill

This is the short public alias for:

• content-search-summarization

Also available as simpler aliases:

• summary
• 内容摘要

Use this skill when you want to:

• search Bilibili, YouTube, or similar public content platforms
• pick the top N relevant results for a topic
• summarize each item in Chinese with links

Primary guidance

1. Prefer opencli for supported platforms.
2. If opencli is unavailable, fall back to Playwright scraping of public result pages.
3. Rank by relevance first, popularity second.
4. Open selected result pages and use metadata to improve summaries.
5. Use a structured output with source, capture time, link, summary, and confidence.

Key rules

• Always include source links.
• Include capture time or say when the timestamp is not visible.
• Do not pretend a full video was watched if only metadata was collected.
• Phrase summaries conservatively when based on public page metadata.
• Add a confidence label and a short caveat when the summary is metadata-based.

Quick invocation template

You do not need to use only /content-summary.

Reliable invocation patterns include:

1. /content-summary
2. use the content-summary skill
3. a natural-language request that clearly asks for content search, ranking, and summary output

Use prompts like:

Use /content-summary to find the top 5 results for <topic> on <platform> and summarize each item with links and confidence labels.

使用 /content-summary 在 <平台> 搜索 <主题>，筛选 Top 5，并输出带链接与置信度的摘要。

Please find the top results for <topic> on <platform>, rank them by relevance, and summarize each item with links and confidence notes.

Output contract

The skill output should always include:

1. search method used (opencli or fallback)
2. keyword and capture scope
3. ranked results with links
4. per-item confidence and caveat
5. explicit note when summaries are metadata-inferred

Pointer

For the full detailed playbook, also see:

• skills/content-search-summarization/SKILL.md in this repository