AgentSkillsCN

summarize

利用现有工具（web_operations、pdf_operations、word_operations、data_analysis）对 URL 或本地文档进行汇总与归纳。

SKILL.md
--- frontmatter
name: summarize
description: Summarize URLs or local documents using existing tools (web_operations, pdf_operations, word_operations, data_analysis).

Summarize

Use this skill when the user asks to summarize:

  • •a web page, article, or documentation URL
  • •an RSS feed item
  • •a local PDF/Word/CSV file

Trigger hints

  • •"summarize this link"
  • •"what is this article about"
  • •"extract key points"
  • •"summarize this file"

Preferred workflow

  1. •Identify source type
  • •URL: use web_operations first.
  • •Local file: use the matching file tool first.
  1. •Extract content
  • •Web page: web_operations.extract_text.
  • •RSS/Atom: web_operations.parse_rss.
  • •JSON endpoint: web_operations.parse_json_api.
  • •PDF: pdf_operations with extract_text.
  • •Word: word_operations with read.
  • •CSV: data_analysis.read_csv then summarize key trends.
  1. •Summarize in layers
  • •One-paragraph overview.
  • •3-7 key points.
  • •Actionable next steps or open questions.

Quality rules

  • •Keep source facts separate from your inference.
  • •If extraction is partial, say what is missing.
  • •For very long content, provide section summary first, then expand on request.
  • •Preserve critical numbers, dates, and names.

Limits and fallback

  • •If a URL blocks scraping, try web_operations.http_request or parse_html fallback.
  • •If video transcript is unavailable, clearly state the limitation and summarize available metadata/page text only.

Output template

  1. •Overview (1 short paragraph)
  2. •Key points (3-7 bullets)
  3. •Risks/uncertainty (optional)
  4. •Suggested next steps (optional)

Source attribution

Adapted from OpenClaw's summarize idea: https://github.com/openclaw/openclaw/tree/main/skills/summarize