web-summary

name	web-summary
description	Summarise a single web URL the user shares: fetch with read_url, produce a compact key-points reply. No external CLI, no API key.
version	0.1.0
tags	["research","summary"]
metadata	{"soulclaw":{"category":"research","triggers":["总结一下","看一下这个链接","这篇讲了什么","summarize this","tldr","what is this article about"],"capabilities":["web_query"],"related_skills":["news-digest","daily-paper-digest","arxiv"]}}

web-summary

Trigger

Use this skill when the user shares a single URL (article, blog post, documentation page, README, news piece) and asks what it says or wants a short summary. Examples:

• "https://example.com 看一下讲什么"
• "tldr https://example.com/post"
• "总结这个 PR 的描述"

Do NOT use this when the user wants:

• A multi-source roundup → use news-digest.
• Academic papers → use arxiv / daily-paper-digest.
• Just to download / extract the page content as-is.

Inputs

• url: the URL to summarise. If multiple URLs, pick the first and ask whether to also process the others.
• focus: optional keyword the summary should bias toward (e.g. "只讲 和定时任务相关的部分").
• length: default "short" (~150 chars body) for IM, "medium" (~400) when the user asks for more detail.

Steps

1. Validate the URL (must be http / https; refuse data: / file:).
2. Call read_url(url). If the response is gated (login wall, 403), say so plainly and stop — never guess content.
3. Strip nav / footer / ad text; focus on the article body.
4. Produce 3 sections in the reply:
   • 1-line elevator pitch
   • 3-5 bullet key points (newest / most actionable first when ordered)
   • 1-line takeaway or "为什么值得看"
5. Keep within length budget. Cite the URL once at the end.

Verification

• Every claim must come from the fetched body — never invent details.
• If body was truncated, say so and offer to re-fetch with a different range / section anchor.
• For tech docs, distinguish "the doc says X" from "X is generally true".

Failure Signals

• read_url returned a paywall, error, or empty body.
• Output reads like the original article rather than a summary.
• Same paragraph paraphrased multiple times (no real compression).
• Numbers / dates / quotes don't match the source.