url-reader

IMPORTANT: This skill MUST be applied automatically whenever a URL (http:// or https://) is detected in any user message.

Fetch the URL content and summarize it in Japanese.

Special handling for blocked sites

Some sites block direct fetch. Use these proxy/alternative URLs:

Fetch strategy (preferred order)

1. •Try web_fetch if it is available in the runtime.
2. •Fallback: use exec + curl -L to retrieve content.

Qiita-specific tip

Qiita articles often expose raw Markdown by appending .md:

• •https://qiita.com/<user>/items/<id> → try https://qiita.com/<user>/items/<id>.md

When the domain is qiita.com and the path matches /items/<id>, try the .md URL first.

Workflow

1. •Detect URL(s) in the user's message.
2. •Rewrite blocked domains using the table above.
3. •Fetch content:
   • •First try web_fetch.
   • •If it fails or is unavailable, run:
     • •curl -L -sS <url>
     • •If the output looks like HTML and the site provides a Markdown endpoint (e.g., Qiita .md), try that.
4. •Summarize in Japanese, including:
   • •投稿者名（分かる範囲で）
   • •本文の要約（箇条書き推奨）
   • •重要なリンクや引用
   • •画像があれば説明
5. •If you still cannot fetch content (paywall/blocked), ask the user to paste the text.

Notes

• •If you create any intermediate/downloaded files, do not write them to /data/ root.
  • •Prefer: /data/idea/_sources/ for idea-channel work.
  • •Create the directory if needed.
• •Always attempt to fetch before saying「読めない」
• •For paywalled sites (e.g., Nikkei, WSJ), state that it is paywalled
• •For PDFs, try fetching directly (often accessible)