web-read

name	web-read
description	Extract clean, readable markdown from a web page with the Defuddle CLI — strips nav, ads, and boilerplate to cut tokens versus WebFetch. Use when given a URL to read or analyze (docs, articles, blogs, changelogs, RFCs). Do NOT use for .md or JSON/API URLs — use WebFetch directly.
user-invocable	true

Web Read

Core Rule

When handed a content URL to read, extract clean markdown with the Defuddle CLI instead of pulling raw HTML through WebFetch — it strips nav, ads, cookie banners, and boilerplate, so the same content costs far fewer tokens. Never silently install a global tool: if Defuddle is missing, surface the one-line install and fall back to WebFetch so the task isn't blocked.

When to Use

When the user gives a URL to read, summarize, or analyze:

• Documentation pages, API guides, RFCs, specs
• Articles, blog posts, release notes, changelogs
• Any standard, content-heavy web page

Do NOT use for:

• URLs ending in .md / .txt — already clean; use WebFetch directly
• JSON / API endpoints — use WebFetch or curl
• Pages behind auth — Defuddle fetches anonymously; use an authenticated path instead
• Vendoring library docs into docs/references/ — that's /references-sync's job (it may call this skill to do the extraction step)

Prerequisites

Check once per session:

command -v defuddle >/dev/null && echo ok || echo missing

If missing, tell the user the install and ask before running it — a global install is a tool change worth surfacing:

npm install -g defuddle

While it's unavailable, fall back to WebFetch rather than blocking.

Process

1. Extract clean markdown:

   defuddle parse <url> --md
   

2. For long pages, save to a file and read only the slice you need instead of inlining the whole thing:

   defuddle parse <url> --md -o /tmp/<slug>.md
   

3. Pull a single metadata field when that's all you need:

   defuddle parse <url> -p title
   defuddle parse <url> -p description
   

Run defuddle --help for the current flag set — prefer it over memorized options, it stays in sync with the installed version.

Output Format

Clean Markdown of the page's main content, with site chrome removed. Use it as the source for summaries, extraction, or analysis; when you quote the page, quote from this cleaned text.

Flag	Output
--md	Markdown — default choice
--json	JSON with both HTML and markdown
-p <name>	A single metadata field (title, description, domain)
(none)	Raw HTML — rarely what you want

Notes

• Defuddle is a global CLI (npm i -g defuddle), not a project dependency — it never touches package.json, so it doesn't trip the Protected-Changes dependency gate. Surface the install anyway before running it.
• The token win is real on heavy pages (docs sites, news, marketing): raw HTML through WebFetch carries layout, scripts, and nav that Defuddle drops.
• Pairs with /references-sync, which can call this to extract vendor docs before writing docs/references/<pkg>-llms.txt.
• Upstream: github.com/kepano/defuddle